How AI engines choose which brands to recommend

When someone asks an AI assistant "what's the best accounting software for a small agency?", the answer usually names three to five products. Nobody outside the AI companies can see the full logic behind that list. But the companies do document the parts that matter to you: which crawlers they use, when they search the web, and how they show sources. Independent research fills in some of the rest. This post puts the documented pieces together.

Two ways an AI engine knows about your brand

1. Training data

Large language models are trained on large collections of text, including public web pages. If your brand was discussed widely before a model's training cut-off, the model may name you even without searching. You cannot change this quickly: it only updates when a new model is trained.

Training crawlers are controlled separately from search crawlers. OpenAI says GPTBot crawls content that may be used in training its foundation models. Google's equivalent control is Google-Extended, a robots.txt token that manages whether content can be used to train future Gemini models and for grounding in Gemini apps; Google says it does not affect inclusion or ranking in Google Search.

2. Live retrieval at answer time

For questions that need current information, which includes most buying questions, the engines search the web and answer from what they find. This is where most of the movement happens, because it reflects the web as it is today.

• ChatGPT: OpenAI says OAI-SearchBot is used to surface websites in ChatGPT's search features, and ChatGPT-User may visit a page when a user's question needs it.
• Perplexity: PerplexityBot is designed to surface and link websites in Perplexity's results and, according to Perplexity, is not used to crawl content for AI foundation models.
• Gemini: with grounding with Google Search, the model decides whether a search would help, and if so automatically generates one or more search queries, runs them and builds the answer from the results.

What happens when an engine searches

The documented flow is broadly the same across engines: the question becomes searches, the searches return pages, and the model writes an answer from those pages and points to some of them.

1. The question is turned into queries. Google says AI Overviews and AI Mode may use a "query fan-out" technique, issuing multiple related searches across subtopics and data sources. A question about "CRM for a small agency" can trigger searches about pricing, integrations and alternatives.
2. Pages are retrieved. To be shown as a supporting link in AI Overviews or AI Mode, Google says a page must be indexed and eligible to appear in Search with a snippet. Pages a crawler cannot reach cannot be used.
3. The answer is written and sourced. Perplexity says each answer includes numbered citations linking to the original sources. Gemini's grounding returns citations that link spans of the answer text to source URLs.

The practical consequence: an engine recommends brands that appear, and are described clearly, on the pages its searches return. If the top results for "best X for Y" are three comparison articles that all list the same five tools, those five tools are the likely answer.

Which kinds of sources get cited

Citation patterns differ by engine and by category, so the only reliable answer is to look at the sources cited for your own prompts. There is good public data for Google, though. Pew Research Center analysed 68,879 Google searches made by 900 US adults in March 2025. It found that Wikipedia, YouTube and Reddit were the most frequently cited sources in Google's AI summaries, together accounting for 15% of the sources listed. Government sites made up 6% of links in AI summaries versus 2% in standard results, and 88% of summaries cited three or more sources.

Two lessons follow. First, answers usually draw on several sources, so being on one page is rarely enough. Second, community and reference sites carry weight, which means what other people say about you can matter as much as what you say about yourself.

What correlates with a brand being mentioned

The largest public study of this is Ahrefs' analysis of 75,000 brands in Google's AI Overviews, published in May 2025. Using Spearman correlation, it found that brand web mentions showed the strongest correlation (0.664) with AI Overview brand visibility. The top three factors were all off-site:

Ahrefs stresses that correlation is not causation, and the study covers AI Overviews, not ChatGPT or Perplexity. Still, the direction is consistent with how retrieval works: a brand that many pages discuss has more chances to be in the pages an engine reads.

What makes a page more likely to be used in an answer

The research that gave Generative Engine Optimization its name tested content changes directly. In the GEO paper (Aggarwal et al., accepted to KDD 2024), researchers from Princeton University, IIT Delhi and independent collaborators rewrote source pages in nine ways and measured how much each page was used in generated answers. The best methods, Cite Sources, Quotation Addition and Statistics Addition, achieved a relative improvement of 30-40% on their main visibility metric. Keyword stuffing, the classic SEO tactic, did not perform well. Tested on Perplexity itself, the methods improved visibility by up to 37%.

This was a controlled experiment on a benchmark, not a guarantee for your page. But it supports a simple rule: pages that state facts plainly and back them up are easier for a model to use than pages full of claims.

Why the recommended brands change between answers

Even with the same inputs, the output varies. SparkToro and Gumshoe found that ChatGPT and Google's AI gave the same list of brands in two responses less than once in 100 runs, and the same order roughly once in 1,000. Categories with few options, such as cloud providers, were more consistent than broad ones such as recent science fiction novels.

So "how do AI engines choose" has a probabilistic answer. They do not pick a fixed list; they make some brands much more likely than others. Your goal is to be in that likely group, and the right way to measure it is the share of many answers that mention you. Our post on checking whether ChatGPT recommends your brand covers the method.

What this means for your brand

1. Let the search crawlers in. Allow OAI-SearchBot and PerplexityBot, and keep your pages indexable in Google. Decide about training crawlers (GPTBot, Google-Extended) separately.
2. Find the pages engines retrieve for your category. Run your buyer questions, collect the cited URLs and group them by domain.
3. Get onto those pages with accurate information: comparison articles, directories, review sites and community threads where your buyers ask questions.
4. Earn mentions beyond your own site. The strongest correlation in the Ahrefs data was with web mentions, linked or not.
5. Make your own pages easy to quote: what you are, who it is for, real pricing and limits, and sourced facts.
6. Measure the share of answers that mention you over time, not single responses.

Google says there are no special optimizations required to appear in AI Overviews or AI Mode, and no special schema.org markup or AI text files are needed. Standard SEO fundamentals apply. Treat any claim of a secret AI ranking trick with caution.

Frequently asked questions

Do AI engines favour brands that advertise?

In Ahrefs' AI Overviews data, branded ad traffic (0.216) and ad cost (0.215) had only weak positive correlations with being mentioned. Paid search does not appear to buy organic AI recommendations.

If I block GPTBot, will ChatGPT stop recommending me?

Not for search answers. OpenAI documents GPTBot (training) and OAI-SearchBot (search) as independent settings, so a site can allow OAI-SearchBot to appear in search results while disallowing GPTBot.

Does Google use a different system for AI Overviews?

Pages must be indexed and snippet-eligible in Google Search to be shown as supporting links, and Google says the best practices for SEO remain relevant for AI features. Google also notes that AI Mode and AI Overviews may use different models and techniques, so the responses and links they show can differ.