Lab reportEngines6 min
How AI engines cite adult brands, and which engines to watch
Abstract
There is no single "AI search". Usage is spreading across engines, most referral clicks still come from one of them, and each picks its sources in its own way. For adult brands, engines' safety policies add another layer. This report sets out what is known and how to test the rest.
01Usage is spreading, clicks are not
Two kinds of data describe the engines, and they tell different stories. Estimates of web visits, modeled from Similarweb's panel and published by Momentic, put ChatGPT at about 54% of visits to the largest AI chatbots in May 2026, with Gemini near 28% and Claude around 9%, a far more even spread than a year earlier.
Referrals, the clicks AI chatbots send on to other websites, remain concentrated. Statcounter recorded ChatGPT's share falling to a low of 76.85% in April 2026, then back to about 79% in August, with Gemini near 11% and Perplexity around 4%. Its CEO noted that ChatGPT had stood above 84% a year earlier. For a site owner the practical reading is simple: many people now ask several assistants, but most of the visits you can see in analytics still arrive from one.
One caution when reading these charts. Statcounter's AI chatbot figures count referrals from standalone assistants. Clicks from Google's AI Overviews and AI Mode arrive as ordinary Google search visits, so they are not in the chart at all. A ChatGPT share of about 79% means ChatGPT sends most of the referrals that can be traced to chatbots, not that it sends most of the visits AI now influences. Search Console's generative AI report is the place to see the Google side.
- ChatGPT79.4%
- Gemini10.9%
- Perplexity4.31%
Share of AI chatbot referrals to websites, August 2026 (Statcounter Global Stats, as reported). Shares move from month to month.
02Being crawled is not the same as being visited
AI companies crawl far more pages than they send visitors back to. Cloudflare publishes crawl-to-referral ratios through its Radar service, and the figures reported for mid-2026 show Google crawling a handful of pages for each visit it refers, while several AI companies crawl hundreds of pages per visit. Allowing an AI search crawler is still worth it if you want to be cited, but it should be a decision made with that ratio in mind, not an assumption that crawling brings traffic. You can work out your own ratio by setting each bot's requests in your server logs against the visits its assistant refers in analytics; a sharp rise in one without the other is worth investigating. The crawler settings themselves are covered on AI search optimization.
03Each engine chooses sources differently
Comparisons of AI citations with Google's organic results now point to a looser link than they did in 2025. Ahrefs found that only about 12% of the links cited by ChatGPT, Gemini and Copilot ranked in Google's top 10 for the same prompt, with Perplexity the most search-aligned, and that 38% of AI Overview citations came from top-10 pages in March 2026, down from 76% the year before. Even Google's two AI surfaces disagree: Ahrefs found AI Overviews and AI Mode citing the same page in only a small minority of cases.
So strong organic rankings still help, most of all with Perplexity and AI Overviews, but each engine reaches past the first page in its own way, frequently to supporting pages on sites it already trusts, or to outside sources that write about you. How we act on that is set out on AI citation optimization.
04Engine by engine
| Engine | Where it finds sources | What to check for your brand |
|---|---|---|
| ChatGPT | Its own search index, built by OAI-SearchBot, plus what it learned in training | That OAI-SearchBot is allowed, and how third-party sites describe you |
| Google AI Overviews and AI Mode | Google's search index, often deeper pages on trusted sites | Indexing, snippet eligibility and safe-for-work answer pages |
| Gemini app | Google's systems, governed separately by the Google-Extended token | Your Google-Extended choice, which the Search Console AI setting does not cover |
| Perplexity | Live search results, with many citations per answer | Ordinary rankings and fresh pages |
| Claude | Its search index, built by Claude-SearchBot | That Claude-SearchBot is allowed; its referral share is small but growing |
05The extra layer for adult brands
Every engine applies its own safety policies, and those policies shape answers on sexual products, sexual health and adult entertainment. The effect need not be a blanket refusal. It can show up as answers that lean on reference and health sources, leave out brand names, or shift with small changes in wording.
In our testing the same question can be answered fully by one engine and declined by another, and answered differently by the same engine on different days. Wording matters too: "how to choose a body-safe toy for a beginner" tends to get a full, sourced answer, while a blunter phrasing of the same need can be refused or answered vaguely. That is why we do not assume how any engine treats your category; we test it. The questions that tend to produce useful, citable answers are factual and safe for work, which is where content effort should go. The foundations are compared in adult GEO vs SEO.
06Which engines to watch for your brand
Start from your own analytics rather than market share charts. Check which assistants already send visits to your site and which your buyers mention, then weigh the engines by that. A US audience leans more on Gemini and Copilot than the worldwide figures suggest; a younger audience may lean on ChatGPT; buyers researching health questions may meet Google's AI features first. Most brands should watch Google's AI features and ChatGPT closely, Perplexity and Gemini regularly, and the others occasionally.
A practical first step is a saved segment in your analytics for visits referred by the assistants' own domains, such as chatgpt.com, perplexity.ai, gemini.google.com, copilot.microsoft.com and claude.ai. Compare it each month with organic search, and look at which pages those visitors land on. The landing pages tell you what each assistant is citing, often more clearly than any tracking tool, and they show you which answers are worth protecting.
07A simple test you can run
- Write ten questions your customers actually ask, in their words and in safe-for-work language.
- Ask each one in ChatGPT, Gemini, Perplexity and Google, signed out where possible, and from the country your buyers are in.
- Run each question three times, because answers change between runs.
- Note if the brand appears, if a link to your site appears, which other sites get linked, and anything said about you that is wrong.
- Repeat monthly with the same questions, and look at the trend rather than any single result.
That is the core of the readiness audit; every other check we run sits in our readiness checklist.
08Official sources
Where to read the originals. Guidance can move on after a guide is published.
QQuestions
Q.01
Which engine should an adult brand prioritize?
Start with whichever your customers use, then the one sending you visits. Across the web, ChatGPT still sends most AI referral traffic while Gemini has gained the most share of usage, so it is worth watching at least both, plus Google AI Overviews.
Q.02
Do AI engines refuse to mention adult brands?
Not necessarily, but each applies its own safety policies, and answers vary with how a question is worded. Test the questions your customers really ask, in plain language.
Q.03
Why does the same question get different sources on different days?
Answers are generated fresh and retrieval shifts as indexes update. That is why we track a fixed set of questions over months rather than trusting one screenshot.
Q.04
Should we optimize for every AI engine equally?
No. Start from the engines that already send you visits and that your buyers use. For most brands that means Google's AI features and ChatGPT first, then Perplexity and Gemini.