How AI picks which brands to name
By the CiteTrace team · Updated 2026-11-10 · 8 min read
Nobody outside the labs knows exactly how an answer gets built, and we are not going to pretend. What we can do is watch thousands of answers and describe what repeats. Everything below is observation from our own runs, labelled as such, plus a few named sources.
Short version: the assistant searches, reads a few pages, and names the brands those pages agree on. Which is why your homepage matters less than you think.
The pipeline, as far as we can see it
- The assistant turns your question into one or more search queries. With web search on, ChatGPT, Perplexity and Gemini all do this; a model with no web access answers from what it memorised in training.
- It fetches a handful of pages and reads them. Those are the citations you see under the answer.
- It writes an answer that favours brands several of those pages agree on, and it tends to produce a short list.
Because step 2 exists, "how do I get named" and "how do I get on the pages it reads" are the same question.
What our runs show
We ran 11 audits (9 companies, 10 buyer questions each, 4 engines per question) on 10 Nov 2026. Four things stood out. These are small samples, so read them as a pattern, not a law.
- The audited company was named in a median of 5 of 10 questions (range 2 to 9). The top competitor was named in at least 7 of 10 in every single audit, and in 10 of 10 in six of them.
- Citations lean on a few sources. In the cited-domain lists of those audits, vendor sites and vendor blogs (HubSpot's blog, Salesforce, Bright Data), review sites (G2, Capterra) and big how-to publishers (Zapier) kept coming back.
- Engines disagree. Totalled over the 110 answers each, Gemini named the audited company 43 times, Perplexity 38, ChatGPT with web search 31, and a ChatGPT model with no web 32. The no-web model is not a safe stand-in for the others.
- Repeating a run changes the result. The same company, the same 10 questions, three runs on one day: named in 5, 5 and 2 questions.
Consensus beats volume
In our category research, the brands named most often are the ones that appear on many different cited pages, not the ones with the loudest marketing. Being on one big listicle helps. Being on six independent pages helps more.
This is also why big, established vendors are hard to dislodge: they are already on every page. A smaller product wins specific questions ("for remote teams", "with a free tier") long before it wins the generic one.
Who writes the pages the AI reads
Three groups, roughly. Vendors writing about themselves and each other, publishers writing roundups, and review sites aggregating customers. Communities such as Reddit appear too, but how much has swung. DesignRush, citing Semrush, reported ChatGPT's Reddit citation share falling from about 60% in early August 2025 to about 10% by mid-September 2025. Whatever the right number is today, it moves.
You control the first group fully, can influence the third with real reviews, and can reach the second with honest outreach.
Why the same question gets different answers
Answers are sampled. Wording, location, time, personalisation and the pages that happen to be fetched all shift the result. Digiday quoted an agency chief executive on this in May 2026: use three tools with the same prompts and you get three different answers.
Which is the reason we track daily and report a trend. Any one screenshot, including ours, is one draw.
What that means for you
- Find out which pages are cited for your questions. That list is your real marketing plan.
- Get accurate, current, fair pages onto it: yours first, then third parties'.
- Judge by trend over weeks, not by one answer.
The full process is in How to get your SaaS recommended by ChatGPT. To see your own citations, run the free audit.
Frequently asked questions
Does ChatGPT use Google or Bing to find pages?
OpenAI documents two crawlers on its site: OAI-SearchBot for search and GPTBot for training. We do not claim to know the exact retrieval pipeline behind an answer; we measure which pages end up cited.
Why does Perplexity cite more sources than ChatGPT?
Perplexity is built as an answer engine around search and shows its sources by design. In our runs, citation lists differ by engine, which is why we track each engine separately.
Does domain authority matter?
Large, established sites appear disproportionately in citations in our runs, so being on those pages matters. Whether a model weights authority directly or simply reads whatever search returns first, we cannot say.
Can I see which engines name me?
Yes. The free audit shows named-or-not per engine for 10 questions, plus the cited pages.
Is any of this stable?
No. Engines, models and sources change. That is why a baseline and a re-run schedule matter more than any single tactic.
See who the AI recommends instead of you
A free 10-question audit: 4 engines, the pages they cite, three fixes to start with.