Blog

How AI search engines pick their sources, and how to become one

Why one website shows up in ChatGPT, Perplexity or Google AI Mode and another doesn't, and what you can do about it.

Key takeaways

  • AI search engines write their answers from pages they first retrieve from a search index (RAG); if you're not in that index, you can't be cited.
  • Google AI Overviews and AI Mode use the Google index, Copilot runs on Bing, and ChatGPT and Perplexity operate their own crawlers.
  • Because of query fan-out, only 37.9% of URLs cited in AI Overviews also ranked in the top 10 in an Ahrefs study from March 2026.
  • Brand mentions across the web correlate more strongly with visibility in AI Overviews than backlinks do (Ahrefs, May 2025).
  • Google's own position: optimising for generative AI in Search is still SEO, with no special tricks or files required.

AI search engines such as ChatGPT search, Perplexity, Google AI Mode and Microsoft Copilot choose their sources by searching an index first, reading a handful of pages and writing an answer from them, with links to some of those pages. AI search optimisation means getting into those indexes and answering the question better than the alternatives.

This article is for marketers and business owners who want to understand why a competitor shows up in AI answers and they don't. The topic keeps growing: Google's AI Mode has been rolling out country by country (in the Netherlands, where I'm based, it launched on 8 October 2025). I explain how source selection works, what differs per platform and what you can actually do. The technical layer (crawlers, robots.txt, schema markup and llms.txt) is covered separately in llms.txt, schema markup and AI crawlers.

How does an AI search engine choose its sources?

Almost always through retrieval-augmented generation (RAG): the language model first retrieves current pages from a search index and grounds its answer in them. The model doesn't decide out of thin air which site is good; it works with what the search system hands it.

Google describes this process for AI Overviews and AI Mode fairly openly in its guide to generative AI features in Search (May 2026). Answers are ‘grounded’ in pages from the Search index and run on the same core systems as the regular results. Other AI search engines work in broadly the same way, although the index they draw from differs. Simplified, it looks like this:

  1. Understand the question. The model works out what the user actually wants to know, including context from earlier turns in the conversation.
  2. Query fan-out. The system splits the question into several related searches.
  3. Retrieve and rank. For each search, pages are pulled from the index and assessed for relevance and quality.
  4. Write and cite. The model writes a single answer and attaches sources to the passages they support.

On top of that, there's the knowledge a model picked up during training. That determines what ChatGPT or Gemini ‘knows’ about your brand when it doesn't search live. Training knowledge only changes with a new model version; the sources in a search-backed answer change as soon as the index does. For AI search optimisation, that second track is the one you can influence most.

What is query fan-out, and why are cited pages often outside the top 10?

Query fan-out means an AI search engine breaks one question into several searches and looks for sources for each of them. A page can therefore be cited for a sub-question without ranking for the main query.

Google's own example is ‘how to fix a lawn that's full of weeds’: the system also searches for things like ‘best herbicides for lawns’. A page about weedkillers can end up in the answer without ranking anywhere for the original question.

You can see the effect in the data. In an Ahrefs study (March 2026) of 863,000 search results and 4 million cited URLs, 37.9% of the pages cited in AI Overviews also ranked in the top 10. 31.2% ranked in positions 11-100 and 31.0% outside the top 100. An earlier Ahrefs study (July 2025) had found around 76% coming from the top 10. Ahrefs points to query fan-out as the explanation.

What does that mean for you? A strong page that also answers the logical follow-up questions has more chances than a page that only targets the head term. But don't create a separate page for every conceivable sub-question: Google warns that mass-producing pages for query variations or fan-out queries can count as scaled content abuse, which is a form of spam.

Where do ChatGPT, Perplexity, Gemini and Copilot get their sources?

Each AI search engine relies on a different index. That's the most important difference between the platforms, and it tells you where to focus.

PlatformWhere the sources come fromWhat to focus on
Google AI Overviews and AI ModeThe Google Search index, via Search's core systems; AI Mode runs on a custom version of GeminiRegular SEO: indexing, allowing snippets, strong content. Merchant Center and Google Business Profile for products and local businesses
Gemini (app)Google's systems; whether your content may be used for grounding is controlled with the Google-Extended tokenThe same foundations as Google Search, plus a deliberate decision on Google-Extended
ChatGPT searchOpenAI's own crawler OAI-SearchBot, supplemented by results from external search partners; ChatGPT-User fetches pages live when a user asksAllow OAI-SearchBot, be well indexed in both Google and Bing, give clear answers and build a recognisable brand
PerplexityIts own index via PerplexityBot; Perplexity-User for live requestsAllow PerplexityBot, publish current, well-sourced pages
Microsoft CopilotThe Bing index (Bingbot)Set up Bing Webmaster Tools, submit your sitemap, check indexing in Bing

The common thread: there's no separate ‘AI index’ you can sign up for. You get into AI answers through the same search indexes as always, plus a few dedicated crawlers. Google says so explicitly in its documentation on AI features and your website: there are no additional requirements to appear in AI Overviews or AI Mode. Your page needs to be indexed and eligible to be shown with a snippet. If you want a refresher on how an index works, read how search engines work.

When I migrated seomotion.nl to a static HTML site in October 2026, I set up both the Search Console domain property and Bing Webmaster Tools right at launch, with an XML sitemap submitted to both. Bing powers Copilot. If you only look at Google, you miss part of the AI landscape, and Bing Webmaster Tools takes about fifteen minutes to set up.

What's the difference between a citation and a mention?

A citation is a link to your page as a source for the answer. A mention is your brand name in the answer text, with or without a link. Both count towards AI visibility, but they come about in different ways.

You earn citations with your own pages: they need to be found and give the best answer to (part of) the question. Mentions mainly come from what others write about you: trade media, comparison sites, reviews, forums and roundups. A language model names the brand it repeatedly encounters in the same context across trustworthy sources.

That shows up in an Ahrefs analysis of 75,000 brands (May 2025). The strongest correlations with being mentioned in AI Overviews were brand mentions across the web (Spearman 0.664), branded anchor text (0.527) and branded search volume (0.392). Domain Rating (0.326) and the number of backlinks (0.218) scored lower. Bear in mind this is correlation, not proof of cause and effect.

Click behaviour also explains why mentions matter. According to Pew Research Center (July 2025), only about 1% of visits involved a click on a link inside an AI summary. A citation brings few direct clicks; being named in the answer is what makes someone look you up later. Google is clear, though: don't chase artificial mentions. You build authority with genuine publicity and solid link building and digital PR.

What makes a page citable for AI search engines?

A page gets cited when it answers part of the question better, more specifically or more reliably than the alternatives. Google calls this ‘non-commodity’ content: information based on first-hand experience or expertise that you won't find in the same words on ten other sites.

Research points the same way. In the GEO paper by Aggarwal et al. (Princeton and others, 2023, presented at KDD 2024), methods such as citing sources and adding quotations and statistics increased visibility in generative answers by up to around 40% in their experiments. Keyword stuffing didn't work. It's a research setup with a test environment, not a guarantee for ChatGPT or Google, but the direction makes sense: concrete, well-supported information is easier to cite.

Just as important is what you don't need to do. Google lists several myths in its guide: splitting content into small pieces (‘chunking’) isn't necessary, there's no ideal page length, and you don't need to rewrite content specifically for AI. Write for people, with a clear structure.

  • Every section opens with a direct answer in 1-2 sentences, followed by the explanation.
  • Headings follow the real questions your audience asks, not just your keyword list.
  • Figures come with a source and date, so an AI system can verify them.
  • The page contains something competitors don't have: your own experience, a comparison, a worked example, a clear point of view.
  • It's obvious who the author is and why they know the subject (see also E-E-A-T).
  • Where it fits, you use relevant images or video: Google mentions this explicitly, and YouTube is the most cited domain in AI Overviews (Ahrefs, March 2026).

If you want to tackle this systematically, it's essentially good content SEO, with extra attention to follow-up questions and evidence.

What can you do on each platform?

Start with technical access and indexing, then work on your best pages, and then on your reputation beyond your own site. This is the order I use myself.

  1. Check access. Look at your robots.txt, CDN and firewall to confirm that Googlebot, Bingbot, OAI-SearchBot and PerplexityBot can read your site. How to configure this is covered in the technical article I mentioned above.
  2. Get indexed in both Google and Bing. Set up Google Search Console and Bing Webmaster Tools and submit your sitemap. Copilot runs on Bing, so without Bing indexing you don't exist there.
  3. Build a question list. Write down 20-30 questions your customers genuinely ask, from early research to ‘which specialist should I choose for X?’.
  4. Test those questions per platform. Ask them in ChatGPT, Perplexity, Google AI Mode and Copilot. Record which sources and brands are named, and whether you're among them.
  5. Improve or create the best page per question cluster. One strong page that answers the main question and the logical follow-ups, not ten thin variations.
  6. Work on mentions beyond your site. Think trade media, industry directories, reviews, your Google Business Profile and guest contributions where you genuinely have something to say.
  7. Measure and repeat. Rerun the test every quarter and check your analytics for referral traffic from AI services and, more importantly, the conversions it brings.

How do you measure visibility in AI search engines?

For Google, there's been an official source since 3 June 2026: the generative AI performance reports in Search Console, showing impressions, pages, countries and devices for AI Overviews and AI Mode, among other things. They're being rolled out gradually, so you may not have them yet. Traffic from AI features is also included in the standard Performance report under the Web search type.

For ChatGPT, Perplexity and Copilot, you rely on your own analytics (referral traffic) and on manual or automated tests with a fixed set of questions. Tools that automate this are useful for spotting trends, but treat the results as a sample. Google puts it bluntly: ‘No third-party tool has access to our internal ranking or AI systems.’

Finally, look beyond clicks. Google says clicks from AI Overviews are ‘higher quality’ (Search Central Blog, May 2025) and recommends measuring conversions. Fewer visitors with clearer purchase intent can work out fine overall. For more on AI Overviews and AI Mode themselves, read my article on Google AI Mode.

My advice: where to start with AI search optimisation

Don't start with tricks. Start by checking whether AI search engines can read your site at all and whether you're properly indexed in both Google and Bing. Google sums it up itself: optimising for generative AI in Search ‘is still SEO’. That holds for the other platforms too, with two points of emphasis: answer follow-up questions better than your competitors, and make sure others write about you.

Want to know where you stand? Read what GEO is if you're after the terminology, or see how I help businesses with AI SEO and GEO: from a baseline per platform to a concrete list of improvements. Nobody can guarantee rankings or citations, me included. What I can do is show you where the opportunities are.

Sources

  1. Google's Guide to Optimizing for Generative AI Features on Google Search (Google Search Central, May 2026)
  2. AI features and your website (Google Search Central)
  3. AI Overview citations and the top 10 (Ahrefs, March 2026)
  4. Brand factors and AI Overview visibility (Ahrefs, May 2025)
  5. Google users are less likely to click on links when an AI summary appears (Pew Research Center, July 2025)
  6. GEO: Generative Engine Optimization (Aggarwal et al., arXiv 2023 / KDD 2024)
  7. Generative AI performance reports in Search Console (Google Search Central Blog, June 2026)

Frequently asked questions

Is AI search optimisation the same as GEO?

In practice, AI SEO, GEO (Generative Engine Optimisation) and AEO (Answer Engine Optimisation) are used interchangeably for the same goal: being visible in AI answers. GEO comes from a 2023 research paper, while AI SEO is mostly the term marketers search for. The underlying work is largely the same.

Can I pay to appear in ChatGPT answers?

You can't buy organic citations or mentions in ChatGPT. Advertising is a separate system: OpenAI even runs a dedicated crawler for it (OAI-AdsBot) that checks ad landing pages. Anyone promising otherwise is usually selling an indirect route, such as placements on sites that are cited often.

How quickly will I see results?

It depends on the track. Search-backed answers draw on a live index, so an improved page can appear once it has been recrawled and reindexed. What a model ‘knows’ about your brand from training only changes with a new model version. Think in months rather than days, and without guarantees.

Does ranking well on Google help me appear in ChatGPT or Perplexity?

Indirectly, yes. Those platforms use different indexes, but the same qualities that make you rank on Google, such as good content, crawlability and authority, help there as well. A high Google ranking is neither a requirement nor a guarantee for being cited by other AI search engines.

Should I write differently for AI search engines than for Google?

No. Google explicitly says you don't need to rewrite or chunk content for AI. What helps for both: give the answer first in each section, name your sources and add something only you can offer, such as first-hand experience or a comparison.

Curious what SEO and GEO could do for your business?

Tell me where your business stands today and where you want to take it. In a short, no-obligation call we will work out which steps will make the biggest difference.

Get in touch →