Claude AI Optimisation

Claude answers most questions from what it already knows, and reaches for its web search tool only when a question needs something current, specific, or outside its training — a live price, a recent event, a fact it should verify rather than guess at. When it does search, it cites what it used, visibly, in the answer. That citation is the prize: not a ranking position, but a named, linked mention inside an answer someone is reading to make a decision. This page is about earning that mention.

How Claude actually picks a source

Claude’s web search is deliberate rather than constant — it decides a question warrants a live check, runs a small number of targeted searches, and then cites the specific pages it drew on directly in its response, usually with a visible link. That citation behaviour is more transparent than some competitors: Claude generally shows you where an answer came from rather than presenting synthesised text with no attribution, which means the sources it does cite tend to be ones with a demonstrable claim to authority on the topic — an original source, a primary account, or a page that states a fact plainly enough to quote with confidence.

Anthropic has been explicit that Claude is built to weigh source quality rather than just source popularity: a page that makes a specific, checkable claim, is clearly authored, and doesn’t contradict itself is easier for Claude to cite responsibly than a page of vague reassurance. In practice that means the same fundamentals that help elsewhere — clarity, first-hand data, a real named author — carry unusual weight with Claude specifically, because its search behaviour is oriented around defensible citation rather than volume of retrieval.

What’s included

Authoritative-source positioning. We identify what would make your page the checkable, citable original for a given claim in your category — first-hand data, a primary account, a stated methodology — because Claude’s search behaviour favours pages it can point to with confidence over pages that merely repeat consensus.

Claude-specific citation tracking. A fixed prompt set covering your buying journey, re-run against Claude’s web search specifically, recording whether it searches at all for a given prompt (it doesn’t always), whether you’re cited, and which competing sources it reaches for instead.

Author and claim clarity. Named authors with checkable credentials, specific claims instead of vague reassurance, and facts stated once, clearly, rather than scattered and softened across a page — since Claude’s citation behaviour rewards pages it can quote without hedging.

Answer-shaped content blocks and entity work, applied at the depth Claude’s search-then-cite pattern rewards: a direct, checkable answer near the top of the page, and a coherent entity presence that lets Claude attribute a claim to a real organisation with confidence.

What we don’t sell: an llms.txt file as a deliverable. Claude does not require or specially favour one for citation purposes — the file signals crawl preferences and access permissions to certain automated systems, but it doesn’t make a page more citable, and treating it as a GEO product overstates what it does.

Proof

Official London Theatre reached 85% AI search visibility and the #1 recommended source position in its category, tracked across a five-year partnership, alongside a 67% increase in referral traffic. Claude is tracked as one channel inside that combined figure, not isolated as a separate published number.

Koyo Loans took a #1-3 ranking for “loans” (95% keyword difficulty) with a 73% lift in organic conversions. We haven’t isolated what share of that traffic or citation activity is attributable to Claude specifically — Claude’s search volume and behaviour are newer and less measured than Google’s or ChatGPT’s, and we say so rather than inflate the claim.

Sons is a client we do GEO work for. We’re not publishing prompt-grid or before/after figures for that account until the numbers are sourced and cleared — named here without figures, per house rule.

How we measure it

Search-trigger rate. Of your tracked prompt set, how often Claude’s web search activates at all for the question — since Claude answers many questions from training without searching, this is a precondition worth knowing.

Citation share when it does search. Of the prompts where Claude searches, how often you’re one of the cited sources.

Claim defensibility audit. A manual review of whether your key commercial pages make specific, checkable claims Claude could plausibly cite, versus vague copy it would have no reason to quote.

Run a first check for free with our GEO visibility score — no sign-up.

What it costs

There’s no separate Claude invoice. This work is included in every GEO and SEO plan at no extra cost — retainers start from £1,250 a month ex VAT. Standalone Claude-only GEO work with no retainer attached is scoped to the brief, and either side can end it with 60 days’ notice at any point. Tell us what you need and we’ll price it before we start.

FAQs (8, visible)

Does Claude search the web for every question?

No. Claude decides on a case-by-case basis whether a question needs current or specific information it doesn’t already have confidently, then runs a small number of targeted searches. Many commercial questions won’t trigger a search at all if Claude judges its existing knowledge sufficient — which is one more reason entity clarity and consistent public information matter, independent of citation work.

How is optimising for Claude different from optimising for ChatGPT or Google?

The mechanics of getting found are similar — clarity, crawlability, a real entity behind the page. What differs is Claude’s visible, deliberate citation behaviour: when it does search, it shows what it used, and it appears to weigh whether a source can defensibly back a specific claim rather than just whether it’s popular or highly ranked.

Can I pay to appear in a Claude answer?

No. There’s no paid placement in a Claude answer and no submission process. Anyone offering guaranteed inclusion is describing something that doesn’t exist.

Do I need an llms.txt file for Claude to cite me?

No. It’s a proposed convention that signals crawl and access preferences to certain automated systems, but it isn’t a requirement or a known ranking factor for what Claude chooses to cite. If you want one for completeness across your AI-facing setup, it takes an afternoon to write honestly — we won’t sell it to you as something that improves citation, because it doesn’t.

Does schema markup help Claude cite my site?

It supports entity resolution and is cheap to get right, but it isn’t a citation lever on its own. Check what your pages currently carry with our schema report.

Why would Claude cite a competitor instead of me?

Usually because their page makes a more specific, checkable claim, or is more clearly attributable to a real organisation and author. Claude’s citation behaviour appears to favour defensible sources over merely popular ones, so a page that hedges or restates general consensus is easy to pass over.

How do you track something that only sometimes searches?

By measuring the search-trigger rate as well as citation share, rather than assuming every prompt produces a comparable result. A prompt set where Claude rarely searches tells you something different from one where it searches often but doesn’t cite you — we report both.

How long before I see a change?

Because Claude’s citation behaviour depends on specific, checkable claims and clear authorship, changes to existing pages can show up within weeks. Building genuine authority for a topic you don’t yet own takes longer — typically three to six months, the same range as other engines.