Having clear goals is vital for every organization. It allows team members to focus on the work that matters and deliver measurable results. Defining such goals is never easy.

But there is a secret weapon that can help you with that, and it’s called OKR. OKR is a goal-setting technique that helps teams establish measurable goals. In this article, I will show how product designers can use this technique.

This article is also available in a video format:

What is OKR

OKR stands for Objectives and Key Results. You might think that OKR has only two components, but in reality, there are three components:

• Objectives are what your team wants to accomplish. Objectives tell the team where to go. They are clear, memorable and inspiring goals.
• Key Results are how the team will measure results towards the objective. There should be 3 to 5 measurable outcomes that contribute to the achievement of an objective.
• Initiatives are actions that the team needs to complete to achieve key results.

How to write OKR

Here is a template that you can use to define OKR.

“We will OBJECTIVE as measured by KEY RESULTS via INITIATIVES.”

OKR vs KPI

Before we dive into individual parts of OKR, it’s essential to explain the difference between KPI and OKR. Many people believe that KPI and OKR are the same thing, which is untrue.

KPI and OKR serve different purposes.

KPI stands for key performance indicator, and it is a type of performance measurement. KPI helps us evaluate how well something is working in our organization. For example, we can track customer retention rate to understand how our product retains customers.

OKR helps to understand how the team can achieve something. For example, what should we do if we want to improve our customer retention rate from 60 to 80?

As you probably guessed, OKR and KPI can complement each other. KPI helps to monitor performance and identify areas where the team can do better. OKR can drive innovation and give you a bigger picture.

Writing objectives

An objective is a meaningful goal that you want to accomplish. It sets the direction for your organization or team.

Think of it as a destination on the map.

Questions to ask before writing Objectives:

• What is the most important area of our business that we need to focus on?
• Why does this goal matter to us?
• How would we write our objective into a statement that inspires our team?

Writing a good objective is not easy. Let’s compare two objectives.

Continue improving collaboration between design and customer success teams to improve the quality of customer service.

This objective is poorly defined because:

• It’s too wordy. 
• Unclear why this objective is important to the company. As a result, no motivation to achieve it.
• There is no information about when the team plans to achieve this objective.

Enhance product usability by the end of 2024.

This objective is well-defined because:

• It’s short and to the point.
• The benefit is clear to the company. Better usability means happier customers.
• Clear information when you want to achieve it.

When writing an objective, remember to keep it simple and focused. It should be like a slogan that fits on a t-shirt.

Writing OKRs is an iterative process — the more you write, the better you will become at it. That’s why I strongly urge focusing on progress over perfection.

Key Results

If Objectives are the aspirational words of a leader, Key Results are the specific numbers and metrics of a scientist. Key results help to specify the objective and measure the progress towards it.

Questions to ask before writing Key Results:

• How will we know if we’re successful?
• How can we measure the impact of our work?
• What metrics help us measure if we were successful?

Let’s compare two key results.

Make the product more usable by following user-centered design principles.

This key result is poorly defined because:

• It’s too vague
• No focus on impact.
• No specifics about how we will measure our results.

Achieve a customer usability score of 80 or above by the end of Q4 2024.

This key result is well-defined because:

• Customer usability score is a measure of usability. 
• We’ve defined the bar that we want to achieve. 
• And when we want to do it.

It’s important to remember that a good key result is an outcome, not output. Output is something that you do, like a task you need to complete. The outcome is a result of what you do.

Initiatives

Initiatives are the things your team will do to achieve the outcome.

Questions to ask before writing Initiatives

• What do we need to do in order to be successful?
• What can we get done in this period?
• Who needs to be involved?

When thinking about Key Results versus Initiatives, remember that key Results are the outcomes you want to happen, while initiatives are outputs — activities you’re performing to get there. Since we wanted to improve customer usability score, we need to invest in user research and usability testing.

When planning initiatives

Here are a few things that you need to consider when planning initiatives: 

• Make sure you have the necessary resources, time, and money, as well as stakeholder involvement.
• Focus. A big mistake many managers make when planning work is being overly ambitious. OKRs help the team maintain focus, but when the team tries to achieve a few objectives at the same time, it’s easy to lose focus. Because each objective may have many key initiatives, it is usually best to select only one objective per quarter and focus on it to achieve good results.
• Measure. It’s vital to define a set of metrics to monitor key initiatives. It will help you understand if you are moving in the right direction.

Best practices for OKR

1. When it comes to OKR, less is more. Don’t set too many OKRs. Avoid putting everything you’re doing in OKR. Focus on critical things first.
2. First, define strategy, then define OKR. If team members don’t understand the organization’s purpose and where it is headed, OKR will not motivate the team to do something. OKR simply won’t work because people will be doing their job without a sense of purpose and inspiration.
3. Transparency is key. Team goals should be visible to anyone in your organization. It will also boost awareness.

OpenAI recently released a long-awaited store for GPTs. Now, ChatGPT Plus subscribers can access a collection of GPTs created by different users. In this article, I want to review the downsides of the current version of the store.

This article is also available in a video format:

GPTs home page

When we navigate to the Explore GPTs section, we will see a page that invites us to discover GPTs that suit our needs. We see a search bar and some categories such as Top Picks, Dall-e, Writing, Productivity.

The first problem that we face is category selection. It’s unclear why Dall-e, Writing, Productivity, Research & Analysis are selected as the top-level categories. There are no categories like Finance, Travel, or Health. Below are the categories from AppStore, and you can see that the list of categories is more relevant to everyday activities.

 The naming of categories at the GPT Store is a bit confusing, too. The category “Dall-e” is basically image generation, but it has the name of the tool rather than the action that the user wants to perform using it.

If you look closely at the actual GPTs, you will notice another problem. You will need to read the text to understand what the GPT does for you. Apart from GPT icons, OpenAI doesn’t offer any visuals. And icons are not always self-explanatory. But reading the text is not the most enjoyable experience. It will require a lot of brainpower right from the start.

Searching a specific GPT

If we have a particular goal on our mind that we want to accomplish, for example, we want to find a finance advisor for ourselves, we will face another problem. Once you start typing your query, you will see that OpenAI offers recommendations based on the keywords you provided rather than analyzing your intention.

What I mean by that is when I say finance advisor, I want to see a list of finance apps. Tech Support Advisor or Jo Ellen Mathews results look irrelevant to my goal.

Apart from the GPT name, OpenAI doesn’t offer any information on what the GPT provided in the list is doing. So, I assume I need to spend time to figure out what the app does. At least for me, it usually leads to a lot of time invested in testing various GPTs before I find the one that I need.

There is another problem. It’s hard to figure out what other users think about the app. Take a look at the Apple AppStore for Yahoo Finance.

Apart from the visuals that communicate the purpose of the app, you see the rating, and you can read not only about the app itself but also what actual users think about it. And identify both the positive and negative sides of the product. 

OpenAI doesn’t offer that. So, there is no way to rely on other people’s options when evaluating the GPT.

Using a custom GPT

Let’s try the first GPT from the search results (I assume it’s the one that OpenAI thinks is the most relevant to my goal) and see how it will help us solve our problem.

Suppose I have 1000 dollars that I want to invest in stock, and I wonder if I should buy Tesla stock in 2024. Finance GPT provided a relatively long response, but it didn’t answer my question. Let’s force the GPT to give a precise answer — yes or no.

The GPT failed to do it. It’s true that finance advising is a tricky thing, and there is a legal risk for the tool to say you should or shouldn’t invest in something. But still, my goal is to get financial advice, not a general response about how to invest money.

I’ve tried many custom GPTs and want to say that most of the work is not really well. Many GPTs feel like half-baked apps that were created back in November 2023 when OpenAI announced custom GPTs. So, I think that OpenAI should review current GPTs and remove all mediocre apps from the store.

History of operations

ChatGPT doesn’t show the logo of a specific GPT in History. So, all results look the same to me, both results from native ChatGPT and custom GPTs. 

So, as you can see here, Finance GPT has its own logo, but the result doesn’t feature it.

Marketing and GPT discoverability

Marketing is an integral part of mobile stores like AppStore or Google Play, and both marketplaces offer a way to promote your mobile app. 

But right now, I don’t see any marketing tools in the GPT store. Consider that many people will try to create their GPS in the coming weeks; without efficient marketing tools baked into the store, there is simply no way to promote your product. 

So, even if you have a very useful GPT, you won’t be able to acquire many users.

In conclusion

I know that the GPT store is only making its first steps on the market, and it’s hard to release a perfect from the first attempt. I truly believe that the OpenAI team will do their best to make it perfect.

Netflix is a popular video streaming platform with over 200 million paid subscribers worldwide. This review covers the first steps of the user journey, such as landing, account setup, onboarding, and engagement.

This review is also available in a video format:

Landing page

The Netflix landing page is pretty minimal. The hero area features a value proposition along with essential details about using the service.

Statements “Watch anywhere” and “cancel anytime” increase the chance that the user will convert.

When you start to scroll, you go through a few sections that describe the benefits of using the product.

At the end of the page, you will find frequently asked questions and answers about Netflix.

Once you reach the end of the page, you will see a Get Started button, which is great because, at this point, you have more information to make an informed decision.

Pros and cons of Netflix landing page

Pros:

• It has a clear value proposition. You understand what this service is all about and what benefits you will get from it.
• FAQ section answers popular questions, so you don’t need to search for the answers anywhere on the web.

Cons: 

• There is no way to explore the collection of TV shows and movies. This can be a problem because you don’t know whether the show you want to watch is streaming on Netflix or not.
• Lack of information about the pricing at this point. Although you can check the FAQ on this page to learn about the pricing, Netflix doesn’t show the price prominently at the beginning. Since Netflix is a paid service, it’s essential to be transparent about pricing.

Account setup

Once you provide your email, you go through a 3-step setup process.

In the first step, you have to provide a password to create an account. You can choose if you want to get special offers in their email. There is no option to use social media signup. So, you cannot sign in to Netflix using your Google account. You have to use an old-fashioned email/password combination.

Next, you need to choose a plan. And Netflix is clear about the benefits of using the service, such as canceling anytime or unlimited watching on all your devices.

But the problem is that Netflix pre-selects the most expensive plan by default. It’s called Premium. Of course, it offers the best video quality. Yet, not all users are willing to spend extra money to get it.

Lastly, you will be asked to provide payment information. Depending on your location, you will have the option to pay using a credit card or PayPal.

If you choose the credit card option, you will see information about the plan you’ve selected and a call to action button, Start Membership.

Pros and cons of Netflix setup flow

Pros:

• Clear step-by-step process. At each step, you have to complete only one action, which makes the interaction really focused.
• Contextual information. Netflix provides all the essential details to help you make informed decisions.

Cons:

• No option to sign in using your social media account. You have to create a password and remember it.
• Paywall. Netflix doesn’t offer free trials. So once you sign in to Netflix, you start your paid membership.

Onboarding

Once you start a membership, Netflix will ask you to create your profile.

You need to provide your name and mention if you’re creating a profile for a kid.

Then, you will need to choose your language.

And select 3 shows or movies that you like. Netflix will use this information to provide personalized recommendations for you.

Once Netflix finds a collection of movies that you are supposed to like, it will guide you through the interface. Netflix uses contextual hints to explain how to interact with a product.

The final step is a modal window with information that you can watch your movies on all kinds of devices.

Pros and cons of Netflix onboarding

Pros:

• Simplicity. Netflix asks for the bare minimum information — your name, whether you’re a kid, your language, and 3 movies you like.

Cons:

• The kid option is unclear. At this point, it’s hard to tell what the difference between Kid and adult content will be. The user will learn that once they start using the service, but it might be unclear during the onboarding.
• Low efficiency of the contextual hints. Many users will simply pass by this type of onboarding because they want to start using a product as quickly as possible.

Engagement

Netflix has a relatively simple UI design. The hero section on the home page features the latest show of the service selected for you. In my case, it is a Crown. 

And then, when you start to scroll, you will browse through various collections of shows and movies. There are a lot of collections to browse, so your eye will scan a lot of thumbnails during the scrolling. Considering the fact that the layout is very monotonous, it will create a really uncomfortable experience.

Netflix tries to break the monotony of layout by adding sections like Top TV Shows In Your Country, but this feels more like a patch rather than a solution to this problem.

Another problem with this layout is that every collection is a horizontal scrolling list. It means that you can scroll the page in two different directions — vertically and horizontally.

All tougher, it makes it harder to choose the movie that you might want to watch if you don’t have a particular title in your mind. Because there are a lot of options to choose from and absolute freedom of choice, users often face a situation called decision paralysis.

When we have a lot of options to choose from, our decision time starts to increase exponentially.

Enough about layout design; let’s talk about personalization. Netflix attempts to help users decide what they want to watch next. Take a look at the cards with movies. 

When you hover the mouse over the movie or TV show thumbnail, you will see a nice preview of this show along with essential information about it. The most notable element in the card is the match score. You see green text like 97% match. This number is calculated based on your selection during the onboarding as well as your history of watching Netflix. The top part of the page — the hero section and the first few collections feature the movies with the highest match scores.

If you click on details for individual TV series, you will see a nice window with information about the show, its episodes, and More Like this section. Netflix thinks about the user journey holistically and wants to guide you to the next step — the next TV show or movie that you will want to watch.

Finally, let’s see if I can find a particular movie on Netflix. For example, I want to watch a movie called The Game. It’s a 1997 thriller directed by David Fincher.

Type The Game in the search field, and you will see a page full of movies, but I don’t see a movie, The Game.

Netflix suggests trying The Game, which is exactly what I’ve typed, by the way. But let’s click the link to see what happens.

Now Netflix tells me that there is no movie The Game in its collection. And it recommends movies and TV series that I might like to watch. The problem is there is no explanation of how this selection was created in the first place. I suppose that I see Crown because I mentioned that I liked Crown during onboarding and Flight Club because this movie was also directed by David Fincher, but as a user, I have to guess how Netflix created this recommendation.

Pros and cons of Netflix engagement

Pros:

• Content personalisation. Netflix collects and analyses information about user watch preferences and uses this to suggest content that they like.
• A holistic view of the user journey. Netflix aims to increase the chances that the user will switch to another show or movie after finishing the current one.

Cons:

• Decision paralysis. Netflix offers a lot of options to choose from and gives a user total freedom of choice. It ends up in situations where users spend a lot of time finding the movies that they want to watch.
• Not really an honest search. When you search for a particular movie, and Netflix doesn’t stream it, instead of saying that there is no movie in its collection, Netflix offers a lot of alternatives that you are supposed to like. Yet, the services don’t create a proper context of how this selection was created in the first place.

This is the end of our review. Let me know what you think about this format in the comments.

ChatGPT is not the only powerful AI chatbot available on the market. We have a range of great tools like Google Bard and Bing Chat, but there is a tool that can be even more powerful than ChatGPT in specific tasks. I’m talking about Claude.

Claude is a text chatbot created by the company Anthropic. This company was founded by former members of OpenAI. Claude recently became available in 95 countries. Previously, it was available only in the US and UK.

In this article, I will show what this tool is capable of. I’ll test it in a few popular scenarios, such as data analysis and writing code.

Conducting product/market analysis

Claud has more recent data than the ChatGPT free version. Its trading data is up to 2023. It makes this tool suitable for tasks like competitor analysis and market forecast.

Here, you can see a response to the tool provided to my question about mobile usage in 2023. I’ve asked to provide “General mobile phone usage statistics in 2023,” and the tool was able to provide this data.

Creating summaries for documents

One notable difference between the ChatGPT free version and Claude is that the latter allows you to upload a file along with the prompt. ChatGPT also allows you to attach files, but this feature is only available in ChatGPT Plus, a paid version of the tool.

Claude can handle up to 75 thousand words as input. It’s more than three times more than ChatGPT can handle. It basically means that you can submit a fairly large document and ask Claude to summarize it.

Here is what will happen when I attach a document (overview of large language models) and simply press Start a new chat. If you don’t provide any prompt along with your file, Claude will analyze the file and share the summary of this analysis with you.

To make the summary a bit more specific, you can provide a prompt like “What are the top 5 key points from this document” along with your file.

You can also ask follow-up questions about this document, such as in my case, I decided to ask about the key innovation elements of the GPT model. As you can see, Claude was able to extract this information from the document I’ve provided.

Extracting specific information from multiple documents at a time

You can attach multiple documents at a time and ask Claude to summarize them or extract specific information. For example, I’ve submitted two research papers about the metaverse economy; one was created by JP Morgan, and another by Deloitte and asked a basic question about the experience that will work best for the metaverse.

The tool used the papers and identified factors that will have a positive impact on the metaverse experience, but it’s not exactly what I need since I’m interested in business models.

It’s true that I should have mentioned keywords like “business” in the prompt, but I believe that the tool should be able to figure out my intention based on the documents that I’ve provided since the documents were all about the state of the metaverse & business opportunities in it.

Let’s change the prompt and ask the tool to identify opportunities for product developers who want to build a metaverse experience in the same dialog. As you can see, the tool likely ignored the papers I’ve submitted and provided a very general response to my question.

A simple trick can help you to force the tool to use the documents you’ve submitted to provide an answer. You need to start with your question, such as “What is the expected evaluation of the metaverse” and add a phrase like “according to provided reports.” And as you can see, the tool used the documents when generating a repose.

Generating code

I will ask the tool to create a bar chart to visualize the US population from 1990 to 2000. The tool created a Python sample even though I didn’t mention the keyword Python. And from the first glance, the code is fine. It imports the most popular library for creating plots for Python, which is matplotlib. And Claude has also explained the logic behind this code. This approach is similar to the one that ChatGPT uses.

Claude limitations

There are a few limitations that you need to take into account when working with Claude.

Not able to generate images

The tool cannot generate images. Here is what happens when I ask Claude to draw a kitten. It simply provides a text description of the kitten.

Limited number of messages during the day

Claude limits the number of messages you can submit during the day. If your conversations are relatively short, you can expect to send around 100 messages every 8 hours. But, since I used a tool to summarize large documents, I was able to submit 5 or 6 messages. After you reach your limit, you will see that there are no free messages, and you can either wait until a specific moment or sign up for a pro version of this tool.

Pricing

Claude Pro version costs $20 per month, and this price is the same as the ChatGPT Plus subscription.

Pros and cons

The benefits of Claude are a larger number of tokens (you have 100k of tokens, which is around 75 thousand words), which is the size of a fairly large document. You can provide files along with your prompt, and Claude has training data up to 2023.

The downsides are a lack of text-to-image generation and a limited free plan. The number of messages you can submit during the day is limited.

Midjourney is an excellent AI image generator, but it has one crucial downside. It’s a free-to-use tool, and the basic plan costs $10 per month. If you use AI image generators occasionally, you likely want to see a free alternative version. And Ideogram AI can be a good option for you.

In this article, I want to explore what Ideogram is capable of and test it in six popular scenarios.

Ideogram AI features

Ideogram is a text-to-image generation tool with an easy-to-use interface. Apart from the input field for a prompt, it allows you to choose the style of your imagery and change the aspect ratio.

You can also remix generated images by applying a new visual style. A new style will be applied to the generated image.

Testing Ideogram

I’ve tested Ideogram for six common tasks

1. Realistic images

The first task is generating realistic images. The prompt I submit is

Ultra realistic photo of a young woman

Like many other AI image generators, Ideogram generates a set of 4 images. When you look closer to the images, you will likely notice that 2 out of 4 images have visual defects. For example, the first image has a problem with the eyes and fingers. The second has a problem with hair.

The other two images don’t have any visible defects, and the last image looks very crisp. But overall, the level of realism is average compared to Midjourney.

2. Creating illustrations

Illustration of a cute dog

One great thing about Ideogram is that the tool generates all images in different visual styles. Unlike tools like Adobe Firefly, which basically create a set of similar images, Ideogram varies the style so you can choose the illustration you like.

Looking closer, you will also notice that some images have minor visual defects. For example, some extra elements on the paw of the first dog. Broken heart in this illustration. The last two images are decent. And I partially like the last image, which looks like a sticker.

3. Generating logos and vector images

Ideogram can generate logos and vector images and the prompt that I use is

Flat vector image of a rocket on a white background

I intentionally mentioned that I wanted to see my image on a white background because it would be easy for me to remove it and turn the image into a vector. Unfortunately, the tool ignored my command and used a grey background for all images. The quality of vector images is okay, but nothing really special.

4. Creating graphic pattern

We might want to use AI image generator to create a graphic pattern to use as a background for web page design.

Colourful hexagon pattern

Even at first glance, all the graphics the tool generated are not good. They have misalignment for visual elements and extra noise. We can’t use such backgrounds without modifying them in Photoshop.

5. Images with text

Ideogram can help generate images with text.

fox holding a sign “let’s get wild”

All images that the tool generates have human-readable text without any gibberish. However, some images have minor visual defects; the last image misses one world.

6. Testing censorship

Lastly, let’s see if we can use Ideogram to generate images of famous people. Tools like Dalee-3 are very censored, so it’s impossible to use them to generate images of famous people or mascots like Nintendo’s Mario.

Elvis Presley is singing

Surprisingly, Ideogram was able to generate an image of Elvis Presley. But as you can see here, all images have visual defects — problems with hands, composition of elements, etc.

Pros and cons of Ideogram

Let‘s start with good things. Ideogram is a free-to-use AI image generator that allows you to generate images with text, and it even allows you to create images of celebrities and mascots like Mario.

At the same time, the level of realism of images is average in comparison to other AI generators. Many images have visual defects, such as problems with hands and visual noise. The prompt accuracy is also not very good, meaning that the tool can ignore some parts of the prompt.

Overall, I would say that Ideogram cannot replace Midjourney, but it can be helpful when you need to use an AI image generator and don’t want to pay for it.

What surprises me is how much attention is directed to ChatGPT and how little we’re talking about Bing Chat. Bing Chat gives you access to the GPT-4 model for free. It’s also available as a sidebar in Microsoft Edge, which makes it possible to use this tool to analyze contextual information (such as specific web pages).

This article shares ideas on how product designers can use Bing Chat. I will discuss both the advantages & limitations of this tool and share ten specific prompts that can be helpful for your desing process.

What Bing Chat can help us with

Bing Chat is a multi-modal tool, meaning that it can generate text and images. It uses the power of the GPT-4 model (yes, the same model that ChatGPT Pro uses) and Dalle-3 image generator and offers AI vision capabilities that allow us to use AI to analyze images such as charts.

Bing Chat has one key advantage over ChatGPT — we can see a link to the source of this information. So unlike ChatGPT, which responds without any references, and we have to either trust it or go to the search engine and validate it manually, Bing does this work for you. It also features a list of resources that can be helpful for our further analysis.

Another great thing about this tool is that it allows you to choose the style of your conversation with the chatbot. It can be more creative or more precise, depending on your goals. When you choose the option creative, you give the tool more freedom on how it can respond, so you likely see that your prompt is used only as a reference when AI provides a result.

Bing Chat limitations

Bing Chat has a character limitation for your prompt. You can type a prompt up to 2000 characters, which is the size of a small text document. So basically, you can submit a few paragraphs of text and ask Bing to analyze them.

Another important limitation the tool has is the number of requests per single session — it is possible to ask up to 5 questions in a single session. What does it mean? It means that we can ask four more follow-up questions before the chat ends. And if we need to ask more than five questions, we need to start another session with Bing.

How product designers can use this tool

I typically use Bing Chat for product research as an assistant that helps me move fast during the first steps of my research.

Here is a collection of prompts that I found useful for my product research.

1. Generating product ideas

We can ask Bing to generate product design ideas for specific types of products. In my case, it will be a food ordering app.

Generate 5 new product ideas for a food delivery app

As you can see here, Bing generates an interesting list of ideas, including features like food waste reduction and interactive cooking classes. It is definitely helpful information for product creators.

2. Conducting SWOT analysis

SWOT analysis is creating a matrix with four sections — strengths, weaknesses, opportunities, and threats. It’s a foundational tool in marketing and product design that allows us to evaluate your product strengths and weaknesses.

Conduct a SWOT analysis of the Uber Eats

Once again, Bing supported every major statement with a link to a specific resource. And provides a list of helpful resources that we can visit to learn the topic.

3. Creating a user persona

One of the challenges of working with AI tools like ChatGPT and Bing Chat is that sometimes, the tool fails to understand your intention.

Create a user persona of a middle-aged office worker who wants to eat healthy food but doesn’t have time to cook

As a result, you need to write a follow-up question to clarify your goal and receive a relevant response.

Use this information to create a persona that I can use in product design

In this case, I had to write a follow-up prompt that clearly stated what I wanted to achieve. And Bing generated the persona for me. Quite interestingly, the actual persona description is at the end of the second response.

4. Learning about user pain points

There are two ways we can collect insights about a product domain. We can either mention a particular product (like we did when we mentioned Uber Eats for SWOT analysis) or specify a category of products. And when it comes to learning about pain points, I believe it’s better to specify a whole category.

What are the most common customer pain points for food delivery apps?

Bing provided a very detailed response with statistical data from specific sources.

5. Generating onboarding ideas

The great thing about Bing is that it uses all the information from your current session when answering your question.

Create onboarding flow for new users of food delivery app

For example, here you can see that Bing knows that we’ve created a persona before, so it’s kinda know our ultimate goal.

Again, the fact that Bing also provides references to the sources makes the onboarding recommendations more credible.

6. Competitor analysis

Bing can structure the information in a specific way; for example, create a companion table with your primary competitors.

Create a product comparison table between Uber Eats and its main competitors

The table can be easily exported to Excel, and most data points mentioned in this table have backlinks to original sources of information.

7. Creating market forecast

Bing is a search engine, and it has a connection to the internet, meaning that it has up-to-date information, and that’s why we can safely use it for things like market forecasts or trend prediction.

What is the projected growth rate for the ghost kitchens over the next 10 years?

8. Suggesting metrics and KPIs

When building a new product, you will likely need to define key metrics upfront before starting an actual design process. And Bing Chat can help you with that.

How can we measure the success of food delivery apps?

I would say that the list of metrics that the tool provides is very conventional for this category of products (food delivery apps). You can expect to measure delivery time and delivery success rate when you’re building a food delivery mobile app.

9. Learning law and regulations

We need to comply with government regulations when we’re building products. Bing provides a summary of the key policies that we need to check before releasing a product on the market.

What are the key regulations for food delivery apps in the US?

10. Using AI vision to analyze visuals

What makes Bing really cool is that it supports AI vision, meaning that we can submit a specific image and ask Bing to analyze it.

For example, here you can see I’ve submitted a table that compares popular food delivery apps, and I asked Bing to give me a list of companies from this table that offer free delivery.

Not only was the tool able to extract this information, but it also validated it so the information about Thrive Market has a link to the source.

The release of DALL·E 3 is one of the most important things in the AI image generation world. DALL·E 3 is the latest version of the image generator from OpenAI, the creator of ChatGPT, and it’s now available in Bing. DALL·E 3 significantly improved images and text interpretation capabilities over its predecessor, DALL·E 2, and now it’s available both in Bing Chat and Bing Image Creator.

What can you do using DALL·E 3?

DALL·E 3 allows you to generate different types of images that you can freely use in your design process. You own the images you create with DALL·E, including the right to reprint, sell, and merchandise.

Here are a few examples of images I’ve generated using the tool, along with the prompts I’ve used:

Reasons to use DALL·E 3?

Why should we use DALL·E 3 when we already have powerful tools like Midjourney and Stable Diffusion? There are three reasons for that.

First, DALL·E 3 is free to use; you don’t need to pay any money to access the tool. Bing offers daily credits that we can use to generate images. But the good news is that the generation doesn’t stop when we use all our daily credits; it simply becomes slower.

Second, DALL·E 3 can be better at specific tasks, such as adding text to images and combining characters in the same image.

Adding text to images

The OpenAI team mentioned that the tool is good at adding text to images. Here is an example of a photo of a person holding a text sign I created using the tool. You can notice that the text is not always not perfect, but it’s better than the text generated by Midjourney.

Combining two characters on the same image

I’ve used the tool to combine Batman and Spiderman in the same illustration. Quite interestingly, DALL·E 3 framed the conversation between Batman and Spiderman into a comix.

Integration in ChatGPT

What makes DALL·E 3 so powerful is that OpenAI integrates this tool into ChatGPT. It will be possible to generate images right in chat using the GPT vision model to analyze the image generated by DALL·E 3 and alter the generated images.

Downsides of DALL·E 3

The tool has a few downsides. The image generation process is a bit slow right now, and sometimes, it might end up with an error message saying that the image generation is unavailable due to high demand.

Another downside is that you cannot customize images. Image generator offers a Customise option, but it leads to Microsoft Designer, where you can use a generated image as additional input for your prompt. You cannot alter the image (for example, remove some objects from the canvas) or change the image aspect ratio in Image Creator.

‎Bard is a chat-based AI tool from Google. It is one of the ChatGPT competitors that offers some unique features. In this article, I want to review the top 3 features I find the most valuable for product designers.

1. Validate the output provided by AI

Lack of links to the sources of information is one of the problems with ChatGPT. When submitting the prompt to ChatGPT and it generates a response, there is no way to validate it automatically. You must do it manually — go to Google, submit the information, and search the sources. All AI tools are prone to hallucination, and it is relatively easy to fall into the trap of misleading information when you skip the validation phase.

Bard offers a feature called Double-check a response. When you use this feature, Bard will provide links and citations to where it found the content.

The response that Google Bard provides is validated on the fly, and the tool highlights the content in green if it finds a citation for the information.

At the same time, the information that cannot be validated is highlighted in orange. If you use Bard for tasks like market research, you can remove the claims that could not be validated before sending the findings to your team members.

2. Submit the image along with the text prompt

At the time of writing this article, the ChatGPT free tier supports only text input. You will likely be able to submit the image along with a text prompt to ChatGPT in the future, but right now, ChatGPT is a text-based chat.

Google Bard allows you to provide images along with the text prompt, and it uses the tool called Google Lens to extract information from the image. For example, you can add the image of the UI you would like to design in Figma and ask Bard to provide a step-by-step guide on how to create it.

3. Generate a public link to the chat so you can share it with other people

The only way to share information generated by ChatGPT is to copy the response as plain text. ChatGPT offers a quick shortcut that allows you to do it right in the context of interaction.

While this approach allows you to copy the content, it’s far from the perfect UX because you have to find a way to share the content — create an online document, write an email, etc. And even when you do that, the other person will receive this information as plain text.

Bard allows you to create a link to the chat and share it with anyone so that they can continue the conversation with AI.

For example, the person who receives a link to a usability testing questionnaire can ask Bard follow-up questions to fine-tune the output generated by AI.

Text-to-image AI tools like Midjourney, Dalle-2, and Stable Diffusion can generate images from plain text. These days the internet is full of AI-generated images, but the question is, “Can text-to-image tools can be used for UI design?”

In this article, we will see how AI tools can deal with regular UI tasks such as creating:

• UI screens
• App icons
• Product images
• Logos
• Mascots

To make our analysis more specific, we will generate UI assets for a particular product type (food delivery app).

Disclaimer: This article doesn’t try to convince you to use AI tools. It’s a pure experiment that shows what AI tools are capable of and what UI assets they can generate.

This article is also available in a video format:

A quick introduction to MidjourneyMidjourney is an online text-to-image tool. To use Midjourney, you need to join Discord. Visit https://www.midjourney.com/, click Join the Beta and follow the instructions to create your account.

Once you do that, you can join any channel with the prefix “newbies-” and start generating images by typing

/imagine [what you want to generate]

UI screens

The most important thing that we need to do when working with AI tools is to write clear prompts. A prompt is a text string that you submit to an AI tool so that it can generate an image for you. The prompt should clearly specify what you want to see.

The better you articulate your intention, the more relevant result you will get from AI.

When writing a prompt, think about how you will describe your design to an actual person and write down an exact sentence. Turn this sentence into tokens (short phrases separated by a comma).

The tokenized format is better for Midjourney because it helps you trim all unnecessary words while keeping essential contextual information.

For example, in the context of a food delivery app, the tokenized prompt might be:

/imagine high-quality UI design, food delivery mobile app, trending
on Dribbble, Behance

By default, Midjourney generates four options for you to choose from. You can upscale options using the command U[number]. Let’s see what the upscaled version of the first design will look like.

If you compare the upscaled version to the previous version, you might notice that Midjourney doesn’t only increase the size and quality of the image but it also slightly modifies the layout. For example, the upscaled version doesn’t feature a red button in the top-left corner of the screen.

You likely notice two significant problems in the images generated by the tool: gibberish text and corrupted food preview images. Unfortunately, these problems are relevant not only for UI screens but for many kinds of images generated by Midjourney.

Our first take wasn’t perfect, so you might wonder how to modify our prompt to achieve better results. It’s worth saying that there is no optimal strategy for writing prompts. You need to experiment with different tokens to find an optimal solution. That’s why you should aim for at least a few tries before you come up with a decent output.

Let’s modify our prompt to see how the tool will react to it. This time we will mention “Figma” because we will likely want to use the assets in the design tool.

/imagine food delivery app, user interface, Figma, Behance, HQ, 4K, clean UI

Let’s see what the upscaled version of the first design will look like.

This version looks more like something we can use in design. The layout has a clear visual hierarchy of elements.

Let’s make another take on UI design using Midjourney. This time we will remove tokens “Dribbble” and “Behance,” leaving only “Figma” but also add a parameter “ — q 2” to increase the quality of the images.

/imagine [what you want to generate] -- q <number>

Here is our prompt

/imagine Food delivery app, mobile app, user interface, Figma, HQ, 4K, clean UI -- q 2

The third option looks promising. Let’s upscale it to see more details.

You can see that the results that Midjourney generates, even after the third try, cannot be used without refinement from a human designer. Nevertheless, this raw output can be very helpful during the early stages of the design process because it serve a visual inspiration. For example, it’s relatively easy to create a moodboard of AI-generated images and use it as a reference for visual designers to create a proper visual language.

It’s also vital to mention that we’ve used a very general prompt for UI design. If we want to see a particular screen (such as a restaurant search page), we should clearly specify it in a prompt as a token so that the system can get us a more relevant result.

Imagery

Imagery is photos and illustrations used in various parts of a product. Imagery is integral to user experience because it communicates essential information to users and sets the right mood. Finding the right imagery can be challenging during product design, so let’s see if Midjourney can help us with that.

App Icon

An app icon is a simple yet very powerful element because it communicates the idea of what the app is all about as well as creates a first impression to users. Even before users start to use a product, the icon is the first thing they see, and its level of craftsmanship directly impacts how users will perceive a product.

Since we are designing a mobile app, we need to clearly specify the token “iOS app” as well as mention that we want to see high-quality visual assets (tokens “high resolution” and “high quality”). And because our app is about food delivery, we might want to mention one particular type of food (“burger”).

/imagine icon for iOS app in high resolution, burger, high quality, HQ -- q 2

As you can see, Midjourney generated a set of ultra-realistic icons. Let’s upscale the second option.

The style that Midjourney uses for icons is great, but it might not work for an actual project because many designers prefer to use minimalist icons. Let’s give it a second try using a slightly different prompt. We need to mention tokens like “minimalism” and “flat design” to signal AI that we want to see simple icons.

/imagine icon for iOS app in high resolution, burger, high quality, minimalism, flat design -- q 2

Now the output looks like something we can use in a real project. Let’s upscale the first option.

Product images

Product images are images that we will show on product pages. Since we design a food ordering app, our image will be photos or illustrations of food. Let’s ask Midjourney to generate images for us:

/imagine burger, high quality picture, high resolution, HQ, 4K -- q 2

Midjourney generated solid images, but they have one major drawback — they are placed on a dark background, and this can be a problem because it will require extra work to extract an object from the background. But we can use a simple trick — ask Midjourney to use a transparent or white background:

/imagine burger on a transparent background, high quality, high resolution, 4K -- q 2

The first image looks great; let’s upscale it.

Now, all we need to do is to remove the background from this image. To do that, we can use online services like Remove.bg. Upload the image to removeb.bg and download the ready-to-use visual asset.

Promo illustrations

Promo illustrations are versatile visual assets that can be used inside the app or on a promo website. Let’s see if Midjourney can create a decent illustration. We want to communicate the idea that our food is delicious. So the prompt we provide is:

/imagine happy man eating a burger, high resolution, high quality illustration

The cool thing is that the images are generated in different styles — from watercolor painting to highly realistic photo illustration. But it is evident that Midjourney had a hard time pairing happiness with eating. Let’s upscale the second image to see if the upscaled version will look better.

You likely notice another problem with the images that Midjourney generates — the problem with legs and fingers. The tool often adds an extra leg or finger when it generates images of people. Yet, in our example, the tool removed fingers and modified the human hand to make it look like a claw.

Logo

A logo is a core element of branding. When it comes to logo design, we have to ensure its meaning is clear to customers and the logo itself is memorable. The great thing about Midjourney is you can ask the tool to generate images in a particular style. When it comes to logo design, we can follow the style of famous graphic designers such as Saul Bass, who is known for designing posters for movies like Vertigo (directed by Alfred Hitchcock).

Here is what Midjourney creates for a prompt:

/imagine minimal logo of a burger, graphic style of Saul Bass

Midjourney indeed uses some elements of Saul Bass’s style, such as typefaces and color scheme, yet the final result doesn’t look like something from his works.

Mascot

A mascot is a graphical object used to represent a business. A well-crafted mascot helps people remember a brand and communicates a certain mood. Let’s ask Midjourney to generate an image of a mascot for a food delivery company mentioning a particular style (Japanese).

/imagine simple mascot for a food delivery company, Japanese style

Midjourney generated a few nice examples of mascots. But only one example can be used with minimal modification (fourth option).

Can Midjourney replace UI designers?

No. At least not right now. The outputs the tools generate are very rough and often require refinement from designers. Midjourney is also very dependent on the prompt (the quality of output it provides can depend drastically on the prompts you provide).

Does it mean this tool has no place in the designer’s toolkit? No. The tool can be beneficial during the early phases of a product design process, such as ideation and visual exploration. Midjourney can be especially useful for moodboarding. It can help a team to compare a lot of visual styles before they can commit to one particular style.

User research is a fundamental part of the design process. The more time and energy a product team invests in user research, the higher the chances of releasing a commercially successful product.

In this article, I will explore whether ChatGPT can be helpful for user and market research. To make this article more specific, I will use ChatGPT to conduct research for a new hotel booking mobile app focussed on the US market.

Disclaimer: ChatGPT cannot be used as a complete replacement for human user researchers. It’s best suited as a digital assistant that can speed up some research activities. Still, product creators should always validate the information the tool provides and spend time working with a product’s target audience to understand real users.

This article is also available in a video format:

1. Competitor analysis

Competitor analysis is an integral part of product design. To design a good product, we must understand who our competitors are and what they do right and wrong.

Identify common issues that all products in the niche have

Let’s start with what all our competitors do wrong first and ask ChatGPT to summarize common issues that mobile users face when they use hotel booking apps.

As you can see, some points that ChatGPT provided are way too general. For example, “Poor User Interface” doesn’t uncover specific UI issues. But a few points that ChatGPT mentioned in this list are super helpful — payment issues and limited customer support. Based on my experience creating booking apps, those are two major areas of friction for many users.

Identify our direct competitors

Now let’s ask ChatGPT to provide a list of direct competitors in our business niche. To make a list more specific, let’s mention in our prompt that we focus primarily on the US market.

Once we have a list of competitors, we can ask ChatGPT a follow-up question to identify the market share that each competitor has.

As you can see, Booking.com and Expedia are two major players in our market. We need to learn what customers like and dislike about Booking.com and Expedia, and we can ask ChatGPT to provide this information. Let’s start with things customers like about the product:

ChatGPT mentioned a few great points, such as transparent pricing and customer reviews and rating. If you have experience using services like Airbnb, you know why transparent pricing is extremely important. Let’s ask ChatGPT to summarize the downsides of using Booking.com

Cancellation policies and limited customer service are two areas in which Booking.com have problems. We need to consider those areas when building our product.

2. Define key business metrics

Before creating a design for our product, we need to define key business metrics. The metrics we set during this step will help us evaluate our design decisions. Let’s see if ChatGPT can be helpful for that.

As you can see, the list provided by ChatGPT contains a few standard metrics, such as retention rate, customer acquisition cost, lifetime value, and NPS. Most for-profit apps rely on those metrics when they evaluate their performance. At the same time, the tool also mentioned a few specific metrics, such as Bookings and Average Booking Value.

3. Identify your users and their behavior

It’s time to learn who will use our product. The better we understand the target audience and their needs, the more chances to create a user-centered design.

Creating user persona

A user persona is an archetype of an ideal user. Product creators use personas to evaluate design decisions.

Let’s ask ChatGPT to create a few user personas for our product.

ChatGPT created 3 user personas — business traveler, budget travler and luxury travler. For each persona, it specified demographics, needs & wants and provided relevant contextual information. Since our app will have one primary persona and a few secondary personas, let’s ask ChatGPT to figure out what will be a primary persona from this list.

Identifying popular travel destinations for users

Along with user persona, popular travel destinations will give us a better idea of how users will use a product. Since we aim for the US market, its vital to identify popular travel destinatiosn in the US.

One cool thing about the response that ChatGPT provided is that it referenced the source (Statista) when it listed popular destinations. We can ask a follow-up question to understand how much money customers spend for each location.

Creating user journey

Once we know who will use our app (our user personas), we need to know how they want to use it. It’s time to dive into the user journey and identify specific steps users will follow when interacting with a product. User journey will give us a birds-eye view of a product experience so that we can use this information when designing an actual application.

Let’s ask ChatGPT to design a user journey.

Storytelling

Storytelling is an essential part of the design process. User personas and user journeys are significant design artifacts that communicate what we know about our users and their habits, but stories can make the experience feel real.

Let’s ask ChatGPT to write a story about a person who uses our product.

Writing use cases for our app

Use cases are specific tasks that users want to achieve using a product. When it comes to a hotel booking app, the primary use case is booking a hotel in a specific area. Let’s see if ChatGPT can help us identify other use cases.

As you can see, the list that ChatGPT provided is not exactly use cases but rather an overview of the app’s functionality. But this information can be helpful during design process.

Identifying interaction patterns that work for our target audience

Last but not least, let’s see if ChatGPT can help us identify interaction patterns that work for our target audience.

Suggesting techniques that can evoke positive emotions

When it comes to product design, it’s essential to make the product both functional and pleasurable for users. Let’s ask ChatGPT to suggest a few ideas to make the user experience more enjoyable.

As you can see, ChatGPT provided a very general list of recommendations that includes using visually appealing images, storytelling and using the right color schemes. Perhaps the only notable thing about this list is the point about personalization. It’s definitely possible to use past booking history and preferred amenities to suggest relevant booking options to users.

4. Reaching the target audience for insights

No matter how much time we invest in user research, we need to validate all our insights about the user and their behavior by conducting interviews and surveys with people who represents our target audience.

Questions for a user interview

Since we know our target user (we’ve identified user persona before), we need to hire people who match these criteria and ask them questions about their booking epxiernece. Let’s see if ChatGPT can help us find the right questions.

Overall, ChatGPT created a solid list of questions that can be used for user research.

AI is in the spotlight in 2023. ChatGPT, released on Nov 30, 2022, is the first app that got 100 million users in two months. AI is breaking into the mainstream because people see practical value in using AI tools. AI tools make work easier, faster, and more efficient. At the same time, AI tools are far from perfect, and in this article, I want to explore the dark side of AI tools and discuss four issues that the tools have.

1. AI tools are prone to hallucination

AI tools are still prone to errors and mistakes. Hallucination is when you ask ChatGPT a particular question, and it provides an incorrect answer. The system rarely says that it doesn’t know something and can write plausible-sounding but incorrect answers. Only after the validation may you find that the answer AI provided was completely wrong.

In many cases, validation cannot be done automatically — it requires having human moderators, people who have relevant expertise in a field, and the time to do the work. Even when you have a large team of moderators, the team might not have the capacity to analyze AI-generated output. No wonder why large resources like Stack Overflow started to ban answers provided by ChatGPT due to many of them being wrong.

Stack Overflow Bans ChatGPT For Constantly Giving Wrong Answers

One of the internet’s largest coding resources has temporarily banned the AI chatbot after users answered programming questions with its responses.

VICEJanus Rose

It’s vital to understand why hallucination occurs in the first place. Unlike tools like Apple Siri that use the internet to find the answers (i.e., a result that Siri provides to the user is basically a reference of an article or video), ChatGPT constructs the answer on-fly word by word, choosing the most likely token that comes next in the sentence. And the model that ChatGPT uses is rooted in prediction (large language models like GPT predict the next best token). So it’s nearly impossible to completely prevent hallucinations of AI, but it is definitely possible to fine-tune the model to minimize the areas where hallucinations occur.

The context in which hallucination occurs is also essential. If a hallucination happens in a business context, the business might easily lose money following incorrect advice that it gotfrom AI. But it’s way too dangerous when it happens in a medical context because it will put human lives in danger. OpenAI introduced a safety mechanism in ChatGPT that prevents users from asking questions about their health.

Yet, this mechanism is not idead and it’s possible to trick the system by slightly altering the prompt.

2. AI tools and the problem of copyright and intellectual property

Both text-to-text tools like ChatGPT and text-to-image tools like Midjourney and Stable Diffusion trained their models on massive quantities of data, basically all publicly available information on the internet. But when the tools generate the response to the prompt, they don’t reference the original sources.

The problem becomes even more noticeable when you use text-to-image tools like Stable Diffusion and Midjourney. The systems are trained on millions of images around the internet. And when you work with text-to-image AI tools for quite some time, you notice that they often replicate existing illustration styles.

No wonder why Stability AI, the company behind Stable Diffusion, was sued by Getty Images, that alleged Stability AI of using millions of images from their photo bank to train the model and didn’t pay a single dollar for it.

Getty Images is suing the creators of AI art tool Stable Diffusion for scraping its content

Getty Images claims Stability AI ‘unlawfully’ scraped millions of images from its site. It’s a significant escalation in the developing legal battles between generative AI firms and content creators.

The VergeJames Vincent

AI tools will likely have to provide references to original sources along with the output. For example, when Microsoft introduced GPT-powered chat in Bing, they added a sub-section called “Learn more” that cites sources the AI tool used to generate the response.

3. AI tools can help people manipulate other people

We tend to think we’re unique in consuming information and evaluating ideas. In reality, we typically follow similar or even the same approaches as many other people in our niche, be it management, design, hiring, etc. We all follow certain patterns, and AI tools are very good at understanding patterns. Plus, tools like ChatGPT were intentionally designed to imitate human conversation.

As a result, the tool has an excellent ability to generate texts that communicate a specific message. For instance, with the right prompt, ChatGPT can write a cover letter that will sound much better than one written by a human.

Large corporations try to mitigate this negative behavior. For example, Microsoft doesn’t allow users to generate personal texts like cover letters using AI chat integrated into Bing.

4. AI tools make it harder for people to understand what is real and what is fake

The rapid evolution of AI technology raises an important question: How will we govern the technology? If we give everyone the power to do whatever they want with technology, we can easily face a huge problem: many users will use the technology for harm. For example, scammers can use AI to copy the voices of people we know (our relatives and friends) and trick us. Imagine receiving a call from someone who sounds exactly like your close friend, and this person asks you to do something urgently (i.e., transfer X amount of money). For the same very reason, we might see a new wave of misinformation (fake news).

We’re entering the era of synthetic reality — reality in which hard to tell what is real and what’s fake.

Unfortunately, government policies developed much slower speed than AI technology progress. If this trend remains the same, we might end up living in a very dark world in which we won’t be able to understand what is real and what is fake.

People choose to use a particular product because they believe it will help them achieve their goal(s). The more popular product becomes, the more tempting its for product creators to add extra features to the product.

But adding more features to the product doesn’t always lead to more value to end users.

Extra features often cause feature creep — a situation when a product becomes overloaded with too many unnecessary features. But how to identify the key features that provide the most value? A feature audit template can help you with that.

Feature audit template

A feature audit template can help you prioritize features according to the frequency of use among your target audience. The template has two dimensions:

• On X-scale, you can see how many people use a feature (a few people, some people, most people, and all of the people).
• On the Y-scale, you can see the frequency of use (none of the time, some of the time, most of the time, and all of the time).

In the perfect situation, all features of your product should be located at the top right corner of the template (all of the people use them all of the time).

Here is an example of the template for the Apple iOS Camera app:

As you can see, all of the users use iOS cameras to take photos, and just a few people adjust the aperture.

When to conduct a feature audit

Roadmap planning is the ideal moment for conducting a feature audit. When you discuss what features you need to focus on in your design, it’s better to stop and think about how people actually use your product. Our hypothesis about features and the value they bring to users often might not match with actual user behavior. The right way to evaluate your design is to:

• Identify core features in your product. Identify features that you think are valuable to your users.
• Track the core feature usage over time to understand the frequency and adoption rate. Use qualitative and quantitive data to understand how real people actually use your product and what features they find valuable.
• Move features to the right quadrant in the feature audit template.
• Analyze the feature audit document you’ve created.

What to do with features that perform poorly?

What to do with the features that your customers are not using? Does it automatically mean that you should remove them? No always. The fact that users don’t use a particular feature doesn’t mean that it is useless.

Here are two popular scenarios that you should consider:

• The feature can perform poorly because of poor discoverability and findability in your user interface. Your users might not know that the feature exists in the first place or might not be able to find it. You can try to increase the adoption rate of the feature by making it more visible to users (i.e., adding visual cues in the interface).
• The feature itself might be poorly designed. Users tend to avoid using such features after the first negative try. The only way to solve this problem is to improve the design of your product.

Feature audit template for Figma

Here is a free feature audit template that I’ve created for Figma:

Content is an integral part of product design. And just like any other product element, content should be well-crafted. The better content is designed, the better experience users will have with your product. UX writers are professionals responsible for writing micro and macro copy for the user interface.

In this article, we will explore whether it is possible to make the life of UX writers easier using ChatGPT. Below is a list of examples of UX writing for the eCommerce website that sells mountain bikes.

1. Call to action button label

The call to action button is one of the most critical elements on the eCommerce website because it directly impacts site conversion. Not only should the button be easily noticeable, but it also should have a proper actionable label. The label should persuade users to click/tap the button. A good label shouldn’t be long — up to 14 characters.

ChatGPT generated a very interesting list of examples, and what’s cool about the options is that the tool added an exclamation point “!” at the end of the label, reinforcing its meaning.

2. Order placed message

The message customers see when they place an order should make them feel great about their decision. And it is possible to achieve this goal by doing two simple things — saying “thank you for choosing us” and inviting customers to try a newly purchased product.

ChatGPT generated a very solid list of examples, and I like the “Get ready to escape the everyday and explore the mountains. Enjoy the ride!”

3. Invitation to review a product

When you ask customers to write a review, you should understand that most customers won’t be motivated to do that, especially if they don’t understand the reason. That’s why it’s vital to communicate the importance of this task.

ChatGPT generated an average invitation, both very long and not very inviting. But the tool also baked a very excellent idea into it — “Your review will be a valuable resource for other customers.” Knowing that your action can help others can motivate some users to write a review.

4. Incorrect password

The error messages that users see when interacting with your service can significantly impact how they feel about it. There are a few popular places where eCommerce service can show error messages, and incorrect password is one of them.

ChatGPT generated a standard message that clearly describes the problem and suggests a few ways the user can do about it.

5. Product is out-of-stock message

Showing “This item is out of stock” is the last thing we want to do on an eCommerce website, yet despite our great intentions, there are cases when the product is unavailable. The message should communicate whether the product is out of stock temporarily or permanently and suggest what the user can do about it. Ideally, the user should be able to leave their email, and the system will notify when the item is returned back to stock.

ChatGPT generated a very useful message and even reassured the users that their data is safe (i.e., email addresses that users will provide won’t be spammed).

6. Discount message

If you are looking for an easy way to sell a product, offer a discount code. Discount codes can work for various types of shoppers — price-droppers (potential customers who look for the moment when the product will be discounted) and impulsive shoppers (potential customers who can act because they have a clear motivation factor).

ChatGPT generated a solid copy with the right message, but it’s too long. The good thing about this tool is that you can ask a follow-up question to optimize the text.

7. Product return policy

Poor customer service is one of the most common complaints for any eCommerce service. And poorly designed product return is one of the things that can lower customer satisfaction scores to zero. To avoid any potential problems customers who want to return a product might face with the service, its vital to communicate the product return procedure.

ChatGPT generated a step-by-step guide for the customers who want to return a product and mentioned a specific time during which the customer or business is expected to do things.

8. Subscribe to newsletter

The larger the social network you build, the more profitable your business becomes. A newsletter is of the channels that you can use to communicate with your real or potential customers. But before people join the newsletter, they should clearly understand the value they receive.

9. FAQ section

A section with frequently asked questions (FAQ) is valuable for both customers (because they find answers to the question they have) and search engines (relevant information from this section can be good for SEO).

Overall, the list of questions that ChatGPT generated is helpful, but some of the questions are very uncommon. For example, “Do you offer financing options for your bikes?” sounds like a random question in this particular shopping experience.

Overall ChatGPT is very helpful for UX writing tasks, but it also has a few crucial downsides. It tends to write long paragraphs of text, even for requests that need one short sentence (like a discount message in our example 5). It also tends to respond monotonously, which can be a problem because people don’t like reading long passages of text, especially if they are written very monotonously. Last, ChatGPT can provide irrelevant responses for some questions (i.e., FAQ example 8).

ChatGPT cannot replace UX writers (at least right now). To make the most of the tool, you need relevant knowledge and skills in UX writing because a UX writer should moderate and fine-tune the output the tool provides.

ChatGPT changed the way we think about artificial intelligence. It's one of the tools that uncover the true potential of AI for all groups of users. But what made ChatGPT so powerful? ChatGPT is an agent simulated by the language model GPT-3.5. And in this article, I want to discuss a few fundamental things that made GPT-3.5 so effective.

Model scale & model alignment

GPT (Generative Pre-trained Transformer) is a large language model that uses deep learning to produce human-like text. The number of parameters defines the model scale. GPT-3 is called "large-scale" because it has 175 billion parameters.

But not only does the number of parameters matter, but also the alignment of the model itself. GPT-3.5, the brain of the ChatGPT, has the same model scale, but it's better aligned than GPT-3. And that's why it seems smarter. The OpenAI team fine-tuned the GPT model to make it more helpful in the context of chat assistant.

Let's discuss what exactly model alignment means:

Reasoning

The primary objective of language models is predicting texts. To predict texts, the model should be good at generating texts. One great thing about large language models like GPT is that they incorporate the mechanism of reasoning. Common sense reasoning is something we as human beings take for granted, but it's an entirely new thing for machines. But reasoning allows the GPT to become so good at simulating human behavior.

Reasoning is the use of natural language processing and machine learning techniques to analyze and understand text-based data. Think of reasoning as a logic that the model relies on when it answers the user's questions. Reasoning enables the model to "understand" the meaning of the text and use this understanding to provide the answer.

There are several different approaches to the reasoning for language models, including rule-based systems, knowledge graphs, and neural network-based models. GPT incorporates neural network-based models called transformers and gives the system the ability to learn complex relationships between words and handle large amounts of text corpora.

Reasoning also works for the prompt that the user submits to the system. The model generates the response in accordance with the prompt style. For example, if you submit a prompt that sounds like a quote from a scientific paper, the model will provide the output in a similar style.

But it's not enough to introduce reasoning into a model. It's also essential to fine-tune the model so it becomes better at following instructions. GPT uses semi-supervised learning—it combines unsupervised pre-training with supervised fine-tuning. Once OpenAI trains a model on a huge dataset in an unsupervised way, they fine-tuned the model to different tasks, using supervised training in smaller datasets. Codex, an AI system that translates natural language to code, is an excellent example of this approach.

Reinforcement learning

If you have experience working with GPT-3, you probably know this system was very sensitive to prompts. It was vital to articulate your intention properly so the system could understand what you wanted and generate relevant answers. GPT-3.5 is less sensitive to prompts because it was fine-tuned by people who evaluated the system's outputs. Think of fine-turning in this way—you show a few answers that the system generates for a particular question to humans and let them choose the best answer. It looks like a ranking system.

But you might wonder, “Okay, it's easy to do that for small data sets, but what about large datasets? How many human hours will it take?” The point is that reinforcement learning works a bit differently than a simple ranking system.

Reinforcement learning is a type of machine learning where the model learns to make decisions based on feedback it receives in the form of rewards. The goal of the model is to maximize its total reward over time. Reinforcement learning is a technique that summarizes human feedback and learns how to use it. The model that uses reinforcement learning tries to predict what answers the humans will react positively to. Since this model learns from humans in real-time, it becomes better and better at predicting. At the end of the training process, AI systems start to imitate humans.

Conclusion

The approach that OpenAI followed when creating GPT-3.5 clearly proves that "bigger is better" does not always work well regarding AI tools. GPT-3.5 has the same scale but is better aligned than its predecessor. Reasoning and Reinforcement learning allow OpenAI to align the GPT model in a way that can better understand user intention and generate more human responses.

Stable Diffusion is an AI model that can generate images from text prompts. The tool is similar to MidJourney or DALL-E 2. You can access the Stable Diffusion model online or deploy it on your local machine. In this article, we will review both approaches as well as share some practical tools.

Online services

DreamStudio

DreamStudio is an online tool created by Stability AI, the team responsible for Stable Diffusion. It provides access to the latest version of Stable Diffusion models (for example, below, you can see that I generated the image using Stable Diffusion ver 2.1–768). Plus, the tool has an impressive generation speed. It can generate an image in less than 15 seconds with default settings.

The tool offers a free trial to its users. As a user, you will have 100 credits that you can use to generate images, so you will likely generate from 100 to 500 images depending on your settings. You can buy 1000 credits for $10.

If you want to learn more about individual settings you can customise, such as the number of steps, cfg scale, etc., consider watching the following video:

Hugging face’s Stable Diffusion Demo

Hugging face provides free access to the latest version of the Stable Diffusion model (ver 2.1). But unlike the Dream Studio, the tool doesn’t allow you to customize properties like the number of steps, scale, etc. And it takes significantly more time for the tool to generate the image (on average, it takes more than a minute to create an image).

Local setup

Stable Diffusion is an open-source model. It means that you can run it on your local machine.

Stable Diffusion WebUI

Stable Diffusion WebUI is a web app that gives you access to Stable Diffusion models. It’s a browser interface based on the Gradio library for Stable Diffusion that you set up from individual libraries. There are two great things about this option:

• WebUI works on all systems (Linux, Windows, mac)
• WebUI gives you access to advanced settings such as choosing a sampling method or restoring faces (this option is extremely valuable if you generate human faces using Stable Diffusion).

But it also has a few significant downsides — you will need to install tools like Python, git and use the command line for some operations. So it will require some technical expertise. Plus, the interface of the app is not very user-friendly. You need to have at least a basic understanding of the settings that the Stable Diffusion model has to offer.

DiffusionBee

DiffusionBee is another software that you can run locally on your computer. Right now, the software is available for macOS users, but in the future, it will also be available for Windows users. The great thing about this app is that it is relatively easy to install and start using. The tool has a simple interface and guides users who want to access advanced features.

The downside of this app is that it uses the old Stable Diffusion model (ver 1.5), while all other tools in our list use 2.1. Plus, the app consumes a lot of hardware resources. You need at least 16 GB of RAM to run this tool comfortably; otherwise, image generation time will take more than a minute.

If you want to see how to install and start using this tool, consider watching this video:

What option should I choose?

Your decision should be based on how you plan to use the tool and what level of technical expertise you have. I recommend starting with Dream Studio because you will have access to the latest version of the model and the tool will have solid image generation performance. If you aren’t afraid of the command line, you should try Stable Diffusion WebUI.