Research · AI answer consistency · October 2026
Ask AI the same question tomorrow. You’ll get a different shortlist.
We analysed 197,314 AI answers to more than 1,650 real buying questions, tracked daily across five leading AI engines (ChatGPT, Google AI Overviews, Gemini, Perplexity and Copilot) over eight weeks in the UK and US. By comparing each answer with the next answer to the same question, we measured how consistently AI names the same brands. Only one answer in four named the same brands as the day before.
of next-day answers named exactly the same brands. Same question, same engine, same country.
is the chance that the brand AI names first will be a different brand the next day.
of the brands AI mentions for a question appear almost every time it is asked: in 9 answers out of 10 or more.
appear rarely, in 1 answer in 10 or fewer. Spot one in a single check and it looks just like a regular.
01 · One question, one week
Seven days, six different shortlists
Let’s zoom in on a real question from our tracking, put to ChatGPT once a day for a week. One brand held the top spot every day. Everything behind it kept changing.
"What’s the best [product] on the market?"
ChatGPT · UK · 13 to 19 Oct 2026
Real answers from reconnAI tracking. Brands shown in the order each answer named them. Brand names and the product category are withheld.
Brand A was named first on all seven days. That is what a strong position in AI looks like: not one good answer, but the same answer every time.
Four other brands took turns behind it. Check on Monday and Brand B is on the list. Check on Tuesday and it has gone. Neither day tells you the real picture.
Snapshot vs aggregated view
Measure for longer and your real visibility comes into focus
The same week of answers, read two ways. A single check shows you one answer. Combining every day shows you each brand’s share of answers, meaning how often it was named across the week. That is a figure you can track, compare and act on.
Looks like a two-brand race, with Brand B a clear second. Brands C, D and E don’t appear at all.
Brand D, missing from Monday’s answer, is the real number two. Brand B, Monday’s runner-up, showed up once all week.
An aggregated view beats a snapshot. The more answers you combine, the steadier the figure gets. Measured over weeks, your share of answers shows how visible you really are, and whether that is changing.
02 · Over time
The longer you wait, the more it moves
Some of the change is random: ask the same question twice and you can get a different answer. Some of it builds up, as answers move further from where they started. Four weeks on, only about half the brands named today are still being named, and the exact same shortlist comes back just one time in six.
The same question, on the same engine, in the same country, compared one, seven and 28 days apart. Based on the five categories tracked daily for all eight weeks, so the figures differ slightly from the headline numbers above.
03 · Why one check misleads
Most brands in an AI answer are only sometimes there
We took every brand AI named for a question at least once in eight weeks, and counted how often it came back. Out of every 100:
8 are named nearly every time, in 9 answers out of 10 or more. These are the brands AI names consistently.
47 come and go, named in more than 1 answer in 10 but fewer than 9 in 10.
45 rarely appear, in 1 answer in 10 or fewer.
A single check can’t tell these groups apart. A brand that appears once a month looks exactly like one that appears every day.
Based on the five categories tracked daily for all eight weeks, across every engine, with each category weighted equally. Rounded to whole numbers.
04 · By engine
No AI engine gives a steady answer
The engines differ less than you might expect. None repeats the exact same shortlist the next day more than about 3 times in 10.
Same shortlist: the next day’s answer named exactly the same brands.
Same brand first: the brand named first did not change.
Brands named again: how many of today’s brands were also named the next day.
| Engine | Same shortlist | Same brand first | Brands named again | Categories in sample |
|---|---|---|---|---|
| Google AI Overviews | 27% | 68% | 66% | 9 |
| ChatGPT | 24% | 64% | 64% | 9 |
| Gemini | 23% | 62% | 64% | 5 |
| Perplexity* | 30% | 71% | 63% | 2 |
| Copilot* | 21% | 76% | 72% | 2 |
05 · What it means
Stop checking. Start measuring.
A screenshot isn’t evidence
One answer is one roll of the dice. Being named today says little about tomorrow, so don’t base a decision on a single check, good or bad.
Visibility is a percentage, not a position
The number that matters is the share of answers that name you, measured every day, for each AI engine and each question your buyers ask.
Consistency is the prize
Only a few brands are named nearly every time. Moving from "sometimes" to "nearly always" is a better goal than getting into a single answer.
Read trends over weeks
A lot of day-to-day movement is random. Look at weekly figures over time before you react to a dip or celebrate a jump.
How reconnAI measures this
Built for answers that never sit still
This study is built on the same daily data reconnAI collects for every customer. When the answer changes every day, measuring it every day is how you get a reliable picture.
Only 1 in 4 answers repeats the same shortlist
Every question, every day. Your buyers’ questions run daily on the AI engines that matter to you, from ChatGPT and Google AI Overviews to Gemini, Perplexity and Copilot. Visibility is reported as your share of answers, not a single snapshot.
Most brands are only sometimes named
Question by question. See where you are named nearly every time, and which questions sit in the "sometimes" zone where work pays off most.
The first brand changes 1 time in 3
Against your competitors. Track how often you are named first, how often a rival is, and which websites AI cites when it names the leaders.
Day-to-day swings hide real change
Trends you can act on. Weekly trends show whether a change is real or just the usual daily swing, so you can show what a campaign actually achieved.
Questions
Common questions
Does ChatGPT give the same answer every time?
No. When we asked ChatGPT the same question on two consecutive days, it named exactly the same brands 24% of the time. The brand it named first was the same 64% of the time.
Why do AI answers change from day to day?
Each answer is written from scratch, and many engines search the web before they reply, so the sources and wording vary each time. Over weeks, updates to the AI models and new content published online can shift answers further. Our data shows both: a big swing from one day to the next, and answers moving further from where they started over time.
Is any AI engine more consistent than the others?
Not by much. Google AI Overviews, ChatGPT and Gemini all repeat the exact same shortlist the next day around a quarter of the time. None of the five engines we track gives a steady answer.
How should I check whether AI recommends my brand?
Ask the same questions repeatedly, on every engine your buyers use, and track the share of answers that name you over several weeks. A single search, or a screenshot from a colleague, can’t tell you whether you are a regular or a one-off.
Method
How we measured it
reconnAI puts the same real buying and research questions to AI engines every day. For this study we used about 1,650 questions from ten categories, asked in the UK and US on the free version of each engine, between 31 August and 25 October 2026. Questions ran daily, apart from one category tracked weekly.
We compared each answer with the answer to the same question, on the same engine, in the same country, one, seven and 28 days later. We left out questions that name the company being tracked. We also never counted a brand that was named in the question itself, because the answer is bound to mention it.
In each category we looked for the competing brands that matter in that market, typically 10 to 30. Brands outside those lists are not counted. Every figure is worked out for each category first and then averaged, so each category counts equally and no single one dominates. For the headline figure, we are 95% confident the true value is between 21% and 29%. All data is anonymised.
How often does AI actually name you?
We run your buyers’ questions every day across the major AI engines and show you your share of answers over time, not a single snapshot.