How Claude (Anthropic) Recommends Businesses

Claude does not return ten blue links. When a buyer asks who to hire or which tool to buy, it mentions you, cites a URL, or omits you. On 6 Oct 2026 this URL was indexed (last crawl 16 Sep), had 5 Search Console impressions and 0 clicks, and sat on Google’s “how claude recommends businesses” SERP. Keep Search Console next to the checker.

First published 7 Apr 2026. This is a full rewrite. The April copy stays as labeled history. Do not reprint its influence table as if it were measured data.

Does Claude rank businesses like Google?

No. Google returns ordered blue links and reports them in Search Console. Claude returns a synthesized answer. Three different outcomes get mixed in sales decks:

A mention is not a citation. A citation is not a recommendation. How to get cited in AI search is the citation job. This page is the Claude-specific recommendation job. GEO next to SEO is the pair.

Claude is also not ChatGPT, Gemini, Perplexity, or Google AI Overviews. Profound’s public citation work (Jul 2026) is often summarized as Claude leaning harder on Brave-style web retrieval than ChatGPT’s listicle/YouTube mix. Treat that as a mechanism hint, not a score you paste under every H2. Re-measure your category.

Who already occupies “how Claude recommends businesses”?

Google’s 6 Oct 2026 US desktop results for the recommend query, condensed from the same-day pulls:

Owner-SERP check: this cluster is B2B marketers and operators, not homeowners, not APA “how to cite Claude in a paper.” Pass. Keep the recommend frame.

How does Claude decide what to name?

Honest version without a fake internal diagram:

1. Training memory. Older facts about brands, products, and categories live in weights. A brand that never appeared in durable public text is easy to omit.
2. Tools and retrieval. Claude deployments with web search or connectors can fetch live pages. What they can fetch has to be public HTML, allowed for the bot, and extractable in the first screen.
3. Corroboration. The same entity sentence on your site, a review platform, and a third-party list beats a beautiful homepage alone.
4. Caution. Anthropic’s product posture is conservative. Thin or conflicting NAP/entity data produces hedges and omissions more often than confident picks.

What we are not claiming: a proprietary Claude ranking coefficient, a “Constitutional AI boost” percentage, or a directory tier list with invented weights. The April table on this URL did that. It is gone.

Google Search Central’s AI-features documentation (confirmed 2 Sep 2026) still applies to Google’s own AI surfaces: no extra AI text file or special schema tickets you in. Indexed + snippet-eligible pages still matter. JSON-LD for AI helps machines parse you. It does not mint a Claude rank.

Worked example: this URL on 6 Oct 2026

Worked example = Scope, B2B AI visibility, this page.

1. Google already knows the URL. Inspection: indexed, self-canonical, last crawl 16 Sep 2026. Sitemap lists it. Breadcrumbs only — no Article/FAQ rich result yet.
2. Demand is tiny. 5 impressions in 28 days. Site-wide the same window was 28 clicks / 4,777 impressions. This is not a head-term cash page. It is a durable entity page for a real operator query.
3. The title already matches the better query. “Recommends” is on the SERP we want. “Ranks” is the slug and the sister SERP of rank-hack posts. Do not retitle to “rank in Claude.”
4. April copy was mechanism fan-fiction. Influence levels (Very High / Medium / Lower) had no sample, no date, no engine split. Unique data for a rewrite is GSC + inspect + SERP of this URL.
5. Claude visibility is still a second scoreboard. Search Console cannot see Claude answers. The AI visibility checker and Scope MCP for agents measure mentions and citations. Do not print dogfood scan IDs or internal won/lost grids on this marketing page.

Operator lesson: protect a low-volume indexed URL that already sits on the right SERP. Raise the extractable answer. Do not churn the slug.

Counter-example: optimizing only your homepage for “Claude SEO”

Counter-example: a team ships a 3,000-word “Claude SEO” essay, adds llms.txt, and skips third-party lists because “Claude reads our site.”

What breaks:

• Retrieval has nothing independent to corroborate. Claude (and every other engine) still leans on lists, docs, and reviews other people host. Off-site consensus moves answers more than on-page word count alone (Ahrefs, Jul 2026 on citations generally).
• Google’s sister query rewards rank-hack junk. If your H1 is “how to rank in Claude,” you compete with LocalBoostly-style posts, not with Anthropic’s own product education.
• llms.txt is not a ranking ticket. July–August 2026 crawler studies put hits on llms.txt near noise (~0.1% of AI crawler requests in public tests). Allow GPTBot / OAI-SearchBot / ClaudeBot in robots, ship HTML, stay indexed. See the llms.txt guide.
• You still need Google SEO. This URL’s only Search Console story is five impressions. Pausing technical SEO would not create Claude demand. It would only risk the indexation you already have.

What looks like a win and is not:

• A comparison table that frames Claude against Google as a replacement. Use both channels.
• Reprinting vendor “% of citations” under the wrong sentence. Stat-provenance: primary source, sample, date — or cut the number.
• A second slug /learn/how-claude-recommends-businesses while this one already ranks into the recommend SERP.

What should marketers put in front of Claude?

Practical surface area, in order:

1. One entity sentence everywhere

Same name, category, customer, and proof on the homepage, About, GBP/LinkedIn equivalent, and the first 40–80 words of money pages. Claude cannot confidently recommend a brand that is three different companies across three hosts.

2. Pages built for questions Claude already gets

SaaS: “best X for a 20-person team,” “X vs Y for mid-market.” Services: category + geo + constraint. Put the answer in the first screen, then proof. Content strategy for AI search is the long-form pattern.

3. Third-party pages Brave and others can retrieve

Review platforms in your category (G2/Capterra for software; the real vertical directory for services), independent roundups, documentation hubs, and earned articles. Your blog alone is a weak graph.

4. Clean technical access

Public HTML (not an empty client shell), no accidental noindex, bots allowed, sitemap lastmod honest. JavaScript rendering and AI crawlers covers the fetch failure mode.

5. Engine-split measurement

Run the same 10–20 buyer prompts on Claude, ChatGPT, Gemini, and Perplexity. Log mention / citation / recommendation separately. Claude ≠ ChatGPT. A ChatGPT win is not a Claude win.

Agencies adding this to a retainer keep the SEO line and add a Claude column — see how agencies add AI visibility to SEO.

How should a marketing team spend 30 minutes?

A checklist you can finish before lunch. No new slug required.

1. Inspect this URL (or your Claude pillar). Coverage, last crawl, user vs Google canonical. Indexed is the bar. Crawled-not-indexed is a different ticket.
2. Pull Search Console page EQUALS the URL (28 days). Five impressions is a finding. Zero rows is a finding. Pull query before a title rewrite so you do not chase a homonym.
3. SERP the recommend query and the ranks query separately. If “recommends” is operators and “ranks” is hack listicles, keep recommends in the H1.
4. Rewrite the first 40–80 words. Definition, one dated fact from this URL or this SERP, next action.
5. Add visible Q: / A: pairs already in the HTML before you care about FAQ schema.
6. Link live money URLs only. Checker, agents / MCP, compare, GEO vs SEO. Never internal-link a draft that is not HTTP 200.
7. Ask Claude three buyer prompts about your category. Record omit / mention / cite / recommend. If the UI is blocked, write blocked — do not invent a citation.

How do you measure Claude without mixing it into Search Console?

Do not report checker CTA sessions as organic SEO clicks. Do not report product-MCP dogfood scores here. Off-site lists still matter; extractable first answers still matter once retrieved. Both halves.

Frequently asked questions

Q: Does Claude rank businesses the way Google ranks pages?

A: No. Google returns ordered results you can see in Search Console. Claude returns a synthesized answer that may mention, cite, or recommend you. On 6 Oct 2026 the query “how claude ranks businesses” was mostly rank-hack listicles; “how claude recommends businesses” was the operator SERP this URL sits on.

Q: Is this page proof we rank #1 for Claude advice forever?

A: No. Same-day SERP pulls moved this URL between top and mid-pack on a low-volume query. Search Console showed 5 impressions and position 8.6. Page-level average position is not a durable H1 trophy.

Q: Does Claude use live web search?

A: Sometimes. It depends on the product surface and whether tools are enabled. Plan for both training memory and retrieval. Public, indexed, extractable pages help the retrieval path. They do not guarantee a mention.

Q: Will llms.txt get us recommended by Claude?

A: No. Allow the relevant bots, ship HTML, stay indexed, and earn corroborating sources. llms.txt is not a recommendation ticket.

Q: Should we pause Google SEO to focus on Claude?

A: No. This URL only exists in the recommend SERP because Google indexed it. Keep Search Console. Add Claude as a second scoreboard with the AI visibility checker.

Q: Does Scope replace SEO tools for Claude work?

A: No. Scope measures whether answer engines name or cite you. Semrush, Ahrefs, and Search Console still do Google. Use both. See Scope next to Profound.

Q: Should we create /learn/how-claude-recommends-businesses?

A: No. This slug is already indexed and already appears on the recommend query. A second slug is a cannibal. Change the title and first answer here.

Run the AI visibility checker, connect the Scope MCP server, or keep the pair framing in GEO vs SEO. Claude is one engine. Google is still the other.