Yesterday we showed that AI product recommendations mostly come from "best X" lists. The most common question we got was simple: do the answers stay the same?
So we asked again. Same 210 questions, same four engines, same settings, one day later. Then we compared each answer with the one from the day before.
Most of them changed.
So a single check of "does ChatGPT recommend us?" doesn't tell you much. Ask tomorrow and you may get the opposite answer.
Here's a real example. We asked Google AI Mode for the "easiest CRM for a team that has never used one".
17 November
What it read
18 November
What it read
On day one, Less Annoying CRM was the first pick, called out for its "zero-frills simplicity". AI Mode had read one page to answer: Zapier's list of the best small business CRMs.
On day two, Less Annoying CRM wasn't mentioned at all. HubSpot was first. This time AI Mode had read two different pages: a guide on Slack's blog and HubSpot's own product page.
Nothing about the CRMs changed overnight. What changed was the pages the engine happened to read. Swap the pages and you swap the answer.
That's the pattern behind the numbers. An AI answer isn't a fixed ranking. It's a summary of a few pages, picked fresh each time.
Not every brand moves the same amount. The higher a brand ranked in an answer, the more likely it was to be named again the next day.
#190%
#2–379%
#4–566%
#6 or lower61%
Same question, same engine, 17 vs 18 November. Based on 5,072 brand mentions.
Brands that show up in a lot of answers in their category kept 73% of their spots. Brands that show up less kept 51%. The difference makes sense: a leader is on most of the pages AI reads, so whichever pages it picks, the leader is on them. A smaller brand is on a few. When the engine picks other pages, the smaller brand drops out.
If you're not a leader, this is the part that matters. Your visibility depends on which pages the engine reads that day. The way to make it stable is to be on more of those pages.
| ChatGPT | Gemini | AI Overviews | AI Mode | |
|---|---|---|---|---|
| Answers that changed | 88% | 95% | 67% | 84% |
| Brands gone the next day | 23% | 33% | 24% | 28% |
| Same #1 brand both days | 71% | 57% | 66% | 58% |
| Same cited pages | 40% | 41% | 47% | 39% |
Gemini changed the most. On 95% of questions, its list of brands was different the next day, and it dropped a third of the brands it had named. Google AI Overviews changed the least, but even there two out of three answers moved.
ChatGPT was the most loyal to its first pick: the same #1 brand 71% of the time. That fits what we saw yesterday. ChatGPT often starts with a shortlist from memory before it reads anything, so its top picks depend less on the pages it finds that day.
Only 42% of the pages cited on day one were cited again on day two. Brands were more stable than that (61% overlap), mostly because the same leaders appear on many different pages.
This connects the two studies. Yesterday we found that 90% of recommended brands were on a list the engine cited. Today we see that the set of cited lists rotates every day. Put together:
| Most stable | Same brands | Same #1 |
|---|---|---|
| Accounting software | 79% | 93% |
| VPNs | 75% | 75% |
| Travel insurance | 69% | 53% |
| Least stable | Same brands | Same #1 |
|---|---|---|
| AI search visibility tools | 48% | 52% |
| Mattresses | 51% | 56% |
| Running shoes | 53% | 33% |
Accounting software barely moved. QuickBooks, Xero, Wave and FreshBooks come up in most answers on every engine, so there's little room for change.
The least stable category was our own: AI search visibility tools. That makes sense. It's a new category with many young companies and no settled set of lists. Running shoes had the most unstable #1: the top pick changed on two out of three questions.
Don't trust a single check. A screenshot of ChatGPT recommending you, or ignoring you, is one sample. 84% of answers changed in a day. You need to ask the same questions many times and look at how often you come up.
Measure share, not yes or no. A single answer named about 6 brands. Across eight answers to the same question (four engines, two days), 14 different brands came up. 30% of them were named only once. Only about 2 brands per question were named in all eight. The useful number is how often you're named across many answers, and whether that number is going up.
If you're not the leader, get on more pages. Smaller brands dropped out half the time. The fix is the same one from yesterday's study: find the lists AI reads for your buyers' questions and get onto the ones that leave you out. The more of them name you, the less it matters which ones the engine reads.
Check the sources, not just the answer. The pages rotate even faster than the brands. A list that was cited last week may not be cited today, and a new one may have appeared. Watch which pages are being read, because that's where the next answer comes from.
Give it time before you judge a change. If you get onto a new list and your brand shows up the next day, that's a good sign but not proof. Watch the trend over a week or two.
Doing this by hand means asking the same questions on four engines every day and writing down every brand and every page. Nobody keeps that up for long. That's why we built Supercite.
The engines search the web each time they answer and read a handful of pages. Which pages they pick varies from run to run. Different pages name different brands, so the answer changes.
Often enough to see a trend: at least weekly, ideally daily. Any single check can be off, because 84% of answers changed in a day.
More than the rest. A brand ranked first was named again the next day 90% of the time. But it was ranked first again only 63% of the time.
Google AI Overviews changed the least (67% of answers changed). Gemini changed the most (95%). ChatGPT kept the same top pick most often (71%).
Be on more of the pages AI reads. Find the "best X" lists cited for your buyers' questions and get onto the ones that leave you out. Brands on many lists show up no matter which ones the engine reads that day.