Skip to main content

AI search

How to rank in AI Overviews and be cited by AI assistants

How to rank in AI Overviews comes down to less than most advice claims. Google's own guide asks for a page that is indexed, eligible to be shown with a snippet, and worth citing.

Book a discovery call

At a glance

What this page is
A plain account of what each AI platform says about being cited
Where it comes from
Each company's own documentation, opened 4 October 2026
What we tested
Nine buyer questions, four assistants, 285 citations
Platforms covered
Google's AI Overviews and AI Mode, ChatGPT, Gemini, Perplexity and Claude

How to rank in AI Overviews, according to Google

Google says its generative AI features, such as AI Overviews and AI Mode, are rooted in its core Search ranking and quality systems. So optimising for AI Overviews (optimizing, in US spelling) starts with ordinary SEO.1

There is one technical requirement. To appear as a supporting link, a page must be indexed and eligible to be shown in Google Search with a snippet.1

Beyond that, Google recommends content that is helpful, reliable and written for people, and that offers a unique point of view. It also recommends a clear technical structure and a good page experience.1

Google also describes query fan-out: the model issues a set of related searches at once to gather more information. We read that as a reason to cover a subject fully, not to chase one phrase.1

This is the whole of what Google asks for. Most AI search optimisation advice goes further, and the next section sets the two side by side.

What the platforms say, and what you will read elsewhere

Much of the advice in circulation is not in any platform's documentation. On we read Google's AI Overview for this search. We also asked four assistants how to be recommended by ChatGPT and how to be cited in AI Overviews. Here are five things they advised, beside what the companies themselves say.

Exhibit 1

Common advice, checked against the platforms' own documents
AdviceWhere we saw itWhat the platform says
Write a 40 to 60 word answer under each headingGoogle's own AI Overview for this searchGoogle's guide says there is no need to break content into tiny pieces or to write in a specific way for AI.1
Add FAQPage or HowTo schemaThe same AI OverviewGoogle's guide says structured data is not required and there is no special schema.org markup.1
Add an llms.txt fileGemini's answer about ChatGPT, as optionalGoogle says no new machine-readable files or AI text files are needed. The crawler pages from OpenAI, Perplexity and Anthropic do not ask site owners for one.1245
Rank in Bing, because ChatGPT uses itGemini's and Claude's answers about ChatGPTOpenAI names no search engine. It says it may obtain a page's address from a third-party search provider or by crawling other pages.3
Get mentioned on other websitesAll four assistants' answers about ChatGPTGoogle says seeking inauthentic mentions is not as helpful as it might seem. OpenAI's two pages say nothing about mentions.123

Sources: the documents numbered at the foot of this page, opened . The AI Overview and the assistants' answers were recorded on and vary between runs.

None of this makes the advice useless. A clear answer near the top of a page helps a reader, and that is reason enough to write one. It does mean no platform says these things are required, so nobody can honestly sell them as a result.

Which crawler decides whether you can be cited

Being cited starts with being read. Each platform uses a different crawler for search, and most keep it separate from the crawler that collects training data.

Exhibit 2

The search crawler for each platform
PlatformCrawlerWhat its owner says
Google's AI Overviews and AI ModeGooglebotThe page must be indexed and eligible to be shown with a snippet.1
GeminiGoogle-Extended, a control and not a separate crawlerIt governs whether your content is used to train Gemini models and to ground their answers. It does not affect Google Search.6
ChatGPTOAI-SearchBotIt is used to surface websites in ChatGPT's search features. GPTBot is separate and collects training data. A robots.txt change takes about 24 hours to apply.2
PerplexityPerplexityBotIt surfaces and links websites in Perplexity's search results and is not used for training. Perplexity asks you to allow it and its published IP ranges.4
ClaudeClaude-SearchBotIt is used to improve search results. Blocking it may reduce your visibility in them. Anthropic says its bots respect robots.txt.5

Sources: Google16, OpenAI2, Perplexity4 and Anthropic5, opened .

Three platforms also run an agent for visits a user asks for: ChatGPT-User, Perplexity-User and Claude-User. OpenAI says robots.txt rules may not apply to those visits, and Perplexity says its agent generally ignores them. A firewall can block any of these crawlers even when robots.txt allows them, so we check both.24

What we found when we tested it

The documentation says how to be eligible. It does not say who gets cited. So on we asked ChatGPT, Gemini, Perplexity and Claude nine questions a UK buyer might ask, and noted every website each one cited.

Three questions were about business services, three about software and three about products. Between them the assistants cited 209 websites. Counting each website once for each question gives 234 cases.

Exhibit 3

234 cases, 199 cited by one assistant only
Cited byCases
One assistant only199
Two assistants20
Three assistants14
All four1

Source: AI Search Ranking research, . Nine questions, asked once of each assistant with web search on. Some answers were cut short by a length limit, so the counts are a minimum. Answers vary between runs.

That is 85% cited by a single assistant. Being a source for one tells you little about the others, which is why each has to be checked separately.

What gets cited depends on what is being bought

The same nine questions show how much the sources change from one sector to the next.

Exhibit 4

Share of citations by type of website, in three sectors
Type of websiteBusiness servicesSoftwareProducts
The supplier's or brand's own website47%40%21%
Review and comparison sites14%22%27%
Newspapers, magazines and trade publishers6%8%36%
Blogs by agencies, consultants and other vendors13%30%1%
Business directories13%0%0%
Retailers and marketplaces0%0%8%
Other6%1%6%

Source: AI Search Ranking research, . 98, 91 and 96 citations. Website types were assigned by us, and some are a judgement. Percentages are rounded, so a column may not add up to 100.

For business services, a supplier's own website carries most weight. For products, publishers and review sites do. The right work differs by sector, so we map the sources for yours before doing anything else.

Being named and being cited are different things

An assistant can recommend a company without citing its website. Across the nine questions the assistants named a brand 263 times. They cited that brand's own website in 56 of them.

Exhibit 5

How often a named brand's own website was cited
AssistantBrands namedOwn website cited
ChatGPT7715
Gemini5817
Perplexity5518
Claude736

Source: AI Search Ranking research, . Nine questions, asked once of each assistant with web search on. Answers vary between runs.

So we record the two separately for every prompt: whether you are mentioned, and whether you are cited.

Why we cite the platforms and not the commentary

We asked the four assistants two more questions: what OpenAI says about being recommended by ChatGPT, and what Google says about AI Overviews.

ChatGPT cited only OpenAI's and Google's own pages: 9 of its 9 sources. Claude cited none of them: 0 of 12. Gemini cited 6 of 25 and Perplexity 8 of 44.

So an answer about what a platform says may rest entirely on other people's articles. Every statement on this page about a platform links to that platform's own document.

What nobody outside these companies knows

No platform publishes how it chooses between two eligible sources. Nobody outside can say how much of a recommendation comes from a live search and how much from what the model learned in training.

Nor can anyone say why the same question gets a different answer an hour later. We do not know either.

What can be done is to remove what blocks you, give the assistants something accurate to cite, and measure the result on the same questions every month. That is the work we sell.

Where to go next

  • By platform

    ChatGPT

    What OpenAI says, what we measured in ChatGPT, and the work that follows from it.

  • By sector

    B2B companies

    How four assistants drew up supplier shortlists, and where they got them.

  • By sector

    Software companies

    Which products the assistants named, and why comparison pages were cited most.

  • By sector

    Online shops

    Which products the assistants named, and why reviews outweighed brand websites.

  • The work

    Our services

    An audit, technical foundations, answer sections and earned mentions, in one programme.

How we measure it

We agree a fixed set of 50 questions your buyers ask and run each one in Google's AI Overviews, ChatGPT, Gemini, Perplexity and Claude. Every prompt is run more than once, because AI answers change from one run to the next. We record whether you are mentioned, whether you are cited as a source, whether the description of you is accurate, and who is cited instead. We also read the generative AI performance report in your Search Console.1

Questions

How do I get cited in AI Overviews?

Make sure the page is indexed and eligible to be shown with a snippet, which is Google's one technical requirement. Then make it worth citing: helpful, reliable, written for people and offering a point of view of its own.1

Is optimising for AI Overviews different from SEO?

Not according to Google. It says its AI features are rooted in its core ranking and quality systems, so SEO best practices still apply. Optimizing for AI Overviews, as it is spelt in the US, needs no special file or markup.1

What is AI Mode?

AI Mode is one of Google's generative AI search features, named in Google's guide alongside AI Overviews. The same guidance covers both.1

How do I get cited by Perplexity?

Start by letting it read you. Perplexity asks site owners to allow PerplexityBot in robots.txt and to permit requests from its published IP ranges. In our test on it cited the most sources per answer, about 14.4

How do I get cited by Gemini?

We have found no separate Google guide for the Gemini app. Google does document Google-Extended, which controls whether your content is used to train Gemini models and to ground their answers. Blocking it does not affect Google Search.6

Do I need an llms.txt file?

No platform we checked asks for one. Google says no new machine-readable files or AI text files are needed for its AI features. It costs little to add, but we will not present it as a result.1

Should I block AI crawlers?

That is your decision, and the crawlers are separate. OpenAI and Anthropic each run a search crawler apart from their training crawler, so you can refuse training and still be shown in search. We show you what each rule does before you decide.25

Do I have to check every assistant?

Yes, if your buyers use more than one. In our test on , 85% of cited websites were cited by one assistant only for a given question.

Related guides

Sources

  1. Google Search Central. Optimizing your website for generative AI features on Google Search. Last updated . Opened .
  2. OpenAI. Overview of OpenAI crawlers. Opened .
  3. OpenAI Help Center. Publishers and Developers FAQ. Opened .
  4. Perplexity. Perplexity Crawlers. Opened .
  5. Anthropic. Does Anthropic crawl data from the web, and how can site owners block the crawler?. Opened .
  6. Google Search Central. List of Google's common crawlers. Opened .

Last updated

Find out how AI search describes you today

On a discovery call we look at how AI assistants describe your business now, and whether we can help.

Book a discovery call