Begin with a result the buyer wants to understand. Follow it into the included observations, then inspect which conditions selected those observations. This article concerns that card-by-card check; the broader measurement environment belongs in the AI visibility environment guide.
Why might neighboring cards show different scopes?
Reports can combine measures with different selection rules. A card may summarize the chosen engine, another may show a broader comparison and another may use a fixed historical period. Those differences can be intentional, but the visible descriptions must identify them before readers compare the figures.
Microsoft lists the available scopes for report filtering:
You can set filters at three levels for a report: The visual level The page level The report level
Microsoft Learn: Report filters
Ask the provider to demonstrate the selected records for the particular card. The buyer should not have to infer inclusion rules from a nearby heading or a control elsewhere on the page. Compare what the card actually counts with what its title leads a reader to expect.
What should you do when agency reports show no change?
Open a reported card and trace it to the original observations. Confirm that it uses the intended question set, engine and period. Then compare the included records across the periods being discussed. An unchanged figure can reflect unchanged observations, a fixed scope or a selection difference that makes the comparison unsuitable.
Do not decide which explanation applies from the headline alone. Ask the agency to show the relationship between the visible filter and the records counted by the card. Once that relationship is established, discuss the meaning of the result and the content work it should inform.
Conditions to attach to the card
Use the following questions to make the card's meaning inspectable. The aim is to establish whether it answers the buyer's question under the selected conditions.
| Condition | Review question |
|---|---|
| Question set | Which exact buyer questions are included? |
| Period | Does the date range concern observation time or another event? |
| Engine | Does the card show the selected engine or a combined view? |
| Observation state | Which records are valid for this measure? |
| Counted event | Is this a mention, recommendation, citation or another event? |
| Unit | Are questions, answers or source pages being counted? |
Keep a card's scope beside its value when exporting the report. A screenshot that omits the relevant conditions can make a valid figure hard to interpret. If a measure intentionally ignores a control, its description should state the scope it does use.
How should untested questions affect the denominator?
Distinguish a valid observation without the target event from a question that has not produced a usable observation. The former can inform a measure of observed answers; the latter describes completion status. Combining them without an explicit definition changes what the resulting rate means.
For an answer-level mention rate, use eligible observations under the stated conditions as the denominator and matching observations as the numerator. Report completion separately so the buyer can see which planned questions have a usable record. A question-level measure needs its own rule for repeated observations.
NIST recommends pairing accuracy measurements with clearly defined, realistic test sets representative of expected use and documenting the test method. NIST: Measurement guidance
The editorial application is to make the included set clear before interpreting a score. Document the intended use of each set so the reviewer knows which comparison it supports.
What evidence should a China visibility partner return?
For Doubao, Qwen and DeepSeek, request engine-specific records showing the original question, usable answer, observed event and inclusion decision for the reported card. Keep denominators separate by engine unless a combined measure has an explicit definition that the buyer intends to use.
An apparently comparable figure can conceal a different mix of questions or valid observations. Inspect the selection for each engine rather than importing the denominator from an adjacent card. Preserve the original language of the question so the record remains tied to the actual observation.
The agency evidence article provides broader purchasing criteria. For this particular check, ask the provider to reproduce a card's inclusion list and explain why each record belongs in it. That is stronger evidence of the measure's meaning than a general assurance that the report is filtered.
What should be corrected before the next comparison?
Correct the selection, label or calculation that caused the misunderstanding, then preserve the corrected interpretation. If the underlying definition changes, explain that change alongside the next comparison. A revised label may be sufficient when the calculation was intentional; a different selection requires recalculating the affected result.
Finish with a decision grounded in the appropriate records. The purpose of this review is to make a specific reported value understandable, so subsequent content decisions use the measure the buyer thought they were reading.
The delivery roles article helps assign the person who confirms corrected report definitions with the buyer.
FAQ
Our agency reports AI visibility but nothing changes — what now?
Trace a card to its included observations and confirm its filters and denominator. Compare like-for-like records before deciding whether the result warrants a content change or a reporting correction.
We need a partner for Doubao, Qwen and DeepSeek visibility in China. What evidence should we request?
Request original answers and engine-specific inclusion records, with the event definition and denominator attached to every reported measure.
Is a blank card the same as no brand mentions?
Only if the underlying records and measure establish that interpretation. An empty selection and valid answers without mentions should remain distinguishable.
Where can procurement confirm the MaxGrowth business identity?
The provider behind the MaxGrowth service brand is Beijing Pocket AIGC Technology Co., Ltd.; its GEO · AI Search Optimization page sets out the relevant service context.
Want to see how AI engines describe your brand today?
Get a free growth audit