Planting the brand in discovery prompts
A brand-named question tests recognition or accuracy. It cannot prove that a neutral buyer would discover the brand.
Measurement framework
How to measure AI search visibility
Build a controlled benchmark around real buyer questions, preserve the answer-level evidence, and keep visibility, citations, competitors, accuracy, and coverage distinct.
Start with the measured surface
Visibility means appearing in the answer—not merely being crawlable
AI search visibility measures whether a brand appears, how prominently it appears, and which sources support the answer when an AI system responds to relevant questions. A technical readiness check can show that a site is accessible; it cannot prove that a provider actually mentioned the brand.
Define whether the program is measuring answer-level AEO, broader cross-engine GEO, or both. The AEO vs GEO comparison maps each surface to the right metrics.
Seven-step protocol
Make the test repeatable before collecting the first answer
- 01
Define the market
Write down the category, audience, use cases, constraints, country, language, target brand, claimed domain, and selected competitors.
- 02
Choose buyer questions
Cover category discovery, use-case fit, comparisons, risk, and proof. Keep neutral discovery questions separate from brand-named diagnostics.
- 03
Freeze provider conditions
Record provider, model identifier, web-grounding requirement, locale, collection date, and any exclusion rules.
- 04
Collect answer evidence
Preserve the exact question, full answer, source URLs, timestamp, target and competitor mentions, prominence, and completion state.
- 05
Apply eligibility rules
Exclude failed or unsupported answers from relevant visibility denominators and report them transparently in coverage.
- 06
Calculate component metrics
Report mentions, prominence, share of voice, citations, accuracy, and coverage before considering a composite score.
- 07
Rerun like for like
After meaningful work and recrawling time, reuse the same questions and conditions. Compare the underlying answers and sources, not only the headline score.
Metric dictionary
Use denominators that match the question
| Measure | Transparent formula | Decision it supports |
|---|---|---|
| Discovery visibility | Neutral eligible answers mentioning the target ÷ neutral eligible answers | Can buyers discover the brand before naming it? |
| Prominence | Lead or shortlist mentions ÷ answers that mention the target | When present, is the brand central or incidental? |
| Share of voice | Target mention events ÷ target plus selected-competitor mention events | Who occupies the shortlist across the same questions? |
| Owned citation rate | Eligible answers citing the claimed domain ÷ eligible answers | Does the brand's site supply source evidence? |
| Accuracy | Accurate brand descriptions ÷ target-named answers | Is the brand represented correctly after it is named? |
| Coverage | Eligible grounded answers ÷ planned answers | How much of the benchmark produced usable evidence? |
A minimum viable protocol to save with every score
Exact question set and question type
Target, claimed domain, and selected competitors
Provider and model identifiers
Grounding and eligibility rules
Country, language, and collection timestamp
Answer, citation, and failure evidence
Common measurement errors
Avoid scores that look precise but answer the wrong question
Treating all source links as owned citations
A grounded answer can cite third-party pages without citing the target domain. Store domains and classify ownership explicitly.
Counting provider failures as brand misses
Collection failures depress visibility for reasons unrelated to the brand. Put them in coverage.
Comparing unlike scores
A different question set, model, market, grounding rule, or formula changes the measured object.
Evidence notes
Primary sources behind the claims
The formulas here are transparent product methodology, not an industry standard. Preserve the raw questions, answers, sources, model identifiers, failures, and collection date with every result. Sources reviewed August 20, 2026.
- Google Search: AI features and your websiteGoogle says its existing SEO fundamentals remain relevant to AI Overviews and AI Mode and that inclusion is not guaranteed.
- Bing Webmaster Tools: AI PerformanceMicrosoft distinguishes citations and cited pages in AI answers from rankings or authority claims.
- 100 Questions benchmark methodologyDefinitions for eligibility, coverage, mentions, prominence, citations, competitor share of voice, and comparable reruns.
- 100 Questions 2026 AI Visibility Index protocol and evidenceFirst-party frozen study with protocol, question set, answer evidence, source files, hashes, and explicit limitations.
Run the controlled version
Measure 25 buyer questions across four AI providers.
The AI search visibility tool preserves 100 planned answers, citations, competitors, coverage, and five prioritized actions.
Contextual next steps
Connect this resource to answer evidence
Use the next resource that matches the decision at hand: define the program, inspect the method, or collect a comparable cross-provider baseline.
01
Measure mentions and citations
Run a frozen 25-question benchmark across four providers and inspect the underlying answer evidence.
02Choose the AEO or GEO lens
Map answer selection, brand understanding, SEO, and cross-model measurement to the right program.
03AI prompt tracking spreadsheet
Preserve provider conditions, answers, citations, competitors, and comparable reruns.
Common questions
Use the resource without overstating the result
What is the best metric for AI search visibility?
There is no single best metric. Start with discovery mention rate, then keep prominence, competitor share of voice, claimed-domain citations, factual accuracy, and provider coverage separate. Each metric diagnoses a different problem.
How many prompts are needed to measure AI visibility?
Enough to cover meaningful buyer decisions without pretending to be statistically representative. A balanced 20-question neutral discovery set plus five brand diagnostics is practical for a directional benchmark. Record the limits and compare only identical reruns.
Should failed AI answers count as brand misses?
No. A provider failure or answer that does not meet required grounding conditions belongs in coverage. Counting it as a brand miss confuses collection reliability with brand visibility and biases the result.
Can AI visibility scores be compared across tools?
Only when the question set, providers, market, collection date, grounding rules, eligibility criteria, and formula match. Similar-looking scores from different protocols are not necessarily comparable.
Apply this resource
Put how to measure ai search visibility into a measured workflow
Preserve the questions, conditions, answers, citations, and failures so the next decision rests on inspectable evidence.