Getting Cited in Claude

A year ago, Claude was a rounding error in AI search. Today a buyer asks it which tool to use in your category, and it names two or three brands as the answer. If your website is an island, the older job was making it easier to find once a buyer had already set sail toward you. Claude asks a colder question. When the buyer never opens a search results page at all, does Claude name your island, or sail past it to a competitor it trusts more?

The short version

Here is why this stopped being optional. Across measurable B2B AI referrals, Claude climbed from a 1.4% sliver to 18.5% in a single year, second only to ChatGPT and ahead of Gemini, Perplexity, and Copilot. “I’ll do ChatGPT first and worry about the rest later” used to be a defensible plan. With Claude now carrying close to a fifth of the traffic, skipping it means handing that share to whoever did show up. This page is part of our broader AI SEO guide, which maps how the same work pays off across every engine. Claude is its own animal, though, and the way it picks sources changes what you put first.

18.5%

of measurable B2B AI referrals now go to Claude

Up from 1.4% a year ago, second only to ChatGPT. Goodie, 2026

84%

of AI citations come from earned media

Not your own pages. Across 25M+ links spanning Claude, ChatGPT, and Gemini. Muck Rack, 2026

3–6

domains cited per AI answer, versus ~10 on Google

Fewer named slots, worth more. Digital Applied, Q2 2026

Claude does not crawl and rank the whole web the way Google does. When web search is on, it sends your query to a search provider, Anthropic uses Brave Search for Claude’s web results, reads the documents that come back, and writes an answer with inline citations to the pages it actually leaned on. Its own crawler, ClaudeBot, gathers content that may inform those answers and future training.

Two consequences fall out of that, and the second one is the whole game. First, the sources Claude can reach are bounded by what Brave surfaces and what ClaudeBot is allowed to fetch, so a blocked crawler quietly removes you from the running before the contest even starts. Second, Claude reads less like a ranking machine and more like a careful editor: it quotes the passage that answers the question most directly, not the page that happens to sit first in the results. Bury your answer in paragraph nineteen and you are asking it to do work it has no reason to do. Lead with the answer instead.

SIGNAL 1

Corroboration across independent sites

Claude trusts a claim far more when several independent sites say the same thing than when you say it once on your own page. Call yourself the leading vendor on your homepage and it counts for almost nothing. Have a roundup, a news mention, and a comparison article say it for you, and it reads as consensus.

SIGNAL 2

Passage-level structure

Lead each section with the answer, then explain. Claude lifts the sentence that resolves the question, so a paragraph that opens with “Claude SEO is…” is far easier to quote than one that builds toward its point three sentences later.

SIGNAL 3

Consistent, current facts

If your pricing, founding date, or category differs between your site, your LinkedIn, and a third-party directory, Claude has no clean fact to repeat and may skip you rather than guess. Pick one set of facts and keep them identical everywhere it can read.

This is the signal most brands have no plan for, and it is the one that decides Claude answers. The numbers back it hard: across recent analysis of 25 million-plus links spanning Claude, ChatGPT, and Gemini, roughly 84% of AI citations point at earned media rather than a brand’s own pages. A reference you write for yourself goes straight in the bin. The same claim, signed by three sources who owe you nothing, gets you hired. Claude reads it the same way.

The discipline that earns a Claude citation also makes the page easier to read, which is a rare case of the engine and the human wanting the same thing. Our answer engine optimization strategy guide walks through the structure move by move.

When web search is off, Claude answers from training data alone, and that is a door you cannot pick on a short timeline. Nobody outside Anthropic knows the exact mix that decides what a model remembers. What you can do is become the kind of widely-referenced source that lands in the next training set, cited and quoted and mentioned across the open web until the pattern is too consistent to ignore.

Trackers with Claude coverage

Profound is the enterprise option, monitoring answers across multiple assistants, Claude among them, and flagging which sources get cited for your prompts. Peec AI is the lighter-weight pick for a single brand without an enterprise contract.

The honest baseline: manual prompts

Run your real buyer questions through Claude with web search on, log which sources it cites, and repeat monthly. Free, slow, and uncomfortably revealing about where your gaps actually are. It tells you exactly where to earn your next mention.