How to Rank in Claude: Fable Is the Headline, but Sonnet Is the Visibility Reality

1. Start with the Claude buyers actually use

Claude Fable 5 is Anthropic's most capable generally available model. It was first released on June 9, 2026, temporarily suspended, and restored globally on July 1. Anthropic describes it as a model for difficult, long-running knowledge work and coding that can sustain tasks lasting hours or days.

That does not make Fable the everyday default. Claude Sonnet 5 is the default model for Claude Free and Pro users. Anthropic recommends Sonnet for most applications that need a balance of intelligence, speed, and cost at scale. Its enterprise consumption guide says to make Sonnet the organization-wide default and reserve Fable for the highest-value, most complex work.

Anthropic has not published a unique-user breakdown showing how many people choose Fable versus Sonnet. Do not turn token volume into a people statistic. The honest conclusion is narrower: public gateway data shows materially more Sonnet traffic, and the product defaults and price difference help explain why.

2. Stop looking for one permanent Claude rank

When Claude answers a recommendation question, it may work from its trained knowledge, the current conversation, files and connected tools, live web search, or some combination of them. If the answer needs current or specialized information, Claude can decide to search. If it searches, Anthropic says the process can repeat several times in one request.

The result changes with the buyer's wording and constraints. Best CRM is a different question from best CRM for a five-person real estate team. Best med spa is different from best med spa near Draper with transparent Botox pricing. A brand may appear for the broad category and disappear when the buyer adds price, location, fit, risk, or urgency.

Anthropic does not publish a master scoring formula that tells marketers how many points a review, backlink, schema field, or keyword earns. No responsible agency can promise a permanent Claude position.

3. Understand Claude's search and filtering loop

Anthropic's public documentation describes a multistep process. Claude interprets the request, decides whether current web information would improve the answer, creates a targeted search query, and can run additional searches as the task develops. The search service returns results, Claude evaluates the evidence, and the final answer cites web sources it actually uses.

Newer versions of Anthropic's web search tool add dynamic filtering. Claude can write and run code that filters search results before they enter its main context, keeping the material it considers relevant. Discovery is only the first gate. Your page can exist and still be filtered out, misunderstood, contradicted by another source, or judged to be a weaker fit for the buyer's constraints.

That makes vague relevance expensive. A page that buries its category, customer, location, price, proof, or limitation gives Claude less reason to keep it and less material to use in a defensible recommendation.

4. Give Claude evidence it can qualify and defend

Claude's final answer is shaped by its training and published constitution. Anthropic says the constitution directly influences Claude's behavior and asks it to be truthful, calibrated, non-deceptive, non-manipulative, and respectful of the user's autonomy. The constitution is not a search-ranking formula. It does help explain why unsupported superlatives and manipulative marketing language are poor raw material for a recommendation.

Do not try to sound like the obvious choice. Publish enough specific evidence for Claude to explain when you are the right choice. Useful evidence includes original research, dated statistics, product specifications, transparent pricing, methodology, comparison tables, expert commentary, customer results with context, and honest limitations.

Customers love us is promotional copy. A dated result with a defined sample, method, and before-and-after measure is evidence. The second is easier for Claude to qualify, compare, and cite.

5. Keep all three Anthropic crawlers straight

Anthropic identifies three bots with different jobs. ClaudeBot collects public web content that may contribute to model training. Claude-SearchBot indexes content to improve search relevance and accuracy. Claude-User retrieves pages in response to a user-directed request.

For live Claude visibility, check both Claude-SearchBot and Claude-User in robots.txt, your CDN, firewall, and bot protection. Allowing ClaudeBot is a separate training-data decision. You can block training access without necessarily blocking live search access.

Crawler access is eligibility, not endorsement. Allowing the bots does not guarantee that Claude will retrieve, retain, cite, mention, or recommend your page.

6. Know what Claude shares with ChatGPT and what is different

The foundation is more similar than different. Claude and ChatGPT can both decide when a question would benefit from the web. Both can turn a conversational question into targeted searches, retrieve results, synthesize information, and cite sources. Neither publishes a complete formula for brand recommendations. Neither gives a business one permanent rank across all users and prompts.

Both reward the same practical fundamentals: crawlable pages, a clear category, direct answers, current facts, credible proof, consistent third-party information, and repeated measurement. In both systems, one meaningful constraint can change the shortlist.

The public explanations differ. OpenAI emphasizes rewriting a request into one or more targeted queries, then evaluating results using meaning, intent, relevance, and recency. Anthropic's developer documentation exposes more of the agentic search loop, especially repeated tool use and dynamic filtering before generation. Claude also separates search indexing, user-directed retrieval, and training across three named bots, while OpenAI separates OAI-SearchBot for search from GPTBot for training.

Claude's published constitution gives unusually detailed guidance about calibrated claims, non-manipulation, and preserving user autonomy. That can shape how a recommendation is framed, although it is not a retrieval score.

7. Make your company easier to keep during filtering

Imagine a buyer asks which AI visibility platform is best for a small agency that needs client-ready reports, tracks Claude and ChatGPT, and costs less than enterprise software. A competitor can win that answer without having the biggest website or the most backlinks.

Claude may find that the competitor states its category and buyer clearly, publishes current pricing, names the models it checks, shows a real report, explains what the product does not do, and has consistent descriptions across its website, documentation, reviews, and partner pages.

If your page says only that you unlock next-generation AI visibility, Claude has to infer the buyer, product, evidence, and fit. A clearer competitor is easier to keep during filtering and easier to defend in the final answer.

8. Test Sonnet and Fable separately, then measure the pattern

Start with the buyer questions people ask immediately before choosing. Include the constraints that change the answer: location, budget, company size, use case, urgency, integrations, risk, and desired outcome. Do not test only best category. Test the real decision.

Use Sonnet as the practical baseline because it is the Free and Pro default and the higher-volume model in available public gateway data. Use Fable as a second test for complex, high-value questions. Save the exact prompt, model, date, search setting, answer, brands mentioned, and citations.

One favorable answer is not a ranking. Track whether the brand appears, whether it is recommended or merely mentioned, which competitors appear, which claims Claude makes, which sources it cites, whether the answer is accurate, and what changes after one specific improvement.