Skip to content

Blog

What 1,575 AI answers about AEO tools show: sources, engines and buying stages

A month of AI answers about our own category, read by engine and by stage. The engines cited different sources, and the shortlist stage was the most crowded.

Updated Jack F, Co-founder

ResearchAEOAI visibilityAI citationscustomer journeyshare of voiceChatGPTGoogle AI Overviews

Now readingThe findings in brief

Across 1,575 AI answers to 25 questions about AEO tools (UK, 6 October to 4 November 2026), ChatGPT and Google AI Overviews cited different sources: vendor sites and documentation in ChatGPT; YouTube, LinkedIn and Reddit in AI Overviews. Brands were named least at the Learn stage and most at Consider, where Profound, Peec AI and Semrush each appeared in about half of answers. No domain held more than 4.0% of the 11,097 citations. One category, one market.

What did 1,575 AI answers about AEO tools show?

ChatGPT and Google AI Overviews built their answers from different sources, and brands were named at very different rates depending on how far into a purchase the question came. Over 30 days we put the same 25 questions, the kind a buyer of answer engine optimisation (AEO) software asks, to three AI engines in the UK and kept every answer. Answyn sells a product in this category, so the brands counted here are its competitors.

Where does an assistant switch to a competitor? At the shortlist. The most-named brand appeared in 13.7% of answers to Learn questions; at Consider, Profound, Peec AI and Semrush each appeared in about half.

Which pages do the engines cite? ChatGPT's five most-cited domains were the sites of vendors that sell AI visibility tools, followed by documentation from Google and OpenAI. AI Overviews' three most-cited were YouTube, LinkedIn and Reddit.

Do the two engines draw on the same sources? Only partly. The vendor sites appeared in AI Overviews at a fraction of their ChatGPT counts, and all 380 youtube.com citations came from AI Overviews.

What does a real citation share distribution look like? Flat at the top. The most-cited domain held 4.0% of 11,097 citations, and Answyn's own domain held 0.2%.

Do ChatGPT and Google AI Overviews cite the same sources?

Only partly. Of the ten domains cited most often across the three engines, five were the sites of vendors that sell AI visibility tools, cited mainly by ChatGPT. Two were official documentation, cited almost only by ChatGPT. The other three were YouTube, LinkedIn and Reddit, cited almost only by AI Overviews.

The ten most-cited domains of 11,097 citations, 6 October to 4 November 2026. Claude's 75 answers make up the difference between the two engine columns and the total.
DomainCitationsShare of all citationsChatGPTAI Overviews
semrush.com4454.0%36459
otterly.ai4363.9%33078
peec.ai4103.7%36050
youtube.com3803.4%0380
tryprofound.com3222.9%29619
linkedin.com2972.7%0296
ahrefs.com2532.3%22519
developers.google.com1801.6%1782
reddit.com1621.5%2160
help.openai.com1301.2%1300

Each engine's own totals make the split plainer. The five vendor domains supplied 28% of ChatGPT's 5,554 citations and 5% of AI Overviews' 4,474. YouTube, LinkedIn and Reddit supplied 19% of AI Overviews' citations and 2 of ChatGPT's.

Our own domain followed the same line: answyn.com was cited 17 times, all by ChatGPT, and never by AI Overviews or Claude.

So, is it worth checking AI assistants separately, or do they all pull from the same sources? On this evidence, check them separately. A reading of ChatGPT alone would have found answyn.com among its sources, and a reading of AI Overviews alone would have found it nowhere. Neither result describes the other engine.

Claude answered each question 3 times, against 30 for each of the other two. That is too few to compare, so we have not set Claude against them.

Which pages did the engines cite?

ChatGPT cited three kinds of page most: vendors' own product, pricing and feature pages; roundups of tools; and official documentation from Google, OpenAI and Microsoft, on developers.google.com, help.openai.com, support.google.com and support.microsoft.com. One roundup, technologyadvice.com's "Best AI search monitoring tools for 2026", was cited in 99 runs across 15 questions without Answyn being named. For research-style questions, ChatGPT cited studies with numbers.

AI Overviews leaned on YouTube videos and Shorts, LinkedIn posts and articles, Reddit threads and Facebook posts. A single YouTube Short, "How to Check Your AI Visibility", was cited in 19 runs.

That bears on another question we track: what pages should we write to get picked up by AI answers? The answer depends on the engine. In ChatGPT, vendors' own product, pricing and feature pages were the most-cited pages in this category, which suggests a company's own site can compete there. In AI Overviews, YouTube, LinkedIn and Reddit held 19% of citations against 5% for the five vendor sites, so the work there includes video and posts as well as pages. The data shows which pages the engines chose, never why.

We have graded the published evidence on what earns citations, including the share that goes to sites a brand does not own, in how brands earn citations in AI answers.

What does a real citation share distribution look like?

Flat at the top, with a long tail. Here a domain's share is its citations divided by all 11,097 in the window. The most-cited domain, semrush.com, held 4.0%. The ten most-cited held 27% between them, which left 73% for domains that each held 1.2% or less. answyn.com held 17 citations, or 0.2%, all from ChatGPT.

Within each engine the leaders held a little more: semrush.com held 6.6% of ChatGPT's citations, and youtube.com held 8.5% of AI Overviews'.

So what counts as a good citation share benchmark for a mid-sized brand? There is no universal figure, and this is one category. In this one, the most-cited domain held 4.0% and the tenth held 1.2%, so a share in low single figures put a domain among the leaders. Compare your share with the domains cited for your own questions, engine by engine, rather than with a figure from another category.

Does ChatGPT search the same way every time?

No. ChatGPT searched the web in all 750 of its runs, usually two or three times per answer in the runs we read. Over 30 runs, one question produced anywhere from 26 to 95 different search wordings, depending on the question.

Each wording can bring back different pages, so a single check of one question samples one path out of dozens. That is why this study counts every answer over 30 days rather than reading one, and why a visibility figure needs a denominator. We worked through what that means for planning content in how much content it takes to move AI visibility.

Which brands do AI answers name at each buying stage?

Few at Learn, the most at Consider, and fewer again at Purchase. Each tracked question carries a buying stage, and AEO Hub groups the stages in three: Learn for questions about a problem, Consider for questions about options and comparisons, and Purchase for the checks a buyer makes before committing.

Mention rates per stage as Answyn reports them, on the answers analysed for brand mentions at that stage. A brand counts when the answer names it in its own words, and each spelling is counted on its own, so the Otterly.AI figures are a floor.
StageQuestionsBrands named most often (mention rate)
Learn7Semrush 13.7%, Ahrefs 12.1%, Profound 5.8%
Consider14Profound 52.1%, Peec AI 49.2%, Semrush 48.8%, Otterly.AI 30.4%, Ahrefs Brand Radar 23.5%
Purchase4Semrush 31.0%, Profound 20.2%, Peec AI 17.3%, Otterly.AI 12.1%, Ahrefs 9.7%

Read down the table. At Learn, the most-named brand, Semrush, appeared in about one answer in seven. At Consider, three brands each appeared in about half of the answers, and two more passed 20%. At Purchase the shares fell again and the order changed, with Semrush ahead of Profound.

That answers one of the questions we track: at what point in someone's research does an AI assistant stop mentioning a brand and switch to a competitor? In this category, at Consider, the moment a buyer asks what the options are. Before it, few brands are named at all. From there, the answers fill with names, and a brand that is not among them has lost its place to the ones that are. At Purchase the leading brands were named less often, and the leader at Consider was not the leader there.

Behind the leaders at Consider came Ahrefs (16.9%), SE Ranking (13.1%), Scrunch AI (11.6%), HubSpot AEO (9.0%) and Rankscale (8.8%). Ahrefs and Ahrefs Brand Radar are counted as separate names because the answers used both.

Answyn was named in 2 of the 1,575 answers, both at Consider, both by ChatGPT and both for "Top answer engine optimisation software for UK marketing teams", on 12 October and 4 November. A brand that is missing at every stage has no switch point to find, because it never held a place to lose.

Two cautions. These figures count how often answers named each product, which says nothing about which product is better. And three of the 25 questions name a product in their wording: one asks whether Peec AI or Otterly.AI is better value, and two ask about moving away from Rankscale and BrandLight. Answers to those questions will tend to name those products, which lifts their figures at the stage where each question sits.

For how to read your own visibility stage by stage, see customer journey analysis in AI search.

Why do brand counts need merged spellings?

Because answers spell the same brand several ways, and a count that keeps the spellings apart undercounts the brand. In Consider answers, Otterly appeared as "Otterly.AI" (262 answers), "OtterlyAI" (71), "Otterly.ai" (39), "Otterly AI" (28) and "Otterly" (20). Scrunch appeared as "Scrunch AI" (100) and "Scrunch" (41), and AthenaHQ as "AthenaHQ" (50), "Athena" (7) and "Athena HQ" (4).

The raw lists behind this study keep each spelling as the answer wrote it, so the Otterly.AI and Scrunch AI figures above count one spelling each. They are floors: the share of answers naming those brands in any spelling is higher, by an amount these lists cannot show. Adding the spelling counts together is no fix either, because one answer can use two spellings and would then count twice.

The sound method is to merge the spellings into one brand before counting, then count each brand once per answer, which is how share of voice is defined. For your own brand, Answyn counts the aliases on the brand profile as the same brand, and a missing alias makes real mentions look like absences. Before comparing share-of-voice figures from any tool, ask how it handles spellings.

Why doesn't being read mean being named?

Because a mention needs the brand in the answer's own words, and retrieving a page does not put it there. In 49 ChatGPT runs across 5 questions, ChatGPT's search retrieved a page on answyn.com and the answer still left Answyn out. Across the whole window, ChatGPT named Answyn in 2 answers.

ChatGPT runs, 6 October to 4 November 2026. 49 runs in all.
Question as trackedRetrieved an Answyn page, did not name AnswynPage retrieved
Trying to understand at what point in someone's research an AI assistant stops mentioning our brand and switches to a competitor instead18 of 30 runsCustomer journey analysis in AI search (Handbook)
Which AEO platform should a small brand team choose to start improving AI search presence?17 of 30 runsThe 10 best AEO platforms for UK businesses in 2026 (Blog)
Top answer engine optimisation software for UK marketing teams8 of 30 runsThe same list
Best software for mapping brand mentions across different stages of the customer journey in AI answers4 of 30 runsAn Answyn page; on 31 October, the customer journey handbook
We want to see where AI engines stop mentioning us during the buying journey, what tool handles that?2 of 30 runsAn Answyn page

The first row is the journey question from the section above. ChatGPT retrieved our customer journey handbook in 18 of its 30 runs and cited studies with numbers, such as buyer-journey surveys and research on query intent from SEO vendors, agencies and publications. The data cannot tell us why ChatGPT preferred them. It does show that a page explaining a concept can be used without its publisher being named. We published this study partly for that reason: it puts stage-level numbers from a real category on the record.

Retrieved, cited and named are different counts. A retrieved page is one the engine's search brought back. A citation is a link the answer shows. A mention is the brand's name in the answer's own words. answyn.com collected 17 citations in the window, all in ChatGPT, against 2 mentions. For what to check when your own pages are read but not named, see why ChatGPT leaves a company out.

What this means if you are not an AEO vendor

One caution first. AEO vendors publish about the same subject their buyers ask about, so their own sites may be unusually easy to cite on these questions, and a category where vendors publish less may look different. With that caution, these are the habits we would take from the data:

  • Read each engine on its own. ChatGPT and AI Overviews drew on different sources here, and answyn.com was cited 17 times by ChatGPT and never by AI Overviews.
  • Expect the most competition at the shortlist. Three brands each appeared in about half of Consider answers, against 13.7% for the most-named brand at Learn.
  • Judge visibility over many answers. ChatGPT searched differently from run to run, so one check of one question tells you little.
  • Merge the spellings of every brand you count, your own included, and count each brand once per answer.
  • If you can see which of your pages an engine retrieves, set that beside the answers that name you. Retrieved and not named is a different problem from never found.

One more habit comes from outside this dataset: check what the answers say about you, as well as whether they name you. Answyn's Brand Facts for the same period, which cover all of Answyn's tracked questions including branded ones, recorded ChatGPT repeating a trial offer Answyn no longer makes, in 6 answers between 10 October and 2 November. Access is by explicit invitation. A Brand Fact marked incorrect, with a correction, does not change what ChatGPT says, but it keeps a record of the claim so you can see whether it stops.

Method and limits

We tracked Answyn's own category, AEO and AI visibility tools, in the UK market in English for 30 days, from 6 October to 4 November 2026. We asked 25 questions in 8 topics, none of which names Answyn, of ChatGPT and Google AI Overviews 30 times each and of Claude 3 times each, last on 30 October. That is 63 answers per question and 1,575 in all: 750 each from ChatGPT and AI Overviews, and 75 from Claude. The answers carried 11,097 citations: 5,554 in ChatGPT, 4,474 in AI Overviews and 1,069 in Claude. We pulled the figures from Answyn on 4 November 2026.

The topics were citation share (5 questions), AI engine coverage (5), content gap briefs (3), buyer journey stages (3), sentiment in AI answers (3), share of voice (2), AI crawler activity (2) and answer engine optimisation (2). Each question carries a buying stage: 7 Learn, 14 Consider and 4 Purchase. Three questions name another product: one asks whether Peec AI or Otterly.AI is better value, and two ask about moving away from Rankscale and BrandLight.

A brand counts as named when it appears in the answer's own words; a name that appears only in a cited link is a citation, not a mention. The stage shares are Answyn's mention rates, as Answyn reports them: their base is the answers analysed for brand mentions at each stage, so a run with no answer, including the 30 AI Overviews runs where no overview appeared, never counts as a brand being absent. Each spelling of a name is counted as the answer wrote it.

Answyn is the tracked brand and sells a product in this category, so every brand counted here is a competitor. We have reported their figures as the answers produced them. They describe how often AI answers named each product, not how good the products are. Claude is not on Answyn's standard daily roster; a plan can add it by agreement.

What the data cannot show:

  • It covers one category in one market, the UK in English, so it says nothing directly about other categories or countries.
  • It has no search or prompt volume, so it cannot say how often real buyers ask these questions.
  • It shows which sources the engines cited and which pages ChatGPT retrieved, never why an engine chose them.
  • Gemini, Google AI Mode, Microsoft Copilot and Perplexity were not collected in this window, and Claude's 3 answers per question are too few to compare with the other engines.
  • The 25 questions are the ones Answyn tracks. A different set would produce different shares.

The collection and analysis ran in Answyn the way they do for any tracked brand: a fixed set of questions asked on a schedule, every answer stored, and every figure traceable to the answers behind it. Access is by invitation, and every plan is quoted for the business using it. If you want the same reading of your own category, enquire.

Questions people ask

At what point does an AI assistant switch to a competitor?
In this study, at the shortlist. The most-named brand appeared in 13.7% of answers to Learn questions, while Profound, Peec AI and Semrush each appeared in about half of Consider answers, and the leader at Purchase, Semrush, in 31.0%. If a brand is missing from Consider answers, that is where competitors take its place.
Do ChatGPT and Google AI Overviews use the same sources?
Only partly, in this category. YouTube, LinkedIn and Reddit supplied 19% of AI Overviews' 4,474 citations and 2 of ChatGPT's 5,554, while the five vendor domains ChatGPT cited most supplied 28% of its citations and 5% of AI Overviews'. A source one engine leans on can be close to absent from the other.
Is it worth checking AI assistants separately?
On this evidence, yes. answyn.com was cited 17 times by ChatGPT and never by AI Overviews or Claude, so a check of one engine would have said nothing reliable about the others. Read each engine on its own before you blend anything.
What kinds of pages do AI answers cite?
In this category, ChatGPT cited vendors' own product, pricing and feature pages, roundups of tools, official documentation and, for research questions, studies with numbers. AI Overviews leaned on YouTube videos, LinkedIn posts, Reddit threads and Facebook posts. Which kind of page wins depends on the engine, so check each one.
What is a good citation share benchmark?
There is no universal one, and this study covers one category. Here the most-cited domain held 4.0% of 11,097 citations and the tenth held 1.2%, so a share in low single figures put a domain among the leaders. Benchmark against the domains cited for your own questions, engine by engine.
Which AEO tools do AI answers name most often?
In this study's Consider answers, Profound (52.1%), Peec AI (49.2%) and Semrush (48.8%), then Otterly.AI (30.4%) and Ahrefs Brand Radar (23.5%). At Purchase, Semrush led at 31.0%. The shares count how often answers named each product, not how good the products are, and Answyn, which sells a competing product, was named in 2 of the 1,575 answers.
Does ChatGPT run the same searches every time?
No. It searched the web in all 750 of its runs here, usually two or three times per answer, and over 30 runs one question produced anywhere from 26 to 95 different search wordings. One manual check samples one of those paths, which is why visibility is worth reading over weeks of answers.
How should share of voice handle a brand with several spellings?
Merge the spellings into one brand, then count each brand once per answer. In this study's Consider answers Otterly appeared under five spellings, from Otterly.AI in 262 answers to plain Otterly in 20. Adding the spelling counts together would double count answers that use two spellings, and keeping them apart undercounts the brand.
Why would ChatGPT read a page and still not name the brand?
The data cannot say why, only that it happens. ChatGPT retrieved an Answyn page in 49 runs that left Answyn out, and for the journey question it cited studies with numbers. A mention needs the brand in the answer's own words, and being among the pages the engine looked at does not put it there.

These figures come from Answyn's own tracking of its own category, pulled from Answyn on 4 November 2026. Answyn sells a product in this category, so the brands counted are its competitors. If you find a figure we have got wrong, tell us at team@answyn.com and we will correct it and date the correction.

See what the engines already say about you.

Answyn keeps the daily record: mentions, citations, and the answers behind every figure.