Methodology
How we measure AI visibility and change.
Fixed buyer questions. Repeated answers. Recorded evidence. We measure mentions, recommendations, citations, and accuracy separately, then compare like with like.
- Fixed buyer questions
- The same question set for the baseline and the retest
- Repeated answers
- Each question asked more than once on each AI platform
- Recorded evidence
- Every answer stored with its sources and conditions
What we measure
Four measures. A clearer picture.
See whether AI names you, recommends you, cites your pages, and describes your offer accurately.
Mention
Your company is named in a valid answer.
How it is counted
Answers that name you, out of the valid answers for the questions in scope.
What it does not mean: Being named is not being chosen. A company listed among several options has a mention, not a recommendation.
Recommendation
The answer suggests your company as an option to choose, not only names it.
How it is counted
Answers that recommend you, out of the same valid answers, counted separately from mentions.
What it does not mean: A recommendation in one answer is not a ranking. Answers vary, so the pattern across repeated runs is what we report.
Citation
A source the answer links to or names as its basis.
How it is counted
Cited sources, grouped as your pages, competitor pages, review platforms, and other sources. Each count states its own base.
What it does not mean: A citation shows what the answer pointed to. It does not establish what a model read or why it named a company.
Accuracy
Whether what the answer says about your company matches the facts your team has confirmed.
How it is counted
Claims about your company, checked one by one and flagged when unsupported, outdated, or incorrect.
What it does not mean: Accuracy is judged only against facts you support. An answer that never names you has no claims to check.
Rules that keep the measures comparable
- No blended score as headline proof: each measure is reported on its own.
- Every share states its base: the valid answers, for the questions in scope, on the platforms and surfaces tested.
- Branded and unbranded questions are reported separately, because a question that names you usually gets an answer that names you.
- Answers from an API and answers from a consumer search surface keep separate labels.
- Excluded answers are listed with a reason and left out of the count. They are never counted as zero visibility.
Explore the method
From buyer questions to campaign priorities
How it works
Six steps from buyer question to campaign work.
The same steps run for a baseline and for a matched retest, so the two can be compared. Each step leaves a record, so a finding can be checked against the answers behind it.
Step 1: Define the buyer question set.
We agree the questions buyers ask during discovery, comparison, and evaluation for the products and audiences you choose. The set is fixed before testing, so every later run asks the same questions.
Step 2: Run repeated answers.
Each question is asked more than once on each selected AI platform and surface, because the same question can get a different answer from one run to the next.
Step 3: Record valid answers and exclusions.
Every answer is stored as it was returned, with the model, wording, location, and time. An answer that fails the collection rules is excluded and listed with its reason.
Step 4: Extract mentions, recommendations, sources, and claims.
From each valid answer we record whether your company is mentioned or recommended, which competitors appear, which sources are cited, and what the answer says about you.
Step 5: Verify claims against your supported facts.
Statements about your company are checked against facts your team has confirmed. Where a fact is not confirmed, we ask rather than assume.
Step 6: Prioritize gaps into Campaign work.
Gaps are ranked by the product, audience, and initiative they affect, and the highest priorities become the campaign plan your team reviews.
How we compare the baseline and retest
Matched retest
A retest only counts when it matches the baseline.
Ask the same buyer questions on the same AI services after publication. Record any differences in testing conditions alongside the comparison. Retest scope and timing are agreed for your program.
| Compared on | Baseline | Matched retest |
|---|---|---|
| Questions | The fixed buyer question set, branded and unbranded reported separately | The same questions, reported the same way |
| AI services | The selected AI platforms and surfaces, each labeled | The same platforms and surfaces, with the same labels |
| Timing | Before the campaign work | After the work is published, from the recorded publication date |
| Conditions recorded | Model, exact wording, location and language, date and time | The same conditions, with any difference disclosed beside the comparison |
What a matched retest cannot tell you.
- It shows what moved within the tested questions, not across the whole market.
- A few repeats of a question do not make a precise market estimate.
- A change after publication is an observed change, not proof that the work caused it.
How search and analytics fit into measurement
Search and analytics
Connect discovery with measured outcomes.
Matched retests, Search Console discovery, and Google Analytics conversion data come together in Results. We measure changes against the baseline and use the connected evidence to estimate impact and inform the next campaign.
- AI answer visibilitySignal Harbor recorded answers
- Mentions, recommendations, citations, and accuracy from the buyer question set, each reported on its own.
- SEO discoveryGoogle Search Console
- Organic queries, page performance, clicks, impressions, click-through rate, and average position.
- Traffic sourcesGoogle Analytics 4
- Identified AI referrals and organic traffic, connected to the landing pages visitors reach.
- EngagementGoogle Analytics 4
- Sessions and engaged visits, examined by source and landing page.
- Conversions and attributionGoogle Analytics 4 key events
- The form submissions, demo requests, and other conversion events your team tracks, alongside the recorded traffic source.
From remeasurement to impact estimates.
We compare the same buyer questions and AI platforms after publication, then examine visibility changes alongside search discovery, referral traffic, landing pages, and conversion events. Impact estimates bring those signals together with the campaign context, assumptions, and strength of the evidence. Recorded outcomes and estimates retain separate labels and measurement windows.
The evidence behind attribution. Analytics identifies AI referrals when a source is recorded. Visits without that source, including some Google AI search traffic, may remain in organic or other channels. We read conversion events alongside publication dates, campaign activity, and other business changes to keep the interpretation tied to the available evidence.
Where your team sees the evidence
Where you see it
The evidence lives in your client portal.
Clients see the evidence, the work, and the progress in the portal, where every finding stays connected to the answers behind it.
- The evidenceInsights
- The buyer questions tested, the answers recorded, the outcome for your company, competitors named, and sources cited.
- The workCampaign
- The campaign plan, the prepared content with its version history, and your decision on each version.
- The progressResults
- Work prepared, approved, and published, kept separate from the matched retest and the measures agreed for your program.
The research behind the method.
Read the research references that inform how we test, the plain language guide to AI visibility, or how the AI visibility audit works.
Your next step
Find out where AI leaves you out.
See your AI Visibility Snapshot, explore the portal in a live demo, and discuss the next move for your team.
The introductory call, AI Visibility Snapshot, and portal demo are complimentary. The company pilot and agency partnerships are paid engagements. See what the Snapshot includes.