> TomHow / Resource / Measurement
How to measure AI search visibility.
AI search visibility is measured by prompt-level evidence, not by a vague score. The clean version records what buyers ask, which tools answer, which brands are cited, which brands are absent, and what cannot be measured yet.
Last reviewed 2026-10-08
> Primary references
Check the platform guidance and the limits.
These are platform requirements, not measured TomHow results. Google eligibility does not guarantee inclusion. OpenAI documents search access separately from training access. Check the current guidance before changing crawler controls or buying an optimisation claim.
> Prompt set
Start with buyer questions, not vanity brand prompts.
The prompt set should reflect how a real buyer researches the category: shortlist prompts, comparison prompts, audit prompts, reporting prompts, and problem prompts.
- Buyer shortlist prompts
- Competitor comparison prompts
- Audit and reporting prompts
- Problem and fix prompts
- Industry-specific prompts
> Platform split
Record each tool separately because they behave differently.
ChatGPT, Claude, Gemini, Perplexity, and Google AI Overviews do not expose the same answer formats, sources, or reporting data. Preserve the split instead of blending everything into one score.
- ChatGPT capture with mode and citations recorded
- Claude capture with answer mode and source availability recorded
- Gemini browser or subscription capture kept separate
- Perplexity capture with answer links and source pages recorded
- Google AI Overview availability logged separately
> Citation scoring
Separate brand mentions from cited URLs.
Record brand mentions and owned-domain source links separately. A competitor source link shows that source appeared in the observed answer; it does not prove a lost sale or replacement. A completed source-bearing answer without the agreed domain records absence for that capture. Failed and source-unavailable captures remain unavailable.
- Brand mentioned
- Brand cited with URL
- Competitor mentioned
- Competitor cited with URL
- No brand cited
> Source types
Track the sources that shape the answer, not just the final brand list.
Classify the exposed source URLs as research, comparisons, service pages, community discussion, profiles or documentation. Open the URL and check what it supports. Source patterns suggest what to investigate; they do not establish that copying a source type will earn inclusion.
- Agency service pages
- Tool comparison pages
- Research and benchmark pages
- Community discussions
- Technical docs and policies
> Corroboration check
Check the public facts without assigning a trust score.
A weak brand footprint can have a decent website and still be absent from AI answers. Measurement should check whether the same entity is repeated on the site, in trusted profiles, and in independent sources.
- Owned evidence: audit page, resource pages, proof notes, report examples, schema, and llms.txt
- Third-party evidence: profiles, directories, reviews, partner mentions, expert posts, and public references
- Description consistency: same service label, same audience, same offer, same founder or company identity
- Source eligibility: pages are crawlable, indexable, text-readable, and internally linked
- Movement proof: rerun the same prompts after evidence ships before claiming progress
> Google AI Overviews
Do not pretend Search Console is a citation tracker.
Google documents AI feature traffic in Search Console and its current generative guidance describes additional eligibility and performance reporting. Check what the actual property and supported API expose. Aggregate search performance is distinct from a preserved prompt-by-prompt answer and source table; an unavailable report is not a measured zero.
- Search Console is useful but limited
- AI Overview answer text needs separate capture
- Citation lists need manual checks or a SERP provider
- Platform gaps should be logged
- Traffic and citation evidence should not be mixed
> Technical access
Measure whether AI systems can reach the right pages.
Check crawl access, indexation, snippet controls and whether the main answer is readable text. Google requires an indexed page eligible for a search snippet to show it as a supporting link in AI Overviews or AI Mode. Other providers have their own retrieval controls. A technical pass establishes a checked condition, not selection or citation.
- Robots and CDN rules checked
- Important content available as text
- Schema matches visible copy
- PerplexityBot and Perplexity-User policy reviewed where relevant
- llms.txt treated as a map, not a ranking lever
> Trend rules
Do not overread tiny movements.
AI answers vary. A clean measurement system uses a stable prompt set, records capture conditions, and avoids claiming progress from one lucky answer.
- Same prompt set rerun
- Same country and mode noted
- Small samples marked as directional
- Prompt changes logged
- Monthly narrative tied to shipped work