Claude AI GEO: How Claude Finds Its Sources, How To Get Cited, And How Claude's Models Score

Claude AI GEO is the work of getting your brand named and cited inside the answers Claude, Anthropic's AI assistant, gives when someone asks a question in your niche.

GEO stands for generative engine optimisation, and this version is pointed at one assistant that shows its sources whenever it searches the web.

The problem it solves is simple, because AI answers name only a few businesses and everyone else gets skipped.

If you'd like my team to look at your chances first, you can book a free one-to-one GEO, AEO AI audit and get a custom game plan.

In this post I'll explain how Claude finds what it cites, what you can control and how to measure it.

Near the end, I'll show how Anthropic's Claude models score on this leaderboard, and why that's a separate question from being cited.

Claude AI GEO at a glance

This table lists what you control and what Anthropic's own documentation says about each part.

Nobody can guarantee what Claude says, but every row above is something you can actually change.

What is Claude AI GEO?

Claude is the AI assistant built by Anthropic.

Claude AI GEO means making your brand the one it names when a buyer asks who's best, what to buy or who the expert is.

It's different from using Claude as a tool to do SEO, which I cover in my Claude GEO SEO post.

Old search gave you ten blue links, and position seven still got some clicks.

AI answers usually give a short list of maybe one to three names, so there's no room for number seven.

How does Claude find the sources it cites?

Claude can answer from what it learned in training or from a live web search.

Anthropic's web search help article says Claude uses a search tool when a question benefits from current information.

It says every web search response includes citations, so you can verify the sources yourself.

You turn web search on from the "+" menu in a chat, and adding "Search the web" to a prompt helps make sure it searches.

Anthropic's page on reporting, blocking and removing content says its web search uses search partners that index web content.

That means your page has to be indexed and reachable before Claude's live search can find it.

Anthropic's three bots, and which ones matter for citations

Anthropic's help article on how site owners can control its crawler lists three bots.

Anthropic says robots.txt rules need adding for every subdomain you want to control.

It says blocking its IP addresses may not work reliably, because that stops its bots reading your robots.txt.

It also supports the Crawl-delay extension if you want to slow crawling down.

If you want to be cited, the practical rule is to leave Claude-User and Claude-SearchBot unblocked.

ClaudeBot is a separate decision about training data, and it's yours to make.

Noindex and gated pages

Anthropic says the noindex robots meta tag tells its partners not to index your content, so it won't appear in Claude outputs that use web search.

It also says password-protected content stays out of web-search-based outputs.

A stray noindex on a key service page is one of the easiest mistakes to make and one of the easiest to fix.

📞 Want to know whether Claude can even reach your pages?

Book a free AEO SEO strategy session and my team will look at your site, your competitors and the keywords you're closest to winning.

→ Get your custom game plan for Google and AI search

Be the source: case-study pages

My first GEO move is to be the source, which means every page carries a real case study, test or result.

The AI can't write about a test only you ran, so your page becomes the one it has to cite.

In my GEO examples video, Google's AI Overview for "best geo expert" names me first and describes me as known for documenting GEO experiments and public testing.

That example is Google rather than Claude, but it shows how a description gets built from pages that document real work.

Be everywhere: consistent mentions on other sites

My second move is to be everywhere, because AI answers lean on what several sources agree about.

I explain it like five friends recommending the same restaurant, because if five good pages say the same name, the AI repeats it.

I run multiple websites with different angles pointing to the same answer, and my subreddit, AI SEO Insider, ranks on page one too.

For Claude AI GEO, that means your name, services and specialism should read the same on your site and every site that mentions you.

Be first: publish on new topics fast

My third move is to be first, which means publishing within 24 hours of a topic appearing.

When Jev AI dropped, I published on "Jev architecture" and "Jev waitlist" that same weekend.

When Claude searches for something brand new, an early page is often one of the few answers it can find.

Google's 1 October 2026 update to its page on using generative AI content says it's "critical to manually factcheck and review all AI-generated content".

Speed still needs a human check, because a fast page with wrong facts just spreads wrong facts.

Answer-first structure

Put the direct answer to each buyer question in the first two lines of the page.

Add an FAQ where each answer lands in the first sentence and stays short.

Anthropic doesn't document any special file or markup that gets a page cited, so I treat structure as good practice rather than a switch.

How to measure Claude AI GEO

Write 20 to 30 questions a real buyer would ask Claude about your niche.

Ask each one in a fresh chat with web search on, and log whether you're named, which pages are cited and whether the description is accurate.

Repeat the same questions every week and compare the logs.

Answers can vary between chats, so judge the trend over several weeks rather than one reply.

My Claude SEO agency guide describes the same idea, which is measuring by re-asking the AIs every week.

What my own screens show, and what they don't

My on-screen proof comes from Google's AI Overviews, Google's AI Mode and Perplexity.

When I ask AI Mode for the best GEO experts to follow in 2026, it lists me first and links to my sites, and Perplexity also lists me among GEO experts.

One of my sites went from zero to more than 1,300 Google clicks a day.

None of those screens is a Claude answer, so I'm not claiming Claude names me.

I share them because the same groundwork gives any AI assistant something real to find.

A first-month Claude AI GEO plan

This is the order I'd follow if I were starting from zero, because it fixes blockers before it spends time on content.

After the first month, the work becomes a weekly loop of one new proof page, one correction and one re-run of the questions.

I'd also add a new page within 24 hours whenever a fresh topic appears in your niche.

What to do if Claude cites something you don't want

Sometimes the problem isn't being missing, because Claude cites an old or private page instead.

Anthropic's page on reporting, blocking and removing content says removing the content from your site is the best way to keep it out of Claude's outputs.

It also lists password protection, a noindex tag and robots.txt rules for media files as options.

For a URL that already appears in Claude's outputs, it says you can email Anthropic with proof that you own the domain.

If the problem is a wrong fact on someone else's site, the fix is to get it corrected at the source, because that's where the AI found it.

Limits of Claude AI GEO

Nobody can guarantee what Claude says, and anyone who promises it is overselling.

Anthropic doesn't publish ranking factors, so each change is a test.

Training-based answers change slowly, because they depend on future model releases.

Allowing the bots removes a blocker, but it doesn't promise a citation.

📞 Ready to see which questions Claude should name you for?

Claim a free one-to-one GEO, AEO AI audit and leave with a plan to rank in Google and AI search.

→ Book the free call

Doing it yourself with the AI Profit Boardroom

Inside the AI Profit Boardroom you get my Agent OS, my SEO skill and checklist and the AI SEO automation section.

Members also get four coaching calls a week and a member map, alongside 3,400+ members.

Which Claude model is answering, and how does it score?

Claude AI GEO doesn't ship its own model, because it's about the answers Anthropic's models give.

Anthropic chooses which models are available in the Claude app, so this board can't tell you which model answered any particular chat.

Here's how the Claude models on GoldieBench score on their live pages today.

Claude Opus 5 has the highest average of these at 8.27, and Claude Sonnet 5 has the lowest at 7.01.

Why model scores and Claude citations are different questions

GoldieBench scores one-shot builds, such as apps and games made from a single prompt.

It doesn't measure which businesses a model names or which pages it cites.

A stronger model doesn't automatically mention your business more, because a search-based answer depends on the pages it can find and what those pages say.

That's why access, proof and agreement matter more for Claude AI GEO than which model is behind the answer.

Where model quality does matter is your own GEO work, such as drafting case-study pages or comparing weekly logs.

You can browse every model on the models page.

Related reading on GoldieBench

My GEO examples post shows four real AI Overview wins from my own screens.

My best Claude SEO agency post compares agencies that sell Claude visibility.

In the same batch, my Claude GEO SEO post covers using Claude as the engine for your SEO work.

If you'd like expert eyes on your plan, book a free strategy session and see the questions you're closest to winning.

Also On Our Network

• Agent Operatorsthe eight-step walkthrough for getting cited by Claude
• AI Income Deskhow businesses get named by Claude
• AI Tool Verdictevery Claude AI GEO tactic rated honestly
• agentos.guidea weekly Claude citation tracker for your Agent OS
• aiprofitboardroom.commy Claude SEO skill explained

Open the door for Anthropic's search bots, build real proof and track the answers weekly, and you'll be doing Claude AI GEO properly.