How to get cited by Claude (Anthropic) as a small business

Why Claude is a different animal

Claude's training data and its retrieval stack are not the same as the ones behind ChatGPT. Anthropic leans toward sources it can treat as authoritative. Well-structured, factual, stable. That bias carries through into how Claude decides what to repeat when a user asks a specific question.

The practical effect for a small business is that raw volume of content does not help much. Claude does not reward a site that publishes three blog posts a week. It rewards a site that states its facts clearly, marks them up with schema, and does not contradict itself across pages. Clarity beats quantity. Specific markup beats clever copy.

Claude is also more conservative about naming businesses at all. For many prompts it will decline to recommend a specific provider and will describe a category instead. Getting past that threshold is the job.

The six signals Claude weights

In rough order of impact for a small business site.

1. Clean, well-structured HTML with semantic headings. H1 for the page topic, H2 for sections, H3 for sub-points. One H1 per page. Lists as real lists, not styled paragraphs. Claude reads the document tree, not the rendered page.
2. Accurate and current JSON-LD. Organization, LocalBusiness or ProfessionalService, Product, and FAQPage where they apply. Fields filled in, not left empty. Claude treats a well-formed block as a statement of fact.
3. Direct answers near the top of pages. The customer question answered in the first 60 words, in plain English, before the marketing copy starts. Answers buried at the bottom do not get quoted.
4. External corroboration. Wikipedia, trade bodies, reputable industry press, local chambers. Claude checks whether claims on your site are supported elsewhere, and raises confidence when they are.
5. Site age and internal consistency. A domain that has been saying the same thing for three years beats one that rebranded last month. Consistent NAP, categories, and positioning across pages.
6. Clear who, what, where, price, contact facts on every relevant page. Not all on one page. On every page a customer might land on. Repetition is the point.

What to ship this week

A concrete checklist. None of these need a developer.

• Add Organization and LocalBusiness (or ProfessionalService) JSON-LD to your site-wide head, with sameAs pointing at your real social and directory profiles.
• Put a FAQPage block with six real customer questions on your main services page. Real questions, short answers, wrapped in schema.
• Rewrite your About page with Person schema for the founder or lead practitioner. Named human, specific years, named town.
• Fix NAP consistency. Name, address, phone number identical on your site, Google Business Profile, Companies House or equivalent, and every directory listing. Mismatches erode confidence.
• Cross-reference your internal links. Service pages link to the About page. About page links to services. FAQ page links to pricing. Claude follows these to build an entity picture.
• Add a plain-text price or price range on service pages. Even "from 75 pounds" gives the model a fact to anchor to. Vague pricing reads as missing data.
• Publish one clean case study with a named client, a dated outcome, and a short quote. That is more useful than ten generic posts.
• Check your site renders without JavaScript. If the key facts only appear after JS runs, Claude's crawl may miss them.

What does not move the needle

Save your time. These either do nothing or go backwards.

• Keyword stuffing. Claude reads for meaning, not density. Repeated phrases read as low quality.
• AI-written filler. Long posts with no specific facts, no named examples, no dates. Claude is trained to spot this pattern and discount the source.
• Fake reviews. Do not. The FTC Consumer Reviews and Testimonials Rule has been in force since 21 October 2024, with civil penalties of 51,744 dollars per violation. Beyond the legal risk, inconsistencies between a glowing review profile and a thin site tell the model something is off.
• Thin "best X in Y" pages with no substance. A page that lists ten providers with no first-hand detail is treated as low value. If you write one of these, it needs real commentary, real contact details and real dates.
• Buying links. Not just a Google penalty risk. Paid link networks tend to have patterns a retrieval system can flag, and the sources are not ones Claude treats as authoritative anyway.

How to tell if Claude is starting to cite you

Monthly check. Open a fresh Claude chat, no memory, and run the same three prompts your customers would ask. Phrasings that work well for testing:

• "What are the top [service] providers in [city]"
• "Who sells [product] in [region]"
• "Can you recommend a [service] for [use case]"

Run each prompt three times in separate chats. Claude's answers vary run to run, so consistency across three attempts is the signal you want. If you are named in two of three, you are in. If you are named in one, you are on the edge. If zero, the work in the checklist above is where to start.

Log the results in a simple sheet. Date, prompt, whether you were named, which competitors were named. Over three or four months that log will tell you what is working.

Happy to answer anything, Bob.