On 28 September 2026 we asked Perplexity the same local-business question twice, a median of 3 hours apart, for 40 questions in the form "best <profession> in <town>, UK". The two answers to the same question were rarely the same list.
The list moves between runs
Only 3 of the 40 pairs came back with an identical list of named firms both times. 1 of the 40 pairs shared no firm at all between the two answers. On average, the two answers to the same question shared 48% of the firms they named between them.
Looked at from the firms' side instead of the pairs' side: 179 of the 324 named firms that appeared across those 40 pairs showed up in only one of the two answers, not both. 324 is the combined total of firms named across the two answers in each of the 40 pairs, counted pair by pair.
We did not change the question between runs. Same wording, same town, same profession, asked a median of 3 hours apart on the same day.
What this looks like in practice
Say a business owner checks Perplexity in the morning, finds their firm named alongside four others, and checks again that evening out of curiosity. In this sample, only 1 of 40 second answers shared no firm at all with the first, and only 3 of 40 matched it exactly; the rest landed somewhere in between, with an average overlap of 48%. If the owner's firm dropped out of the second answer, that alone doesn't mean anything changed about the business, the listing, or the competition.
A firm named in one answer and missing from the same question a few hours later hasn't necessarily done anything to earn or lose that place. The same goes for a competitor named once, in one answer.
What we don't know
We measured that the list changes; we did not measure why. It could be that Perplexity draws on a different subset of the same sources each time, that a source itself updated between runs, or something else in how the answer gets built. We report what 40 repeated questions returned on one day; the mechanism is outside this data.
This is Perplexity only, in basic mode, from one day, 28 September 2026: only the inline citation links in each answer were counted, not the full source panel. We have not run the same repeated-question test against Google's AI Overviews, so this article makes no claim about how stable its answers are. The pairs also come from a small group of professions, so read them as a close look at one slice of the wider run, not a picture of all of it.
Why this matters for measuring
If a business owner checks Perplexity once, sees their name, and concludes they're "in", or checks once, doesn't see it, and concludes they're "out", either conclusion can be wrong within a few hours. A single answer is a sample from a distribution, accurate at the moment it was taken. It does not show what the same question would have returned an hour earlier or later, and this data says that difference is often not small. That's the reason we run each query 3 times before reporting whether a business appears, rather than trusting one screenshot (see how we measure).
This sits inside a larger run: 496 answers to "best <profession> in <town>, UK" across 105 professions on 27–28 September 2026, basic mode, only inline citation links counted. The 40 repeated pairs are a subset of that run.
What to do with this if you track your own visibility
Treat a single check as a data point. If you see your business named, check again later the same day before you credit whatever you just changed. If you don't see it, check again before you conclude the work isn't paying off. Where possible, compare several runs over several hours or days instead of one.
The same caution applies to anyone selling AI-visibility monitoring, including us. A tool that reports a single daily or weekly check is reporting one reading of something that moved this much within a few hours. It's worth asking any provider how many times a question is run before a result is reported, and whether "gone" or "gained" is based on more than one answer.
Method
- 40 questions, each asked twice (40 pairs of answers) on 28 September 2026, median 3 hours apart, format "best <profession> in <town>, UK".
- Part of a wider run of 496 answers, 456 distinct questions, 105 professions, 27–28 September 2026, Perplexity basic mode, only inline citation links counted; the full source panel was not expanded.
- No business named in this article; we report counts and overlap only.
If you want to know how your business appears across repeated checks, request a free audit.