A useful AI visibility report lets the reader move from a headline rate to the exact answers behind it, then to a specific decision. The fictional sample below demonstrates counts and denominators; it is not a customer performance claim. Keep completed and unavailable records visible, explain classification choices and preserve market, language and interface conditions. For recurring reviews, show which comparable question records gained or lost a signal as well as the net change. This makes a report more useful than a score that hides the underlying movement.
Define a sample that can be repeated
Illustrative setup: 10 fixed English buyer questions, one named market, one agreed platform interface and two observations per question. The plan contains 20 observations. Each record keeps its question ID, exact text, timestamp, interface, available answer and source links.
In this example, two planned observations are unavailable and 18 completed answers can be classified. The unavailable records remain in the report. They are not treated as completed answers with no brand mention.
Use the correct denominator
| Measure | Illustrative count | Calculation |
|---|---|---|
| Availability | 18 completed / 20 planned | 90.0% collection completion |
| Brand mention | 6 completed answers mention the brand | 6 ÷ 18 = 33.3% |
| Owned-domain citation | 4 completed answers link to the agreed owned domain | 4 ÷ 18 = 22.2% |
| Recommendation | 3 completed answers recommend the brand under the stated rule | 3 ÷ 18 = 16.7% |
| Unavailable | 2 observations could not be classified | Reported separately; excluded from visibility rates |
These measures overlap. A recommendation may also mention the brand and cite a source. The rates should not be added together. Count each completed answer at most once per metric under the chosen rule.
Retain an observation record
| Field | What it preserves |
|---|---|
| Question and segment | Exact wording, audience, buying stage and question ID |
| Conditions | Market, language, interface, account context and timestamp |
| Response | Answer evidence and collection status |
| Sources | Cited URL, domain and source role where available |
| Classification | Mention, owned-domain citation, recommendation and accuracy notes |
| Reviewer | Responsible reviewer and reason for any manual correction |
A recommendation needs an agreed rule: for example, the answer presents the brand as a suitable choice for the described need. A mention in a warning or rejection is not a positive recommendation. Accuracy and sentiment can require human review.
Turn a result into a specific action
A low owned-domain citation rate might lead to a review of source pages. An incorrect product description might lead to a factual correction and a check of conflicting sources. A recommendation followed by poor inquiries might point to a mismatch in the audience or offer.
These are possible interpretations, not causal conclusions from the fictional table. The analyst should inspect individual records and compare the finding with the website and commercial evidence before choosing an action.
Preserve the basis of a comparison
- Keep the question set and classification rules stable.
- Record changes to interface, market or account conditions.
- Show planned, completed and unavailable observations in each period.
- Present a percentage-point change separately from a relative percentage change.
- Keep lead and revenue reporting outside the visibility denominator.
The sample describes the questions observed under those conditions. It is not total platform exposure, global market share or a reliable ranking position. Our managed reports use the agreed scope and actual collected evidence rather than these fictional counts.
Show how one observation becomes an action
Illustrative record, not a live result: question Q04 asks whether a fictional supplier supports multi-location deployment. The completed answer names the supplier and links to its deployment guide but does not present it as a suitable choice. Record mention: yes; owned-domain citation: yes; recommendation: no. Preserve the exact answer and source URL in an actual report so another reviewer can check that interpretation.
Suppose the linked page explains installation but leaves account permissions unclear. The proposed action is to obtain the approved permissions model and add that explanation to the relevant page. The product owner approves the facts; the publishing owner checks the live update. Do not claim that this action will cause a recommendation. It addresses a documented buying question, whose subsequent visibility can be reviewed separately.
Explain gross movement as well as net change
Hypothetical follow-up to the 18-answer example: a second comparable review also records six mentions. Four corresponding observation positions retain a mention, two lose it and two previously absent positions gain it. The headline remains 6/18, but the underlying pattern changes. Inspect whether the gains and losses concern different buying stages, questions or factual contexts before declaring the result stable.
Use this comparison only where question, platform and observation design can be meaningfully paired. Keep unavailable positions visible and show unmatched records separately instead of inventing replacements. Finish the review with a short action log: what was observed, which evidence supports the interpretation, who owns the next step and when it will be checked. The log gives the report operational value even when the headline rate does not move.
Common questions
Are these TANTU AI customer results?
No. Every count on this page is fictional and is labeled to demonstrate the reporting method.
Can we see the structure before a purchase?
Yes. This page and the deliverables page explain the structure. Any additional example must be shared with appropriate permissions and context.