Answers

How LLMs choose B2B sources

They do not publish a public ranking formula, and ChatGPT, Claude and Google's AI answers (AI Overviews and AI Mode) do not all work the same way. What we can say is that these systems retrieve sources and then generate an answer, and they tend to use pages that already answer the question clearly, with distinctive proof. Being indexed is not enough. On 5 Oct 2026, all 5 pages on simonhawtin.com were indexed and still earned zero citations in a signed-out check.

What Google has actually published

Google is the only one of those three that has spelled out the mechanics in public documentation. Its generative AI features use retrieval-augmented generation, often called grounding: they retrieve relevant, current pages from the Search index, then draw on those pages to generate a response and show links to the supporting sites. They may also use query fan-out, which means issuing several related searches across subtopics to build the answer. AI Overviews and AI Mode may use different models and techniques, so the links they show can differ, and AI Overviews often do not appear at all.

To be eligible as a supporting link, a page must be indexed and eligible to be shown in Google Search with a snippet. Google also says distinctive, useful, non-commodity content matters more than special AI markup, and that you do not need an llms.txt file or extra schema for these features. It is also explicit that meeting the requirements does not guarantee a page will be crawled, indexed or served.

What I have seen outside Google

ChatGPT and Claude do not publish an equivalent guide. I would not pretend to know their ranking weightings. What I can report from my own signed-out checks is more ordinary than a formula.

Indexed pages were ignored until the content was easier to use as an answer, with dated proof, and until the site was tied more clearly to me as an entity. On 14 Oct 2026, ChatGPT, Claude and Copilot named me only from LinkedIn, Finextra and scraped profiles, not from my site, and mixed me up with other people of the same name. ChatGPT, logged out, first cited simonhawtin.com on 26 Oct 2026, on 4 of 5 weekly prompts, then on 3 of 5 on 2 Nov and 9 Nov. Google's AI Overviews cited the site earlier, on 14 Oct, for branded and topic queries such as demand gen and ABM leadership, but did not cite it on the weekly prompt set I could measure on 26 Oct and 9 Nov. Copilot has not cited those weekly prompts in the weeks I could measure. I have not repeated a weekly Claude check since 14 Oct, so I will not claim later Claude results I do not have. The engines move at different speeds, and they often draw on third-party pages they already trust, which is why mentions on those pages matter as well as your own site.

What this means for a B2B or HR tech query

When a People leader asks ChatGPT to shortlist HRIS platforms, or a talent acquisition lead asks Claude to compare applicant tracking systems, the model is not browsing your product pages in order. It is assembling an answer from sources it can retrieve and is willing to trust for that question. A features page that never answers the question is easy to skip. A page that answers it first, with proof a buyer can check, is easier to use. So is a mention on a review site or listicle the model already cites.

That is why Agentic AEO, run on Grok Bot, starts with the buyer prompts, measures signed out, and then works on both your pages and the third-party sources already appearing in the answers. The case study has the dated results. If you want this done on your own prompt set, book a free discovery call. The AI search audit, if it's a fit, is scoped on that call.

Sources

  1. Google Search Central, Optimizing your website for generative AI features on Google Search (RAG / grounding, query fan-out, non-commodity content): https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
  2. Google Search Central, AI features and your website (eligibility, query fan-out, AI Overviews vs AI Mode, no special markup, indexing not guaranteed): https://developers.google.com/search/docs/appearance/ai-features
  3. Simon Hawtin, How I built an agentic AEO system for my own site, plus weekly notes for 26 Oct, 2 Nov and 9 Nov 2026: https://simonhawtin.com/case-studies/agentic-aeo

FAQ

01

Do AI assistants have a public ranking algorithm for citations?

Not one that is published in full. Google has described retrieval from its Search index, grounding, and query fan-out for AI Overviews and AI Mode. ChatGPT and Claude have not published an equivalent ranking guide.

02

Is being indexed enough to be cited?

No. Google says eligibility does not guarantee a page will be served. On my own site, all 5 pages were indexed on 5 Oct 2026 and still earned zero citations in a signed-out check.

03

Why do competitors appear in an answer when we do not?

Often because the model is already using their site, or a third-party page that mentions them, as a source for that question.

04

Do ChatGPT, Claude and Google's AI answers choose the same sources?

Not reliably. Google notes that AI Overviews and AI Mode may use different models and techniques. My weekly checks also show engines moving at different speeds: Google cited my site on some topic queries from 14 Oct 2026, ChatGPT from 26 Oct, and Copilot not at all on the weekly prompts I could measure.

Related questions

Free discovery call

Start with a free discovery call

A free 30-minute call. Together, on the call, we look at whether buyers in your category use AI to shortlist, whether you show up, and whether you can act on the fixes. If it isn't worth doing, I'll tell you. The £595 audit and the 3-month service both start here.

  • Where you're named and cited, and who's named instead
  • The pages AI engines already rely on
  • What to fix first