Research

AI doesn't pick brands. It reads lists. We checked 1,473 of them.

839 AI answers, 1,473 cited 'best X' pages, 21 categories. What actually gets a brand into the answer, and what to do about it.

Preliminary: based on day 1 of a 7-day run (17 November). We'll update every number with the full week on 24 November.

Most advice on getting recommended by ChatGPT is about your own website: add schema, write an llms.txt, use more headings. We wanted to know what actually decides which brands make it into an answer, so we looked at what the AI read before it answered.

We asked ChatGPT, Gemini, Google AI Overviews and Google AI Mode 210 buying questions ("best CRM for a small team", "cheaper alternative to Zendesk") across 21 categories. Then we fetched the pages each answer cited. 43% of all citations were "best X" or "top X" lists, so we downloaded those 1,473 lists and checked which brands were on them.

The short version

  1. 90% of the brands AI recommended were on a list the engine cited for that answer. For the brand ranked first, it was 94%.
  2. A brand on none of the cited lists was recommended 2.5% of the time. On one list, 22%. On three or more, 44%, and on ChatGPT 65%.
  3. It holds for the same brand. A brand was recommended in 27% of the answers where it was on a cited list, and 5% of the answers where it wasn't. 272 of 313 brands did better when listed.
  4. The lists are a short, findable set. In each category, ten sites carry 38–90% of the list citations. Many are small blogs and vendor blogs, not big publishers.
  5. Writing your own list works too. When an engine cited a brand's own "best X" page, that brand was recommended 68% of the time, against 14% otherwise.
  6. Dates matter. ChatGPT added "2026" to its web searches for every single question. 83% of the pages Gemini cited had 2026 in the title or URL.

So the practical question isn't "how do I make AI like my website". It's "which lists does AI read for my buyers' questions, and am I on them?"

What we did

We asked the 210 questions once on each engine on 17 November 2026, as a US user in English, giving 839 answers with 6,867 citations. A citation counted as a list when its title or URL said "best" or "top". That gave 2,938 list citations pointing to 1,535 distinct pages. We fetched them the same day and got 1,473 (the rest blocked us).

For every answer, we checked each brand in that category against the text of the lists the answer cited. A brand counted as recommended when the answer put it forward (in a heading, in bold, as a link, in a table or at the start of a list item), not when it was only mentioned in passing. The method is the same as in our first study.

Finding 1: if you're not on the lists, you're almost never in the answer

Cited lists naming the brandRecommended in the answer
02.5%
121.5%
231.6%
3 or more44.1%

Out of 3,284 recommendations in answers where we could read the cited lists, 2,943 (90%) went to a brand on at least one of them. The answer's top pick was on a cited list 656 times out of 695.

By engine, the jump from zero lists to three or more:

EngineOn 0 cited listsOn 3+ cited lists
ChatGPT1.6%65.0%
Google AI Mode3.0%48.4%
Google AI Overviews3.3%47.6%
Gemini2.1%37.5%

Isn't this just big brands being everywhere? Partly, which is why we compared each brand with itself. Across the 313 brands we saw both on and off the cited lists, the same brand was recommended in 26.6% of answers when it was listed and 4.6% when it wasn't. Wave, the accounting app, was recommended in 26 of the 33 answers where a cited list named it, and none of the 5 where none did.

This is still a correlation. Lists and answers both favour brands that fit the question. But the engines told us themselves which pages they read, and nine in ten of the brands they picked were on those pages.

Finding 2: the lists that matter are a short list

Each category had 21 to 78 different sites supplying lists, but the top ten carried most of the citations: 38% in accounting software, 47% in CRM, 63% in email marketing, 90% in home security.

They aren't all household names. In CRM, the most-cited list sites included two CRM vendors' own blogs. In email marketing, one of the most-cited lists came from a small email tool. In AI search tools, the three most-cited list sites were all companies in the category.

That means getting on these lists is a task, not a dream. Find the ten or so lists AI cites for your buyers' questions, check whether you're on them, and pitch the ones you're missing. A small blog that updates its list every quarter is easier to get onto than a top-ten Google ranking.

Finding 3: write the list yourself

Brands publish "best X" pages that include themselves, and the engines cite them. In 120 cases an answer cited a list written by a brand in that category, and that brand was recommended in 82 of them (68%). In every other answer, brands were recommended 14% of the time.

Be careful with this one. A list where the author ranks itself first and dismisses everyone else is obvious to readers. The pages that get cited are the honest ones: a real comparison where your product is one of several, with who each option is best for.

Finding 4: each engine reads different things

ChatGPTGeminiAI OverviewsAI Mode
Answers citing a "best X" list88%96%85%70%
Answers citing YouTube0%11%58%23%
Answers citing Reddit0%27%38%34%
Recommended brands whose own site is cited51%5%13%11%
  • ChatGPT reads your homepage. It cited a recommended brand's own site about half the time, and 70% of the brand pages it cited were homepages. 80% of its answers quoted prices. If your homepage doesn't say what you cost and who you're for, ChatGPT is filling that in from somewhere else.
  • Gemini reads lists, almost only lists. Two thirds of its citations were "best X" pages, and only 2% were brand sites.
  • Google's AI engines read YouTube and Reddit. AI Overviews cited a YouTube video in 58% of answers. ChatGPT cited neither.

Finding 5: put the year on it

ChatGPT records the web searches it runs before answering. For all 210 questions, it added "2026" to its search. After the year, the words it added most were "comparison" (37% of questions), "pricing" (34%), "reviews" (29%) and "features" (22%).

Gemini's citations followed suit: 83% had 2026 in the title or URL. A list titled "best CRMs of 2024" is invisible to a search for "best CRM 2026".

What to do this week

  1. Write down the 10–20 questions your buyers ask AI. "Best X for Y", "cheaper alternative to Z", "X that integrates with W".
  2. Ask all four engines and collect the lists they cite. Note which ones name you and which don't.
  3. Pitch the lists you're missing. Start with the small, recently updated ones. Offer what list writers need: a clear "best for", current pricing, a screenshot.
  4. Publish your own honest comparison, with the year in the title, and update it when the year changes.
  5. Fix your homepage for ChatGPT: price, who it's for, and the job it does, in plain words near the top.
  6. If you sell to consumers, get into YouTube and Reddit, where Google's AI looks.
  7. Re-check monthly. Lists get updated, and new ones replace old ones.

Doing step 2 by hand for 20 questions on four engines takes an afternoon. Supercite does it every day: it asks your buyers' questions on all four engines and lists the pages AI cites that don't mention you yet, so you know exactly who to pitch.

Method and limitations

  • One day, one market. Day 1 of a 7-day run, US English, one answer per question per engine. AI answers vary from run to run.
  • Lists are found by title. A page counted as a list when its title or URL said "best" or "top". Some lists are missed and a few non-lists are included.
  • Brand on the page, not ranked on the page. We checked whether a brand's name appears anywhere on the list page, so a brand mentioned in passing counts as "on" the list. We fetched pages on the same day, but a page can change between the AI reading it and us reading it.
  • Correlation. The same-brand comparison removes the "big brands are everywhere" effect, but not every difference between questions.
  • Data. The answers, brand lists and counting script are in the first study's downloads.

FAQ

Do I still need to optimise my own website?

Yes, especially for ChatGPT, which cites brand homepages about half the time. But on Gemini and Google's AI engines, a recommended brand's own site was cited in only 5–13% of cases. Your site gets you described correctly. The lists get you picked.

How do I find out which lists AI cites for my category?

Ask your buyers' questions on ChatGPT, Gemini, AI Overviews and AI Mode and open the sources on each answer. Supercite automates this daily and flags the pages that don't mention you.

Is writing my own "best X" list spammy?

It is if you rank yourself first with no real comparison. An honest comparison that includes competitors, with who each one suits, is the kind of page the engines cite.

Why does the year matter so much?

ChatGPT added the current year to its searches for every question we asked, and most pages Gemini cited carried that year. Pages without it are less likely to come up in those searches.

All posts →

Make AI search your next revenue channel

Start free trial