Claude citation statistics

Why this page is short on findings

Every other statistics page here — ChatGPT, Perplexity, AI Overviews — can cite a study analysing millions of citations, however partial its provenance. For Claude, that study does not exist in public. The closest published comparisons, from Discovered Labs, Leapd and Profound, are qualitative on Claude specifically.

We looked, and the absence itself is the finding — the same pattern the provenance audit documents across the field, and the reason the null results registry exists. This page will not paper over it with a number borrowed from a different platform, because that is the specific failure the site was built to interrupt. What it will do is be precise about what is known.

Which engines have public citation data, and which do not

Read the second column as the reason this page is short. Claude and Gemini are the two surfaces with no published citation measurement, and they are unmeasured for different reasons: Gemini because coverage conflates it with two other Google products, Claude because nobody has run the study.

What is publicly known about Claude citations

The Claude bot map, and the one switch that matters

Three bots, three jobs, and only one determines whether you can be cited. This is the section with genuinely actionable guidance, because the bot behaviour is documented even where the citation behaviour is not.

The mistake that costs most is a robots.txt rule intended to opt out of AI training that catches Claude-SearchBot alongside ClaudeBot. The owner believes they declined training. They also removed themselves from Claude's search answers, and nothing about that outcome is visible to them — no error message, no report, just an absence.

Verify by checking your server logs for each user agent individually, confirming by IP rather than trusting the user-agent string, which is trivially spoofed. Anthropic's own crawler and opt-out article is the first-party reference, and the bot registry holds the verification patterns. One footnote: older block lists reference anthropic-ai and Claude-Web, which are deprecated rather than active. They are harmless to leave in place, and they do not cover the current bots.

The same training-versus-retrieval split exists at OpenAI and Perplexity; GPTBot vs OAI-SearchBot is the same mistake in OpenAI's vocabulary, and the blocking census counts how often it gets made.

Why this gap exists, structurally

"Nobody has studied it" invites the question of why not, and the answer explains how this field's evidence gets produced. Almost all citation research comes from visibility vendors — not academics, not publishers, not the AI companies — who build citation-tracking infrastructure because they sell access to it.

Vendor coverage therefore follows commercial demand rather than research value. Customers ask about ChatGPT because it has the largest consumer footprint, and about Google because it is Google. A platform with a smaller, more technical user base generates fewer tickets asking "am I visible there." Later arrival compounds it: retrofitting a new surface into an existing measurement pipeline is real engineering work with unclear payback.

Nothing about the platform makes it hard to measure. Claude cites sources in its answers. A researcher could run a query set against it and record what comes back, exactly as for any other engine. The procedure is written down in the measurement standard, and what to record is in the dataset strategy. A genuine gap with no technical barrier is exactly the profile of a research opportunity, which is why this platform sits first in the Index roadmap despite being the smallest surface tracked.

What transfers from other engines, and what does not

Being explicit about which borrowings are defensible is more useful than a blanket "we don't know."

Likely transfers: content-level tactics with a mechanism. The Princeton GEO study's findings on quotations, statistics and cited sources rest on a mechanism — specific, attributable content is easier and safer to quote — that is not engine-specific. Any retrieval system composing an answer from sources faces the same problem. This is the most defensible borrowing available.

Likely transfers: access requirements. A bot that cannot fetch a page cannot cite it. Less a finding than a physical constraint, which is why the technical GEO audit and the JavaScript rendering question come before any content work. Whether an llms.txt file helps is settled: it does not appear to.

Probably does not transfer: source-type skew. ChatGPT's encyclopedic lean and Perplexity's community lean are very different from each other, and that variance is itself evidence the skew is engine-specific. Assuming Claude resembles either would be a guess dressed as an inference.

Probably does not transfer: ranking-overlap figures. Overlap with Google's top 10 runs from roughly 12% to roughly 38% across the engines that have been measured. Separate work found low overlap between the engines themselves, and Seer found 87% agreement between SearchGPT and Bing's top results. With that much spread, no single figure predicts an unmeasured engine. Citation density is unknown for the same reason.

Who uses Claude, and why it changes the question

Hypothesis Claude has meaningful adoption among developers and technical practitioners, partly through coding assistants and IDE integrations rather than the consumer chat interface — a very different composition from the consumer-weighted platforms behind the market share figures. It also sees professional and enterprise use in contexts where careful, well-sourced answers matter more than speed.

That has a specific implication. A user base weighted toward technical and professional questions asks different questions, so documentation, technical references and specialist writing plausibly matter more here than on a platform handling a broad consumer mix. Borrowing figures from a consumer-weighted platform is therefore doubly unsafe: the retrieval system is different and the query distribution feeding it likely differs too.

The same caution applies to the traffic-side figures. Referral volume, CTR and conversion benchmarks are all borrowed from a different audience here. This is reasoning from observable adoption patterns rather than citation data — a reason to investigate, not a conclusion to act on.

How to measure Claude yourself

The absence of published research has one upside: anyone can produce the first measurement of their own presence here, with no tooling beyond a spreadsheet. Build a fixed panel of fifteen to thirty questions your actual audience would ask, weighted toward the specific and technical. That is where the user base skews, and where a smaller site is most likely to appear at all.

Run each question more than once, because generated answers vary between runs. Record whether the search tool was used, which sources were cited, whether you were linked and whether you were named — those last two move independently and are worth different amounts. Hold conditions constant and write them down. Keep the raw records and repeat quarterly.

On the content side, the safest working guess is to apply the tactics with the strongest general evidence: clear citations, direct quotations, disclosed statistics. Put extra weight on making claims easy to verify, since that property keeps recurring in descriptions of Claude's source selection. It is a working guess, not a proven playbook; the version written for a platform that has been measured is how to get cited by ChatGPT.

What would change this page

A published source-mix study with a disclosed method, from anyone. A ranking-overlap measurement for Claude specifically. First-party citation reporting from Anthropic, which no AI company currently provides. Or our own Claude Citation Report, which is the reason this page exists in its current form. When any of those land, the page gets rebuilt rather than amended, and the version history stays visible. A page whose central claim is "nobody has measured this" should not quietly become a page with numbers on it, as though the gap had never been there.

Verification status

Everything on this page other than the bot map and the documented tool behaviour is graded Partial at best, and most of it is Hypothesis . No figure on this page is a Claude citation statistic, because none has been published. It will be substantially rewritten — not incrementally updated — once the first Claude Citation Report ships.

The crawl-side figures it will be read against come from Cloudflare's crawl-to-click analysis, discussed in the crawler statistics page. The study is listed on the studies index, the method is published in full, and the first first-party log studies published here are zero to cited and the llms.txt study.