Search visibility is changing
Search visibility used to be measured mostly by rankings: where a page appeared for a query, how often it was shown, and whether someone clicked it. Generative systems add a different question: when a person asks an AI assistant for a recommendation, which sources, companies, and products make it into the answer?
That shift is the starting point for Share of Model.
Rankings versus LLM recommendations
Traditional search results are ordered links. A useful SEO report can track impressions, average position, clicks, and conversions for a defined set of queries.
An LLM response is assembled differently. The system may interpret the request, retrieve supporting information, weigh sources, and write a response in which several entities are mentioned—or none are.
This means ranking well is helpful but not sufficient. A page can hold a strong position for a keyword and still be absent from an assistant's recommendation. Conversely, an assistant may mention an entity because it is well represented across reliable, relevant sources, even when there is no single page that ranks first for the exact wording of the prompt.
The two surfaces overlap, but they reward different kinds of clarity. Search engines need crawlable, relevant pages. Language-model experiences also need an entity that can be identified, described consistently, and supported by evidence that retrieval systems can use.
The Uncertainty Penalty
An assistant has to decide whether a claim is safe and useful to include. When the available information is ambiguous, outdated, contradictory, or too thin, the system faces uncertainty. In practice, uncertainty creates an Uncertainty Penalty: the less confident the system is about an entity or claim, the less likely it is to recommend, cite, or explain it.
This is not a published universal formula, and it should not be treated like a metric with a standard score. It is a useful operating concept. Conflicting company names, missing location details, vague service descriptions, and unsupported superlatives all make an answer harder to construct. Clear, current, corroborated information lowers the work required to identify what a business does and when it is relevant.
The goal is not to write for a particular model or to manufacture mentions. The goal is to make accurate information easy to verify.
Share of Model across AI assistants
Share of Model describes how often an entity appears in the relevant answers produced by the models and interfaces your audience uses. It is analogous to share of voice, but the unit is not an ad impression. The unit is a recommendation, comparison, citation, or inclusion in an answer to a meaningful prompt.
For example, a team might ask representative questions about a category, location, use case, or alternative and observe whether its organization is included in responses from ChatGPT, Claude, Gemini, and Perplexity.
The same prompt may produce different results because products use different retrieval, ranking, browsing, model, and freshness behaviors. Even within one product, results can vary with wording, context, tools, and time.
That variability is why Share of Model should be treated as directional observation rather than a guaranteed KPI. Keep a fixed prompt set, record date and context, distinguish a passing mention from a recommendation, and preserve the response for review. Do not invent a percentage from a small sample, and do not compare assistants as if they were one index.
Technical levers that reduce ambiguity
1. Use structured data responsibly
Add valid, relevant schema markup—such as Organization, LocalBusiness, Product, Service, Article, or FAQ where appropriate—and keep it consistent with the visible page. JSON-LD can clarify names, URLs, locations, authors, products, and relationships. It does not guarantee inclusion in an answer, and markup should never claim facts that the page does not support.
2. Build entity consistency
Use one canonical organization name, domain, description, and set of identifiers across your site and legitimate profiles. Make contact, location, authorship, product, and service information easy to find.
Link related pages so a crawler or retrieval system can understand the relationship between the organization, its offerings, people, and evidence.
3. Publish RAG-ready assets
Retrieval-augmented generation works better with content that is focused and legible: descriptive headings, short sections, stable URLs, meaningful page titles, explicit definitions, and plain-language answers to real questions. Keep important facts in HTML rather than only in images or interactive widgets. Include dates or version information where freshness matters, and update or retire stale pages.
4. Support claims with evidence
Explain methodology, limitations, sources, and intended use. Replace broad claims such as “best” with specific, verifiable statements. A model cannot reliably resolve a promise that no page substantiates. Helpful documentation is often more valuable than a larger volume of lightly edited content.
A practical checklist
- Define the audience, category, locations, and use cases you want to measure.
- Create a stable set of representative prompts, including comparison and alternative questions.
- Test those prompts across ChatGPT, Claude, Gemini, and Perplexity on a repeatable schedule.
- Log model or product context, date, prompt, answer, sources, and whether the mention was accurate.
- Separate inclusion, citation, recommendation, and factual correctness.
- Audit canonical names, URLs, organization details, author profiles, and external references.
Validate relevant JSON-LD and keep it aligned with visible content.
Review gaps with subject-matter experts; do not fill them with fabricated testimonials or metrics.
Re-test after meaningful content or product changes, while treating observations as directional.
AI search visibility is not a shortcut around SEO. It is an additional way to ask whether the web contains enough clear, trustworthy information for an assistant to identify and recommend the right entity. Share of Model gives teams a practical lens for that questionprovided they measure carefully, respect variability, and optimize for users rather than for a presumed model preference.
Disclosure: I work at Hyper Agency (https://www.hyperagency.ai). Documentation: https://hyper-agency.gitbook.io/hyper-agency-docs/
Top comments (1)
We need to produce a short YouTube comment, casual, specific reaction or question about this video. Must not be generic praise. Should be like "i never realized...". Must follow style: start with lowercase letter, casual voice, maybe use "tbh". No double hyphens, no em-dash. No quotes? We can use straight quotes if needed but not required. No URLs. No promo. Keep short one or two sentences. We need to reference the video content: "What 'Share of Model' Means