AI Visibility Reporting
Rank tracking answers where you sit. AI visibility reporting answers whether you get mentioned at all. Ivan Sp reports on AI visibility by tracking query coverage and recommendation frequency inside AI answers. Ranking first and being the business an AI recommends are measured differently, and a standard rank tracker reports neither.
What Gets Measured
AI visibility reporting tracks two figures over time. Query coverage records how many of a defined query set produce an answer mentioning the business. Recommendation frequency records how often the business appears when the answer names specific providers.
Both figures get recorded per answer engine, because ChatGPT, Perplexity, and Google AI Overviews draw on different sources and return different results for the same question.
Building the Query Set
The query set gets built from how customers actually ask, which is closer to conversational phrasing than to classic keyword strings. “Who does commercial security monitoring in Edmonton” belongs in the set, and “commercial security Edmonton” belongs in a rank tracker.
Queries get grouped by intent so the report shows where visibility is strong and where it is absent. A business named consistently for its main service, and never for an adjacent one, has a specific content gap rather than a general visibility problem.
What the Monthly Report Contains
The report covers five items:
- Query coverage and recommendation frequency, compared against the previous month
- The queries where the business is named, and the queries where competitors are named instead
- Which pages get cited, where the answer engine names a source
- Factual errors in how answer engines currently describe the business
- The content or authority work indicated by the gaps
The fourth item produces the most urgent findings. An answer engine stating an incorrect service area, an outdated price, or a service the business no longer offers is actively costing enquiries.
Honest Limits of the Measurement
Answer engines return different results to different users and change their outputs frequently. A single check is a sample rather than a measurement, so figures get recorded across a fixed query set on a fixed schedule to make the trend meaningful.
No public API reports “share of AI answers” the way a rank tracker reports position. The numbers in this report come from running the query set directly, which is why the set stays fixed once agreed. Where visibility gaps trace back to weak source coverage, the fix runs through citation authority building on platforms answer engines read.
Sample Your Own Coverage Twice
I suggest a test that shows the variance immediately.
- Write down five questions a buyer would ask before choosing a provider like you.
- Ask all five in an answer engine and record whether you are mentioned.
- Wait an hour, open a fresh session, and ask the identical five questions.
- Compare the two sets of results.
Most people are surprised by how much the two runs differ. That difference is exactly why a single run should never be reported as a figure.
Where the Standard Reporting Advice Breaks Down
Would your AI visibility number look the same if it were measured twice in one afternoon?
- The usual advice
- Add AI visibility to the monthly report, since clients want to know how they appear in AI answers.
- What holds up in practice
- A single monthly observation of a system that varies by session is closer to an anecdote than a measurement. The reportable unit is a stable query set sampled repeatedly, with the variance stated rather than hidden.
Picture a report claiming the business appears in seventy percent of AI answers, based on one run through twenty queries in a single session.
A second run the same afternoon returns a different figure. Nothing changed on the site, and the number in the report was never stable enough to compare against next month.
Ivan Sp samples the same query set repeatedly and reports the range alongside the figure. A trend across months means something, and a single reading does not.
A precise-sounding number from an unstable measurement is worse than no number. Decisions get made on it, and the next month’s reading contradicts them for no reason.
A report you cannot reproduce is a report you cannot act on. Stating the variance is what makes the trend usable.
Four Things Commonly Assumed About AI Visibility Reporting
Which of these describes AI reporting you have seen?
I meet these four assumptions whenever AI visibility is being reported.
| Commonly assumed | What actually holds |
|---|---|
| AI visibility can be measured like a ranking. | Rankings are comparatively stable. Answers vary by session and personalisation, so the honest unit is a sampled trend with a stated range. |
| One run through a query set is a measurement. | A single observation of a varying system is an anecdote. Repeated sampling of a stable query set is what produces a comparable figure. |
| Appearing in an answer is the goal. | Being described accurately matters as much. A mention carrying a wrong claim is worse than no mention at all. |
| The query set should grow each month. | A changing set makes month-to-month comparison meaningless. The set is fixed first and extended deliberately, with the change recorded. |
| Traditional rankings can be retired. | Organic search still carries the majority of traffic on most sites. AI visibility reporting sits beside it rather than replacing it. |
The number you put in front of a client has to survive being checked. Stating the range is what keeps the whole report credible when somebody runs the query themselves.
A Worked Example: One Query Set, Two Runs
Consider twenty queries sampled twice in one afternoon. The figures below are illustrative, written to show why variance has to be reported. No client engagement is described here.
| Measure | Run one | Run two |
|---|---|---|
| Queries where the business appeared | 14 of 20 | 11 of 20 |
| Queries where it was named first | 4 | 6 |
| Answers linking back to the site | 9 | 8 |
| Answers containing a wrong claim | 2 | 3 |
| Headline figure if reported alone | 70 percent | 55 percent |
| Honest reported figure | 55 to 70 percent across two runs | Same |
I report the final row. The first five rows are real observations and none is the number on its own.
Reporting a single run is quoting the temperature from one thermometer reading. Accurate at that moment and useless as a climate record.
The wrong-claim row usually matters more than the coverage row. An inaccurate description reaching buyers is a problem no amount of coverage offsets.
How This Sits Beside Traditional Reporting
AI visibility reporting supplements classic search reporting rather than replacing it. Clicks, impressions, click-through rate, and average position still come from Google Search Console, with branded terms filtered out so unbranded performance is visible on its own.
The two report types move independently. A page can gain AI citations while losing ranking position, or hold position while disappearing from AI answers entirely.
How the Reporting Runs
I fix the query set before the first measurement, since a moving set cannot be compared.
- Build a query set covering the questions buyers genuinely ask before choosing.
- Fix the set at roughly twenty to forty queries, and record it as the baseline.
- Choose the surfaces to sample, based on which ones the audience actually uses.
- Sample each query at least twice per reporting period, in separate sessions.
- Record whether the business appeared, where it sat, and whether the answer linked back.
- Record any factual error in how the business was described, verbatim.
- Report coverage as a range across runs rather than as a single figure.
- Correct the underlying facts wherever an answer carried a wrong claim.
- Track the trend across months rather than comparing individual runs.
- Extend the query set only deliberately, and record the change alongside the figures.
How This Engagement Is Scoped
I scope visibility reporting by query set size and by the number of surfaces sampled.
- What arrives
- A fixed query set, a monthly report with ranges, and a list of factual errors found.
- Typical duration
- Ongoing, delivered monthly after a two-week setup.
- What the client supplies
- The buyer questions worth sampling, and the canonical business facts.
- Included
- Query set construction, repeated sampling, range reporting, and error correction at source.
- Not included
- Any promise of coverage, and any claim that a figure is precise.
- What changes the estimate
- Several surfaces, several product lines, or a large query set.
- How it is quoted
- After a short scoping call, once the query set size is agreed.
Tools and Primary Sources
The references below cover the surfaces sampled. Each opens in a new tab.
- Google AI features documentation describes how AI Overviews draw on and attribute web sources.
- Search Console Performance report supplies the organic baseline this reporting sits beside.
- Schema.org Organization type carries the facts an inaccurate description usually needs corrected at source.
- Google structured data policies governs the markup used to state those facts correctly.
Common Questions About AI Visibility Reporting
Can AI visibility be tracked automatically?
Partially. The query set gets run on a schedule and the results recorded, which automates collection. Judging whether an answer describes the business correctly still requires review.
How large should the query set be?
Large enough to cover every service and intent that matters, and small enough to run consistently every month. Most engagements settle between 20 and 60 queries.
What if an AI system says something wrong about my business?
The correction runs through the underlying sources rather than through the model. Fixing the website, the structured data, the listings, and the third-party profiles changes what the model reads next time.
Discuss This Work
The first call is a free 30-minute consultation with no pitch attached. You receive an honest assessment of where the site stands and what can realistically be improved.