Using ChatGPT for UX Writing
Try writing copy with me, my colleague, and ChatGPT!
Hi! It’s Luda and Nastya. We work together as UX writers and write clear and empathetic texts for interfaces. Over the past couple of months, we have been experimenting with ChatGPT in various aspects of our work, and we have compiled the results in this article. It’s also available in Russian.
We have roughly divided our experiments into three categories:
- Content and interface text generation
- More creative tasks, such as writing a 404 error message and using tone of voice
- Assistance in copy editing
Let’s take a look at them one by one.
Content generation for UI elements
The most obvious application for AI is simply asking it to do a part of our work, which is to write UI copy like errors, emails, or forms. We approached the СhatGPT with several tasks that closely resembled our real ones. And here’s what we got:
- Copy for a notification informing the user about a new payment method
We wouldn’t use any of these options as the final version of the copy because not everything in them aligns with our style guide and general UX writing principles. However, we can definitely take note of some phrases. For example, “give it a try/go,” “say hello to..,” and “the latest addition to…” is useful, and might not have immediately come to mind.
If you need a more concise text, which is often the case when writing UI copy, you can ask the AI to shorten these options.
- Title and button for a modal window about insufficient funds
Here the AI worked on a modal window offering the user to top up their balance to access the content. ChatGPT did well with the button text and also managed to shorten the title as requested. The only issue for us was the capitalization of every word, which contradicted our style guide. These details can be specified in the prompt, just like the character limit, but sometimes it’s faster to skip this step altogether and adapt the AI’s output to the company’s guidelines on your own.
- Subject of the email about a new feature
The context is quite simple: there’s a new functionality for our partner, and we need to notify them about it via an email. Let’s imagine we have the email body ready to go, but we end up struggling with the subject line (maybe you can also relate to that).
We fed this text to ChatGPT and received 5 options for the subject line. Some of them were suitable and could be used as is, while others required some minor editing. Overall, they were good. It’s nice to see that the options have different syntactic structures and aren’t just variations of synonyms. Similar to the notification, we can at least pick up a couple of useful phrases.
- Texts for minor errors
This time, we decided to “train” the AI with our corporate rules and included a small list of guidelines in our request. Unfortunately, not everything worked as expected. AI persistently inserted “please” and “sorry” even when we explicitly asked it not to do so.
The error texts themselves turned out fine in the end, but interacting with the AI and making additional edits took as much time as it would have taken for a human writer to create this text from scratch.
- Message about user consent for data storage
The next task for ChatGPT was more challenging in terms of structure — we asked it to write a message with a title, body text, and a button.
What went well: the message turned out to be cohesive. The title aligned with the button and wasn’t repeated in the body. It clearly and concisely communicated what was required from the user, and even included a short encouraging phrase, “Your data is safe.”
What was puzzling: the chat showed a tendency to assume and add information that wasn’t requested. For example, AI assumed that having a transaction history would help users make more informed purchases in the future. It resulted in a user-friendly message highlighting the benefit to the reader. However, this may not always be relevant or necessary. ChatGPT developers warn users about such hallucinations of the AI, so it’s just something to keep in mind.
In some situations, this could be valuable and useful, but blindly copying the AI’s output is not recommended.
- Purchase addition form
The last task for ChatGT as a UX writer was to write texts for a form where users store their purchase history using a transaction ID and their email.
The chat did well with the task, except for initially writing long texts and using capitalization in the headings and buttons (again). Additionally, the AI primarily used a casual tone and added a few “engaging” exclamation sentences, which a human UX writer would have to adapt to their Voice & Tone. After some editing, we got the texts that could be used in the actual product.
Assistance with creative tasks and tone of voice
In addition to writing neutral copy, AI tools can be used as an idea generator: you can ask AI to come up with a couple of jokes, rewrite texts in different tones, or generate variations of product names.
- Jokes
We turned to ChatGPT for a couple of light-hearted and non-offensive jokes to make verification success a bit more joyful. We were pleasantly surprised that the jokes turned out to be original, not clichéd, and not overly sweet. Not all the variations were super successful, of course, but some key phrases with human editing could serve as a good base for a nice little joke.
- Different voices and tones of text
ChatGPT can also play with the voice of the product by writing the same copy for different audiences. For example, stylizing the onboarding flow in the pet owners’ app:
From the AI’s response, it’s clear that it didn’t interpret our task correctly in some places and even added details that weren’t asked for, like the VIP mention in the first variation. However, overall, the results are impressive — the lexical and grammatical choices align with what you’d expect from the selected tone.
Such a request can also be used to estimate roughly where your current text falls on the tone spectrum and what you can try to move closer to the desired end.
- Name for a new feature
If you and your team have been brainstorming for hours trying to come up with a name for a new feature, why not involve ChatGPT as an idea generator? It’s possible that it may propose something absurd or simply echo your ideas. But there’s always a chance that its output will provide you with an idea, phrase structure, or even a ready-to-use feature name. And if you like a certain suggestion, you can always let ChatGPT know and ask it to generate more ideas based on that one.
- Creative text for a 404 error
Finally, that only error where creativity isn’t only allowed, but also encouraged. Yes, you can also entrust the task of writing a creative text for a 404 error to ChatGPT. Just remember to specify what actions the user should take in case of an error, so that the generated texts are relevant.
Assistance in text editing, working with dictionaries, synonyms, and more
Alright. What if you’re not yet ready to trust AI with the writing and just want to use it when editing? We tried using ChatGPT as a thesaurus, spell checker, and contextual dictionary. Here’s what we got:
- Synonym search
Using AI as a dictionary is very convenient because you don’t have to switch to other tabs. An additional advantage is that you can find less obvious synonyms from different parts of speech if they slip your mind. You can also find synonyms for entire phrases, not just individual words or fixed expressions. For comparison, we used Thesaurus, which we occasionally used before.
Thesaurus didn’t find a phrase “coming soon”, but the synonyms for the individual words are almost the same.
- Proofreading and spellchecking
For text proofreading, we usually rely on plugins, but it was interesting to try out ChatGPT as well. We translated a paragraph from the draft of this article using Google Translate and intentionally added errors:
The AI did an excellent job of correcting typos and grammatical errors, such as verb tenses and articles! The only thing we noticed was that it omitted the phrase “A few of them can be taken” because we accidentally removed a comma, and it didn’t recognize it as a separate thought. This slightly altered the meaning of the statement, as the original text conveyed the message that “some things can be used as is.” As always, it’s important to review the edited text to make sure it’s still true to your original copy.
If you write in your second language like we do, you often question the compatibility of words. Typically, to check a preposition that goes with a word, we pull out a contextual dictionary. We tired ChatGPT to see how it would handle this:
We asked it to provide examples of using the word “deceive” in a sentence. We saw that the preposition “into” is often repeated, and we clarified its meaning.
The simplicity of it all is the most appealing aspect — no need to search through other tabs and services or think about how to find the right word or phrase. However, there’s a complication: we can’t be 100% sure that the answer will be correct. We think ChatGPT can be used as a dictionary if you have a good command of the language and can notice when it produces nonsense. For completely unfamiliar words, we’d still turn to the good old dictionary. But as a spell checker, the AI works fine — you just need to read through the edited text.
- Generating copy alternatives
We took it a step further and decided to ask the chat to suggest alternatives to our text. This can be useful when you’re stuck and everything seems off, but you don’t have time to generate more ideas.
It’s really cool that the suggestions turned out to be diverse in structure and tone. There’s a lot to choose from. But let’s not forget to verify the generated text for accuracy. The AI added “See how easy it is to use!” on its own, which, overall, doesn’t contradict the essence, but it demonstrates how ChatGPT sometimes adds its own message and raises user expectations. We definitely don’t want that.
- Summarizing text
Lastly, we asked the chat to summarize the content of the text. This feature will be interesting not only for writers, but also for anyone analyzing large volumes of text.
A part of an article we asked ChatGPT to summarize:
The chat did a great job with the task and saved us some time. This time it was accurate and didn’t add anything of its own.
Conclusion
ChatGPT performed excellently in editing and proofreading the text. With minimal editing, it showed good results for simple interface copy like errors, modal windows, and forms. Slightly more editing was required for creative copy like 404 errors and texts with different tones, but AI worked great for generating jokes and finding suitable phrases. Nothing was too irrelevant because we made sure to provide detailed context, but it’s worth noting that ChatGPT tends to add features and advantages to the product, so it’s important to verify the answers not only for compliance with the style guide but also for plausibility.
ChatGPT can be a great assistant for UX writers, speeding up routine tasks like writing error copy, or generating creative ideas. Unlike a human writer, if not given a limit, the AI can generate a very long text and may not always adhere to style guide requirements. However, in some aspects it’s similar to a human writer: both require a brief and context to deliver the desired outcome.
Understanding this, your non-writing teammates may try to bypass you as the UX writer and turn to ChatGPT for copy. We’ve also been there, so we recommend explaining to the team why unedited text may not be suitable for their product. The examples from this article may help you present convincing arguments. This topic is also actively discussed in the Microcopy & UX Writing Facebook group, where you can find additional insights.
The desire to speed up the process is understandable, especially when there are few writers on the team. To avoid being seen as the grammar police, you can create templates for the team to use when seeking assistance from ChatGPT (providing maximum context) and a concise version of the style guide, allowing them to edit the text themselves. You can also establish guidelines for when texts can be written by the AI without review and when it should require a thorough check. The question is whether teams are willing to make that effort. For now, we keep advocating for reviewing all texts and still giving the final say to the UX writer.
ChatGPT is actively evolving, along with other AI tools. While we were gathering materials for this article, ChatGPT gained more resources, courses, and materials about its usage and advanced settings — like this article. We’ll continue to explore AI tools in our work and daily lives, and if we come across anything significant, we’ll be sure to share it.
Anastasiya Myasnikova thank you for collaboration on this article!
