How Do We Get Cited
by Claude? Know
Our Playbook

On this page 12 sections

1. The Short Answer
2. Why Claude Is the Hardest AI Tool to Win
3. Claude Answers Two Ways
4. How Claude Picks Its Sources
5. Case Study: Named by Claude and Three Other AI Tools
6. Schema Markup: How to Make a Page Easy to Cite
7. Why Claude Citations Need Off-Site Brand Mentions
8. Our Claude Citation Process
9. What a Claude Citation Is Actually Worth
10. What Does Not Move the Needle
11. Where to Start
12. FAQs

Normally, we track four AI assistants for clients. Claude sends the least traffic of the four. It is also the hardest one to game.

On one campaign, Claude drove 173 sessions out of 1,639. On another, it showed up as a brand-new channel with 10 users over a full year.

Those are small numbers, and we will not dress them up. What makes Claude worth the effort is the bar it sets. It hedges more than the others. It names fewer brands. And it leans harder on sources it can point to.

So when Claude does cite you, it usually means your signals are already strong enough that the other tools picked you up months ago. A Claude citation is a late confirmation that the answer engine optimization work is real.

Why Claude Is the Hardest AI Tool to Win

Ask most assistants for the best provider in a category, and you get three brand names. Ask Claude the same thing, and you may get a list of criteria instead, with no names at all.

That is the whole problem in one sentence. Getting named means clearing a higher bar, not shouting louder.

It also makes Claude a useful test. If your brand shows up in a Claude answer, your entity signals are in good shape. If it does not, something upstream is weak.

There is a second reason to care, and it has nothing to do with traffic volume. Claude holds a small share of consumer chatbot visits, roughly 9% of generative AI web traffic against ChatGPT’s 52.7%, according to Similarweb data. Its share was 1.6% a year earlier.

Source: Similarweb data via PPC Land

The business side looks nothing like that. Menlo Ventures surveyed around 495 enterprise AI buyers and found Anthropic holds an estimated 40% of enterprise LLM API spend, and 54% of the coding segment against OpenAI’s 21%.

Source: Menlo Ventures

Read those two figures together. Claude is a small consumer channel and a very large professional one. If you sell to developers, technical buyers, or enterprise teams, Claude reaches a bigger slice of your market than its traffic share suggests.

Claude Answers Two Ways, and Only One of Them Cites You

This is the part most guides skip, and it changes what you should work on.

Claude does not search the web for every question. It answers from what it already knows unless the question needs something current, and it only shows citations when it actually retrieves a page. No search means no link.

Independent research by Josh Blyskal put a number on it. Across a set of tested prompts, Claude ran a web search for 36.6% of them and answered the other 63.4% from training knowledge alone.

So there are two ways to be named, and they need different work.

1. 01

   Path one, the training corpus:

   Claude names brands it already knows from the web it learned on. You cannot optimise your way into that in a quarter. It comes from years of consistent presence, coverage and mentions.

2. 02

   Path two, live retrieval:

   When Claude does search, the game is ordinary retrievability. Crawl access, a clean page, a fact it can lift and credit.

Both paths reward the same underlying work. The difference is the clock. Path two can move in months. Path one is the reason a Claude citation usually arrives last.

How Claude Picks Its Sources

Claude has its own crawlers, and there are three of them. Anthropic’s crawler documentation lists ClaudeBot for training, Claude-SearchBot for search indexing, and Claude-User for fetching a page when someone asks Claude about it.

Each one needs its own line in robots.txt. Blocking ClaudeBot does not block the other two, and you have to repeat the rules on every subdomain.

Anthropic states that blocking Claude-User or Claude-SearchBot may reduce your visibility in Claude’s answers, which is the whole game here. It also warns that blocking by IP is unreliable, because that can stop the bot reading your robots.txt at all.

Anthropic now publishes its crawler IP ranges at claude.com/crawling/bots.json, so you can verify a bot is genuine. We check all of this on day one of every technical SEO audit.

Claude searches through a partner index.

Anthropic lists Brave Search as a subprocessor for web search, first spotted on its Trust Center, and has never announced the arrangement publicly. The observed behaviour matches. In the same Blyskal research, 79.2% of Claude’s cited URLs also appeared in Brave’s top ten across roughly 400 queries.

Treat that as a strong correlation rather than confirmed architecture. The practical read stands either way. If you are invisible to general web crawlers, you are invisible to whatever index Claude is querying.

Claude rewards facts you can check.

A page that says “we are the leading provider of X” gives a large language model nothing to work with. A page that says “median approval time across 400 accounts was 26 hours” gives it a line it can quote and credit.

We rewrote a lot of client copy on that one point alone, and it’s now baked into our SEO content writing briefs.

Claude cross-checks before it commits.

One claim on your own domain is weak. The same claim on eight sites with no relationship to each other is strong.

Ahrefs put numbers on this in a study of 75,000 brands: branded web mentions correlated 0.664 with brand visibility in Google AI Overview, against 0.218 for backlinks.

That study covers AI Overview rather than Claude, and the authors are clear that correlation is not causation. But the mechanism is the same one Claude uses. A model learns your brand from what other people write about it.

Claude wants structure it can read fast.

Clear headings. One idea under each. FAQ blocks. Comparison tables. Structured data does not rank you here, but it decides whether a model has to work out your meaning or can simply lift it.

Case Study: Named by Claude and Three Other AI Tools

Our client sells agency ad accounts to media buyers. Their buyers check everything before they spend, so a lot of them now hand that research to an AI assistant.

We have worked on the site since launch, close to two years. By the end of that period, all four major tools named this client first for its core category question. ChatGPT, Gemini, Perplexity, and Claude.

AI sessions reached 1,639 for the period, up 111% on the six months before. ChatGPT sent 936. Gemini sent 342. Perplexity sent 189. Claude sent 173.

ChatGPT936

Gemini342

Perplexity189

Claude173

1,639 AI sessions in total, up 111% on the previous six months.

Three parts of the build earned the Claude result.

Data pages, not keyword pages:

We built pages full of numbers: Facebook ads statistics, Google ads statistics, and ad cost benchmarks. Each figure is stated plainly with its source named and dated. That is a format an assistant can quote without hedging.

Across those six months, the Facebook ads statistics page grew 675% in Search Console, and the Google ads statistics page grew 3,850%. Both now pull a large share of the AI traffic. Our statistics page work follows the same template every time.

One brand description, used everywhere:

Directories, profiles, and other people’s write-ups all describe this brand the same way. What it is. What it does. Which platforms it covers.

SEOs call this entity consistency, and it runs wider than your own site. Your company profile, your directory listings, and any reference entries about you all feed the same picture. When those descriptions clash, Claude hedges. When a dozen sources agree, it commits.

Mentions on other people’s sites:

Claude reads the live web before it answers, so much of this work happened nowhere near the client’s own domain. Earning those mentions is the job of digital PR, and most SEO plans skip it because it does not move a rank tracker.

Four assistants. Four different systems. One matching answer. That is far harder to fake than any single tool’s output, and the full breakdown is in our AI search optimization case study.

Schema Markup: How to Make a Page Easy to Cite

We ran a separate campaign for a medical cold chain packaging maker. On-page work only. No new content, no link building, no rebuild.

We optimised 13 pages, four buying segment pages and nine product pages. One H1 on each. Meta titles rewritten with the focus keyword first. Self-referencing canonicals. Real alt text. Internal links built by hand. Then FAQ blocks carrying FAQPage, Product, and Service schema. That is the standard scope of our on-page SEO work.

The result: 9 of 20 tracked keywords started appearing in Google AI Overview. Seven keywords sat at position 1. The term “blood bank shipper” went from outside the top 100 to number one. Zero backlinks were built.

The lesson carries straight over to Claude. Say a model needs to answer “what is a blood bank shipper.” The easiest source to use is a page with an FAQ block that answers that exact question in a form a machine can parse.

Why Claude Citations Need Off-Site Brand Mentions

The hard part of getting cited by Claude is that you cannot do it from your own website. You need other sites saying things that match.

In the same Ahrefs study, brands in the top quarter for web mentions averaged 169 AI Overview mentions. The quarter below them averaged 14. The bottom half barely showed up at all. That gap is the reason we treat off-site work as the main event and not the follow-up.

Three placement types do most of that work, and they are not interchangeable.

1. 01

   Editorial posts on truly relevant sites:

   Topical relevance beats domain rating here by a wide margin. A DR 45 site that is genuinely about your industry counts as a category signal. A DR 78 general blog counts as noise. We place these through niche guest posting, and for anything B2B or software we start with SaaS SEO publishers.

2. 02

   Expert quotes and resource pages:

   Curated lists on industry and university sites last for years, and they carry a hint of endorsement that a normal guest post does not. HARO link building is the fastest route in, because a quoted expert is exactly the kind of source a model likes to credit.

3. 03

   News and trade coverage:

   Editorial mentions on regional and national outlets give a model a source with clear standing. These take the longest to land and cost the most, which is why we treat them as the last layer, not the first.

You do not need all three. You need enough matching mentions that your story holds up whichever source the model happens to read that day.

Our Claude Citation Process

Every AEO campaign we run follows these seven phases, in this order.

Swipe to see the full table

That last row matters more than it looks. Asking the same question with search enabled and disabled tells you which of the two paths is working, and that decides whether you push on content or on off-site mentions next.

What a Claude Citation Is Actually Worth

Four things worth knowing before you plan a budget around this.

The traffic volume is small:

Claude sent 173 sessions on our best campaign. Cloudflare measured Anthropic’s crawl-to-referral ratio at 70,900 page requests for every referral sent back over one measured week.

If your business case rests on Claude referrals paying for themselves by month three, the business case is wrong. Treat a citation as proof the entity work is landing, and take the clicks as a bonus.

Referral data runs low:

AI tools often strip the referrer, so what you measure is a floor rather than a total. Some of what your reports call direct traffic came from an assistant.

Answers shift, but less than they look:

Ask the same question two ways and the wording and brand order change. The substance moves far less. We test in batches and report the pattern across a month, not a position on a single day.

Two years is normal in a tough category:

The ad account client took that long. A jewelry client of ours moved much faster because its category was less crowded, though Claude still only sent 10 users across that year. The month-by-month detail is in our AI SEO case study. Anyone quoting you 90 days for a hard niche is describing a different job.

What Does Not Move the Needle

Three things get sold as Claude tactics and do not hold up.

• Publishing an llms.txt file.

  Anthropic publishes one for its own developer docs, which is probably why the idea spread. There is no evidence Claude reads yours. It costs nothing, but it should never come before crawl access and content work.

• Stuffing keywords into headings.

  Claude is not matching strings. A heading that reads like a real question beats one padded with a keyword, because the answer underneath is what gets lifted.

• Chasing domain rating.

  A high DR placement on an unrelated site adds almost nothing to a model’s picture of what category you belong in. Relevance is the signal.

Where to Start

1. 01

   Start with your own pages. Plain answers, real figures, FAQ blocks, and matching schema. It is the cheapest step, and everything after it depends on it.

2. 02

   Then check your crawl access. Open ClaudeBot and Claude-SearchBot in robots.txt, and make sure your main content does not need JavaScript to appear. This takes an afternoon and blocks every other result if you skip it.

3. 03

   Then build matching mentions on sites that are truly about your subject, and keep your brand description identical on every one of them. A steady monthly pace beats one large burst, which is why most of our clients run link building packages rather than buying 40 placements in a quarter and stopping.

Want the full campaign numbers? They are in our SEO case studies. If you would rather talk it through, book a call and we will run a prompt audit on your category first.

FAQs