Book a demo
Product
AI visibilityCompetitorsCitationsGEO score and page fixesPrompts and demandAI crawlers and referralsContent StudioAgents and automationsAurora strategistCommerceEarned mediaLocal networkAPI and MCPThe whole platform
Solutions
Brand and marketingSEO and GEO teamsContent teamsPR and communicationsE-commerceMulti-location and localAgenciesExecutives
Free tools
Free AI visibility scanPrompt set generatorAI readiness checkTemplates
Resources
Learn GEOUse casesHoliday AI Index 2026BlogNewsletterMethodologyDevelopersChangelogIntegrationsSecurity and trust
PricingAboutSign in
Check your AI visibilityBook a demo
Methodology

How Lighthouse measures AI visibility

The method behind every number in Lighthouse, including what it does not measure yet.

How does Lighthouse measure AI visibility?

Lighthouse types each tracked prompt into the AI engine’s own consumer page in a headless browser, in a fresh session, on a home internet connection in the market being measured, and stores the full answer, its sources and a screenshot. AI visibility is the share of measured answers that name your brand, and an answer that could not be collected is left out of the count rather than scored as a miss.

Last reviewed

Which engines does Lighthouse measure?

ChatGPT, Gemini, Google AI Overviews, Google AI Mode and Qwen are measured today, on their logged-out consumer pages. Claude, Perplexity, Copilot, Grok, Meta AI and DeepSeek have no logged-out experience, so they are asked signed in, in the same kind of browser, set up to behave like a real user. Until those accounts are live, those engines are shown as not measured, with that reason, rather than as zero.

How is each answer collected?

  • Every prompt, in every run, is asked this way. There is no API route and no fallback to one.
  • The prompt is typed verbatim into a new conversation, in a fresh headless browser, logged out, with no cookies or history.
  • The browser connects through a home internet provider (ISP) in the market being measured, checked before each answer is kept, at desktop size.
  • When an answer cannot be collected this way, it is recorded as not measured, with the reason, never filled in another way.
  • Model choices and search settings are left at the page’s defaults, as a buyer would find them.
  • The answer text, every cited source and a screenshot (with a checksum) are stored, so any number can be opened and checked.

How often is each prompt asked?

Once per engine in each run. Runs repeat on the schedule you choose: daily, weekly or monthly, weekly by default. The sample builds up across runs, which is why trends are more reliable than a single run.

How is share of voice counted?

Share of voice counts only category prompts: the questions that name neither you nor a tracked rival, such as “best hiking boots for wide feet”. A brand earns one mention per prompt when any measured answer to it names the brand, and your share is your mentions divided by the mentions of you and your tracked rivals. When no category prompt was measured, share of voice reads not measured, never 0. Brands owned by the same company are listed apart and left out of the rankings. Questions that name you or a rival are measured separately, because naming a brand in the question decides who appears in the answer. This definition is version 2.0.0, in use since November 9, 2026; earlier points on the trend were recounted with it.

How are changes reported?

A change from one run to the next is calculated only on the prompts both runs asked. When the prompt set changes, the trend is marked on the date it changed. Each method is versioned, and every run is stamped with the version it used, so a definition cannot change without notice.

What does Lighthouse not measure yet?

  • Markets beyond the United States and the United Kingdom, and languages other than English.
  • Claude, Perplexity, Copilot, Grok, Meta AI and DeepSeek, until their signed-in accounts are live.
  • Repeated samples of the same prompt within one run.
  • Prompt volume from a panel of real AI conversations; prompt demand comes from search data.

Questions people ask

Is this scraping?

Lighthouse uses headless browsers to ask the engines’ public pages the same questions a buyer would, logged out, and records what the page shows. Engines with no logged-out experience are asked signed in, the way a real user would see them.

Why not use the engines’ APIs?

API answers can differ from what buyers see on the consumer page, because the page adds its own search, ranking and formatting. Lighthouse never measures through an engine’s API: measuring the page a buyer uses keeps the number close to what buyers read.

See what AI says about you.

A real Lighthouse scan of your site: your AI visibility, who is named instead, and what to fix. Free to start.

Check your AI visibilityBook a demo