Jaxon Parrott
BriefWritingFAQMachine RelationsPressAbout
← Writing

How to Get Recommended by Claude

Jaxon Parrott
Jaxon Parrott
July 23, 2026Updated November 14, 2026Machine RelationsGlossaryPR & AI search

I read a brand that appears in other answer engines and disappears in Claude as a trust gap, and I fund earned media, original evidence, and a consistent entity before I fund another metadata pass. Claude gives brands the highest owned-domain citation share of the engines in TripleDart's account, 9.1%, double ChatGPT's 4.5%. Separately, per Adobe as TripleDart cites it, AI-referred visitors converted 31% better than visitors from other channels in the 2025 holiday shopping season, across all AI engines and not Claude specifically. The same design that makes the citation valuable is the design that refuses a thin recommendation.

The method for getting into those answers is How to Get Cited in Claude AI Answers. That page is the deeper resource, the method.

What I do not import from the other engines

ChatGPT pulls live results through Bing, so a conventional ranking can become an answer. Perplexity cites widely. Gemini leans on Google's own index. Those paths have familiar levers. Claude was trained with Constitutional AI methods, which mentionagent argues make it favor high-trust sources over promotional content, and its training corpus is weighted toward long-form journalism, books, academic literature, and high-trust web sources. The model is willing to decline a recommendation when the evidence is thin. I treat that refusal as information about the record, not as a bug to route around.

If a buyer prompt returns us in ChatGPT and Perplexity and omits us in Claude, I do not commission a Claude-specific trick. I treat the omission as the stricter reading of the same evidence. The technical buyers, developers, and researchers in Claude's user base are often the people a B2B company already wanted. Anthropic's reported run-rate went from $1 billion early in 2025 to $14 billion annualized by February 2026. I do not need a larger user-count story than that to care about the omission. The omission is the brief.

Three crawlers, three clocks

Claude runs three crawlers, and I do not brief them as one "Claude SEO" task.

ClaudeBot builds training data. What it indexes can become part of the base the model answers from when someone asks for a recommendation in a category. Influence there is measured in model updates, which means months, and it runs through sources Anthropic's selection treats as credible. I do not promise a training-corpus result inside a quarter.

Claude-User fetches a page when Claude needs something current or a person asks it to read a URL. Timely, specific pages matter here, on a shorter clock.

Claude-SearchBot is Anthropic's search crawler, and Anthropic documents an in-product web search tool. When search is on, Claude retrieves from a search index reported to be Brave's (TechCrunch, March 2025: Anthropic "appears to be using Brave"), not Bing or Google. There is no backdoor through Bing's or Google's ranking. Either the brand is in sources that index will retrieve, or it is absent from the searched answer. Live-search influence, once a credible source is crawled, can show up in days or weeks. The structural work of earning the source is still measured in quarters.

I put those three clocks on one page so a team does not celebrate a schema deploy as if it had changed ClaudeBot.

The three signals I fund, in order

Earned media first. I expect a visibility program made of posts, listicles, and sponsored placements to be faint here. I fund placements in publications this corpus already treats as journalism, and I do not fund a social calendar and call it a Claude plan.

Named expertise second. TripleDart says Claude rewards answer-first content with named experts, in contrast to Google's keyword-optimized pages. Author credentials, a visible method, and an organization a reader can identify all move with citation probability. Schema helps a crawler find a page. It does not create the trust. I publish one study we can defend before I publish ten explainers that cite someone else's study.

Entity consistency third. Full entity resolution, again in that method's framing, is what lets Claude treat a brand as a known entity rather than an anonymous site. The name, the description, and the core claims have to agree across independent sources: reviews, earned placements, industry databases, technical docs, and our own pages. Ahrefs, on 75,000 brands, found branded web mentions correlating 0.664 with appearance in Google AI Overviews, against 0.218 for backlinks (Spearman). That study is AI Overviews, not Claude. I read it as evidence about how models weight web text in general, and I say so in the memo. The operating conclusion I keep is that being talked about, unpaid, matters more than being linked. I do not paste the 0.664 onto a Claude forecast.

What I do with a volatile score

One brand in the TripleDart dataset peaked at 13.6% visibility and fell to 0.64% within two months. I show that pair to anyone who wants to declare victory off a single cycle. The brands that hold a position are the ones whose authority is structural. A formatting win that is not backed by earned sources is the kind of win that can evaporate inside a quarter.

Utility content outperforms ordinary blog posts by 6x to 30x in Claude citations in that account. Competitor-comparison queries deliver 2x to 3x higher mention rates than generic category queries. I commission the page that helps a buyer decide: a tool page, a diagnostic, a pricing comparison, a method note. I do not commission a keyword article and hope the 6x applies to it. The range is wide because the underlying pages are different. I copy the job, a decision aid with evidence, and I do not copy the top of the range into a goal.

The test I run is the same buyer prompt in Claude, ChatGPT, Perplexity, and Gemini. Something a prospect would type: the best options in our category for our kind of buyer, how to solve the problem, alternatives to the incumbent. Presence in the engines with wider citation and absence in Claude is manufactured visibility that clears a lower bar. I write that sentence without turning it into a claim that any one engine is ignorant of a person. The subject is whether the brand's evidence survives a stricter source test.

The stack I use so this does not collapse into "write more Claude content" is Machine Relations. The weighting I am trying to earn, as distinct from a formatting pass, is earned authority.

I also separate the audience fact from the citation fact so a growth chart does not become a content plan. Soar's reported Anthropic run-rate path, $1 billion early in 2025 to $14 billion annualized by February 2026, says the surface is large enough to deserve a read every month. Adobe's 31% AI-referral conversion premium, as TripleDart quotes it, says AI-referred sessions can be worth more than a raw user count; it is not a Claude-specific number. Neither number tells me which sentence to publish. The 6x to 30x utility range and the 2x to 3x comparison-query rate tell me that. The 9.1% versus 4.5% owned-domain comparison tells me our own pages can carry more of Claude's citations than ChatGPT's, and still tells me that most citations are not our domain. I put the business figures in the demand paragraph and the citation figures in the production paragraph, and I do not let a run-rate slide approve a listicle.

The decision

I query the buyer prompt before I query our name. If we are missing, I buy one earned placement that states a checkable claim, publish one original number with its method, and align the entity across the profiles and docs a technical buyer would find. I schedule the next read in a later week, and I judge Claude-SearchBot on a weeks-scale and ClaudeBot on a model-update scale. I do not buy a metadata project as the first response to silence.

Scope: brands that sell to technical buyers and already appear, or hope to appear, in ChatGPT, Perplexity, Gemini, and Claude. The figures above are TripleDart's (including the Adobe conversion figure it quotes), Ahrefs' 75,000-brand correlation for AI Overviews, Soar's reported Anthropic run-rate path, and the crawler split Soar and Anthropic's docs describe. It does not transfer the 9.1% owned-domain share into a promise that our domain will be cited at 9.1%. That number is a corpus-level share of citations pointing at owned domains, and it is a reason owned pages can matter more here than they do in ChatGPT's 4.5%. It is not our target.

What would change my mind: a disclosed Claude-specific measurement, not the AI Overviews correlation, showing that on-page formatting and social distribution move citations as much as earned media. I also treat metadata as the first fix if our pages were absent from Claude-SearchBot while the same claims were already present in sources Claude cites elsewhere. In that case the record exists and the retrieval is the hole. I do not change the order because a visibility score spiked and might repeat the 13.6% to 0.64% collapse.

FAQ

How is a Claude recommendation different from a ChatGPT one, in the numbers I cite?

ChatGPT's search runs through Bing and cites broadly. Claude's search retrieves from a search index reported to be Brave's, not Bing or Google. Owned-domain citation share is 9.1% in Claude against 4.5% in ChatGPT in that account. I brief those as corpus properties. I still earn the third-party sentence, because 9.1% means the other citations are not our website.

How long should I tell the team to wait?

Months for ClaudeBot, because the training corpus moves when Anthropic updates the model. Days to weeks for Claude-SearchBot after a crawled source is indexed. Quarters for the earned placement, the original dataset, and the entity alignment. I do not report a schema change as a Claude result on a shorter clock than the crawler it could have affected.

What should we publish if we can only ship one page?

A decision aid. Utility pages, tools, diagnostics, pricing comparisons, and method notes outperform standard posts by 6x to 30x in the TripleDart Claude citations, and comparison queries mention brands at 2x to 3x the rate of generic category queries. I put one original number and its method on that page, and I prefer to have one defensible page rather than a pile of restated statistics.

What do I do with the Ahrefs correlation?

I label it. Branded mentions correlated 0.664 with Google AI Overview appearance, backlinks 0.218, on 75,000 brands. The study is not a Claude measurement. I use it as a reason to care about unpaid mentions in web text generally, and I keep Claude decisions tied to Claude-specific evidence and the three crawlers.


About the author

Jaxon Parrott founded AuthorityTech in 2018 and coined Machine Relations in 2024. He writes about earned authority, AI discovery and founder judgment.

Continue reading

Does PR Help With AI Search Visibility?

PR improves AI search visibility when it creates third-party evidence an engine can retrieve and cite. The coverage has to say something specific.

May 8, 2026

A Brand Strategy for AI Search Starts With Evidence

Build an AI search brand strategy around a defensible claim, clear identity and independent evidence. Match the next investment to the actual discovery gap.

June 30, 2026

Is PR Dead?

PR is not dead. The clip-and-share-of-voice scoreboard is. Earned media now has a second reader, the AI engines buyers ask first, and that is the job I measure.

October 24, 2026

Sections

  1. What I do not import from the other engines
  2. Three crawlers, three clocks
  3. The three signals I fund, in order
  4. What I do with a volatile score
  5. The decision
  6. FAQ
AuthorityTechMachine Relations
XLinkedIn
© 2026 Jaxon Parrott
GlossaryTopicsSkillsArchiveContactPrivacy