AI Insights reads every tester's feedback in your test, written survey answers and video transcripts alike, and turns it into a report of feedback categories and the positive and negative insights inside them, each backed by the testers' own words.
AI Insights is a plan feature, so it may not be enabled on your account.
When does it run?
It runs on its own when a test finishes, and you can start an analysis yourself at any point, with or without your own categories and instructions.
Use it at the end of a test to get from two hundred open text answers to the five things that matter, midway through a multi-week test to check whether the early complaints are still the main ones, before a readout when you need quotes that prove a point, or when you have a specific question, such as how testers reacted to onboarding, and want the report organized around it.
What makes it more than a summary is the evidence. Every insight shows how many testers raised it and links to each tester's exact quote, in the survey or at the second in the video where it was said, so you can check any claim before you act on it. Nothing is reported on the word of one tester, and nothing is invented to fill a gap.
All plans include basic AI features, and Subscription and Enterprise plans add advanced AI features and customizations.
What the analysis reads
Each run reads every tester's written survey answers and the transcripts of any videos they submitted, treating a transcript like a long written answer. Issue reports are not part of the analysis. The report is built in four stages, which you see as progress while it runs: Reading feedback, Finding themes, Extracting insights, and Writing summary.
Two rules shape what ends up in the report. An insight is only reported when it is supported by at least two testers, so a single strong opinion never becomes a finding. And each tester contributes at most one quote per insight, so one talkative tester cannot make an issue look more widespread than it is. A report holds up to ten feedback categories.
When an analysis runs
Automatically when the test completes, provided at least ten testers took part.
Weekly during a multi-day test with at least ten testers, as long as new feedback has arrived since the last run.
Whenever you start one. Open your test and click AI Insights in the left menu. On a test with no analysis yet, click Generate AI Insights. Once a test has a run, click + New Custom AI Analysis to start another, with or without custom settings.
Each test allows five manually started analyses in any seven day period, and only one can run at a time. Processing time depends on how many testers are in the test; the page can be left and revisited, and the report appears when the run finishes. If a run fails, the page says so with Retry Generation, which keeps any custom settings.
Reading the report
The AI Insights page has three sections, from the shortest read to the most detailed.
By Top Insight
Up to three positive and three negative insights with the most tester support, each with a short title, the number of testers behind it, and the full finding. View all positive insights and View all negative insights jump to the full lists.
By Feedback Category
Every category the analysis found, ranked and sorted into Positive, Mixed, and Negative columns with the share of positive or negative evidence in each. A category counts as positive or negative only when one side accounts for more than 60% of its evidence; otherwise it is mixed, which is often where the most interesting findings sit.
All Insight Categories
The full report. Each category card shows its overall sentiment and how many insights it holds, followed by the insight cards inside it.
Use the search box to find insights and quotes by keyword, and the Positive and Negative chips to show one sentiment at a time.
Click View 6 Testers on an insight to expand its evidence: one quote per tester, with the tester, the date, and the survey it came from.
Click View in context on a quote to open it where it was given. For a written answer that is the survey results page; for a video quote, the play badge shows the timestamp and opens a preview of that moment without leaving the report. See AI insights for tester videos.
When one of your own requested categories has too little matching feedback, it still appears with a note explaining what was found, rather than being padded out.
Custom analysis
Click + New Custom AI Analysis to open Start Custom AI Analysis and shape the next run.
Under Surveys to analyze, untick any survey you want left out entirely, for example a screening survey or a logistics form. Unticked surveys are not read at all.
Under Feedback categories or themes to analyze, type a category and press Enter for each theme you want reported, up to ten. Your categories always appear in the report, in the order you gave them, and the analysis fills any remaining slots with themes it finds on its own.
Under Custom analysis instructions, describe what to emphasize or leave out, for example focusing on onboarding and ignoring pricing feedback.
Under Apply these settings to, choose This run only or All future runs. Saved settings apply to every later run, including the automatic ones, until you change them; Remove saved settings clears them.
Click Start Test-Wide Analysis.
Runs with custom settings are marked (custom) in the run list, and a bar above the report reads This analysis used custom settings with a summary of what was excluded or requested, plus Run again without settings. For the details and limits, see Guiding the AI Insights analysis with your own categories.
Working with runs
Every analysis is kept as a run, so you can compare what the feedback looked like at different points in the test.
Use the Run dropdown to switch between runs. Each is listed by date and tester count, with (latest) on the newest and (custom) on runs with custom settings; a run still in progress or failed is labeled as such.
Open Run options to Run again, which starts a new run with the same settings and counts toward the weekly limit, or Remove, which hides that run from the page without affecting the others.
The Export, Print, and Share buttons at the top always act on the run you are viewing. Share copies a link to the page for a colleague with access to the test.
Export downloads the whole run as an Excel workbook with categories, insights, evidence, and per-tester sheets.
AI insights on the test dashboard
Learn more about the AI Insights on the Test Dashboard.
Your test's home page summarizes the latest run so you do not have to open the full report to know where things stand.
BetaIQ is a product quality score. It is meant as an estimation for the overall product quality based on feedback provided during the test. The score takes into account the amount of positive / negative feedback along with sentiment weight, and any NPS or high-level rating questions (like star ratings) that are part of your surveys.
Beneath it are three tiles: Issues, a letter grade based on how many testers were blocked and how many high severity issues remain open; NPS from any 0 to 10 recommendation question, or overall Sentiment when the test has none; and Test health, a grade for how well testers are engaging and completing activities on time. The score is calculated when the test ends, and weekly during multi-week tests.
The AI Insights section below it lists the top Positive findings and Negative findings, the Top positive categories and Top negative categories, and Tester voices, a handful of representative quotes. Where an insight is supported by video, it notes how many of its testers said it on video. View full report opens the AI Insights page.
The Test Overview section alongside it shows testers recruited, issues submitted, messages sent, and the Avg Engagement Score, a per-tester measure of engagement quality based on the ratings their feedback receives, completion, timeliness, and issue reports.
Things to know
The score and the report are only as good as the questions. Asking for praise gets praise, asking for problems gets problems, and bug hunts naturally grade lower on issues. Treat the BetaIQ score as a directional guide within one test, not a comparison against other products.
A run can finish with nothing to report, for example when a test has too little written feedback or the two-tester rule leaves nothing standing. The page says so plainly rather than showing an empty report.
Custom instructions guide emphasis and wording. They cannot change the report's structure, and text that tries to redirect the analysis itself is ignored.
Exporting needs full admin access to the company. Support and read only admins can view, print, and share the report.
Video quotes come from transcripts, so videos without speech contribute nothing. Individual videos also carry their own insights and exports; see Exporting the AI insights for one video and Downloading a video transcript.
The per-question AI tools on the survey results page, such as sentiment, key phrases, word clouds, and summaries of a single text question, are separate from the test-wide analysis. See Reviewing survey feedback and Analyzing survey result charts.