User research is one of the most important things we can do to design a truly valuable, successful product. That should be one of the most streamlined parts of our process.
ChatGPT (or any other generative AI) can greatly help with that.
Let me share some of my favourite prompts for speeding up user research.
One of the most important tasks in user research is recruiting the right candidates, so let’s make sure we nail it:
If you have an existing user persona in place, you can quickly convert it into key criteria for a screener:
Prompt: Convert this persona into inclusion/exclusion criteria I can use in a screener.
Drafting screeners tends to be time-consuming, but that’s actually one of the areas AI is really efficient at.
If you used the prompt I shared earlier, ask it to turn those criteria into actual questions (sometimes the model does it itself with the first prompt, sometimes it doesn’t):
Prompt: Use these inclusion/exclusion criteria to create a short yet accurate screener for user research candidates.
Alternatively, here’s a more loose prompt, just be very descriptive of whom you want to recruit:
Prompt: Draft a screener to recruit [target]. For each question, explain what it screens for and how it maps to the research question. Include one ‘trap’ item to catch satisficers.
Keep in mind that the “explain what it screens for and how it maps to the research question” is for you to double-check the screener; don’t include it in the version you send out.
Check screeners for bias. If the draft was already made by an AI, use a different model to double-check:
Prompt: Review this screener for priming/leading language and propose an order that minimizes bias. Mark any items that should be randomized.
A surprisingly strong side of AI models? Rewriting legal jargon into comprehensible language. A small but valuable detail:
Prompt: Rewrite this consent section at a Grade-6 reading level. Keep the legal meaning intact. Provide two shorter alternatives for mobile.
Just double-check the new version with your legal team.
I know sending AI messages is still taboo sometimes, but AI does speed up the whole process:
Prompt: Write 4 invite variants (email, in-app nudge, forum post, LinkedIn DM) to recruit [target]. Keep it under 80 words, avoid ‘survey/feedback’ language, and emphasize the value and time commitment.
I ask it to avoid the word survey/feedback because they sound like work, and people don’t like to work!
Nowadays, I never start my guides/scripts from a blank page. Instead, I let AI have the first shot here. Yes, there’s a lot of fixing and adjusting, but I still find it more efficient than starting everything from scratch:
I always let AI take the first shot at creating an interview plan and list of questions, and then just edit it slightly:
Prompt: Create a 45-minute semi-structured interview guide with [target persona] to answer [research question]. Include warm-ups, neutral probes, and story-based questions.
Remember, always ask users for stories; that’s where most insights are.
Whether the questions were created by you or by AI, there’s no harm in double-checking for irrelevant questions:
Prompt: Given my guide, list questions that are leading, solution-oriented, or out of scope. Provide neutral rewrites.
Usability testing scripts are one of the areas that AI seems to perform extraordinarily well, unless you test some novel or very nuanced flow:
Prompt: Draft tasks for a remote unmoderated test for [flow]. Define success metrics, acceptable hints, and post-task questions that diagnose friction without blame.
If you already have a prototype/design, feed it to the model as well.
ChatGPT can be a powerful co-pilot during the session. You don’t need a fully agentic, expensive tool; copy-pasting works too, especially if you have a dedicated note-taker:
Whenever a topic is intriguing or you get stuck, paste a part of the live transcript and ask the model to help you with follow-ups:
Prompt: As I paste notes, suggest two neutral follow-ups I could ask next to deepen understanding of goals and constraints.
Tracking the sentiment of the interviewee in real-time can help you notice hidden verbal cues faster and steer the interview accordingly:
Prompt: Summarize the emotional tone of this answer (e.g., confused/skeptical/delighted/resigned) and why you think so. Keep it one line.
I use that one mostly for longer and more confusing answers.
Near the end of the session, copy-paste the whole transcript and ask what further questions might be worth asking before closing the interview:
Prompt: Based on the interview so far and the research question for this interview, what follow-up questions should I ask before ending the interview”
This really helps avoid that nasty feeling of “Damn! I wish I had asked her X!”
Interviews are great and should be the main source of your research insights. Still, surveys can be a great add-on to your research:
I often get inspired/surprised by some of the questions the AI comes up with:
Prompt: Generate X non-leading survey questions to answer [research question].
Include a mix of frequency, agreement, and scenario-based formats. Add one or two open-ended questions.
If you already have your survey questions, you can ask ChatGPT if it would add anything else.
Whether created by you or by the AI, double-checking for bias is always a good practice:
Prompt: Review this survey for bias, double-barreled questions, and ambiguous terms.
Rewrite problematic items and explain the reasoning for each fix.
There are usually five things I’m interested in when analyzing survey results:
Prompt: Provide a quick summary of
most common answers
most polarized answers
surprising outliers
main differences by segment*
main insights from open-ended questions
*if self-segmentation is part of the survey/screener.
Surveys often reveal interesting correlations that are not that easy to spot at first:
Prompt: Identify potential correlations or relationships between survey answers. For each, explain what might be the reason and if the link is strong or speculative.
Remember not to confuse correlation with causation.
It doesn’t matter how great your interviews were if you don’t draw specific insights and next steps from them. Let’s use AI to make sure we haven’t wasted time:
If you are not transcribing your interviews, start now. It makes further analysis much, much easier, especially with this prompt:
Prompt: Turn this transcript into
- executive summary
- ~5 main findings
- open questions and follow-ups
That’s probably the biggest time-saver on the whole list 🙂
Always check summaries generated by AI for hallucination:
Prompt: Check the summary below and compare it with the transcript. To each statement, connect a relevant quote. If no quote can be found, remove it from the summary.
Excessive? Not really. AI can hallucinate even when summarizing things.
Sometimes it takes a whole day to summarize everything we learned into a short and meaty one-pager. Or five minutes with a simple prompt:
Prompt: Generate a CEO-ready one-pager from these outputs: problem, what we learned, user quotes, business impact, next steps.
Even if you want to go through the process of summarization yourself (which has its own benefits), an AI one-pager can give you another view on the topic.
One could go on and create a list of 200 prompts here, but there’s no point. Prompting is not coding. You don’t need very specific prompts, just some common sense. Give AI the same instructions you would give to an employee.
The list here serves more as an inspiration. Choose prompts you need, edit them, and create new ones. Just go and start experimenting. Each hour invested in mastering AI prompting will yield dozens (if not hundreds) of saved hours in the long run.
LogRocket's Galileo AI watches sessions and understands user feedback for you, automating the most time-intensive parts of your job and giving you more time to focus on great design.
See how design choices, interactions, and issues affect your users — get a demo of LogRocket today.
A practical guide to AI in UX design, covering predictive UX, generative assistance, personalization, automation, and the risks of overusing AI.
AI wireframe tools are everywhere but they don’t all work the same way. I tested Visily, UX Pilot, Uizard, Mokkup AI, and Figma Make to see which tools are best for non-designers, fast iteration, and serious UX work.
Small actions can have large consequences in complex systems. Here’s how UX designers can manage dependencies so users feel informed and in control rather than blocked or blindsided.
This article examines when hero sections are necessary in digital products, when they create friction, and how to evaluate them using UX goals, primary actions, user flow impact, and real-world alternatives.
Learn more →
