When you start a test-wide AI analysis you can tell it which feedback categories matter to you and add instructions about how to read the results. The finished report then covers the categories you asked for, in the order you gave them, and follows your instructions when deciding what to emphasize.
This is worth doing whenever you already know what the test is meant to answer. If you are testing an onboarding rewrite, asking for "Onboarding", "Sign-up friction", and "First impressions" guarantees those appear as their own sections rather than being merged into something broader. If pricing feedback is out of scope this round, an instruction to set it aside keeps the report focused on what you are deciding. The analysis still surfaces themes you did not ask for, so nothing unexpected gets lost.
All plans include basic AI features, and Subscription and Enterprise plans add advanced AI features and customizations. Customizing an analysis is one of those advanced capabilities, so it may not be available on your account.
How to start a custom analysis
Open your test and go to the AI Insights page.
Click + New Custom AI Analysis above the results.
In Feedback categories or themes to analyze, type a category and press Enter. Repeat for each one, up to ten.
In Custom analysis instructions, describe how you want the feedback read. For example, focus on the onboarding experience and leave pricing feedback out.
Click Start Test-Wide Analysis.
You can fill in either field on its own. Leaving both empty starts a standard analysis.
What your categories do
Every category you request appears in the finished report, in the order you entered them, and tester feedback that fits one of them is grouped underneath it.
The analysis still derives its own categories from the remaining feedback, so themes you did not anticipate are not lost. Those appear after yours.
A report holds at most ten categories in total. Your requested categories are kept first, and the remaining slots are filled with categories found in the feedback.
Things to know
Each category name is limited to 200 characters. They work best as short labels of one to three words, such as Battery life or Onboarding.
Instructions are limited to 2000 characters.
A category you asked for that has too little matching feedback still appears in the report, with a note explaining what was actually found. Nothing is invented to fill it.
An insight has to be raised by more than one tester before it is reported. This applies to your categories too, which is why a requested category can appear with no insights under it even though a single tester mentioned it.
Instructions guide emphasis and wording. They cannot change the report's structure, and anything that tries to redirect the analysis itself is ignored.
Runs started with custom settings are marked (custom) in the run dropdown. Select one and hover its Custom tag to see the categories and instructions it used, and any export you take matches the run you have selected.
If a custom analysis fails, Retry Generation starts it again with the same categories and instructions.
Each test is limited to three manually started analyses in any seven day period, and a new one cannot be started while another is still running.
+ New Custom AI Analysis appears once the test has at least one analysis run behind it.