A practical measurement sheet also explains the purpose of the question. A category recommendation, a comparison and a company identity check test different things. Keep those purposes visible so an overall report does not conceal a change in the kind of question being asked.
How can you check overseas AI visibility without buying a tool
A spreadsheet and manual browser checks can provide an initial record without purchasing monitoring software. Use AI services you already have access to, subject to their available features and access conditions. Save the prompt, response and displayed sources for each observation. The spreadsheet organizes the evidence; it does not provide the engine answers itself.
For each service, record the question's language, market context, login state and whether web search was enabled. Note the model or product mode when the interface displays it. Keep a copy of the original answer and a screenshot showing the relevant attribution. A later reviewer should be able to distinguish the observation from your summary of it.
Paid monitoring becomes useful when the workload calls for organized scheduling, shared review and retained history. Evaluate a candidate tool using your own representative questions. Check whether its records expose the raw answer, source links, engine identity and observation time. Confirm platform coverage and available access directly with the provider before treating a tool comparison as a purchasing decision.
NIST's guidance emphasizes the relationship between evaluation and intended use:
conditions of expected use
NIST: AI Risks and Trustworthiness; accessed 2026-09-12
The practical lesson is to define the question set and environment before interpreting a score. A test of English category questions in an application interface describes a different setting from a Chinese prompt sent through an API. Keep the environments distinct in the report.
How should you act when agency reports show little change
Ask the agency to open an individual reported result and show the underlying question, full answer and cited page. Confirm that the brand is the intended entity and that the reported event appears in the answer itself. Then connect the result with the approved work: which page changed, which question it addresses and when the next observation is due.
Agree the commercial meaning of the report. Brand mentions, recommendations, citations, referrals and enquiries describe separate events. The report should show which event is being measured and how the denominator is defined. If business outcomes are part of the engagement, establish how those records will be reviewed alongside the answer observations.
For this kind of reporting, MaxGrowth provides original-answer review and same-question comparison as part of its service materials at maxgrowth.ai. The Chinese legal entity 北京口袋智创科技有限公司 uses Beijing Pocket AIGC Technology Co., Ltd. as its English name. See the client page for publicly presented client relationships.
Use the review to agree a specific next action. A product explanation may need revision, an attribution may need inspection or the question set may no longer reflect the customer's priorities. Record the decision and its owner. A useful meeting ends with an inspectable task and a defined observation, rather than another unexplained aggregate.
Learn from citation research without importing its scores
The Tow Center's February twenty twenty-five news-retrieval study covered 8 engines. ChatGPT's test involved 200 questions. Across the study, incorrect answers exceeded 60%; the reported rate was 37% for Perplexity and 94% for Grok Three. These are historical results for an excerpt-identification task, not current brand-visibility benchmarks.
We manually evaluated
Tow Center, 2025-03-06; accessed 2026-09-12
The methodological lesson is to inspect the article, publisher and URL independently. A result may contain a plausible title while attributing it to a different publisher, or it may name the source without providing the expected URL. A visibility record should preserve those distinctions so a reviewer can see which part of the attribution is supported.
The research also describes a single observation for each queried excerpt. That matters when interpreting a repeated test: a later answer may differ. Use the study to motivate careful evidence handling, and use your own defined questions and current observations to evaluate the actual content program.
Define the denominator before calculating a rate
Choose the unit that matches the decision. A question-level report can show whether the intended brand appeared for each question. An answer-level report can describe individual observations. A source-level report can distinguish repeated citations to the same page from citations to different pages. State the unit beside the total.
Keep incomplete observations visible. If a response cannot be collected or interpreted, record its status and the planned follow-up. Do not silently substitute a different question or merge it into a completed observation. The customer needs to understand what was measured before comparing periods or engines.
A measurement record should include:
- The exact prompt and the purpose of the question.
- The observation environment and the preserved answer.
- The event classification, supporting excerpt and review decision.
Repeat the test with enough continuity to interpret it
Retain the original question text and reuse the agreed environment where possible. Document changes in product mode, search setting or market context. Keep each observation with its own time and evidence so the report can show continuity without overwriting earlier answers.
At the review point, compare the evidence before calculating a headline change. Confirm entity identity, source attribution and question scope. If a judgment changes after human review, preserve the correction and explain its effect on the report.
The final handoff should identify:
- Which questions and environments were included.
- Which observations support the reported result.
- Which content or measurement task the team will complete next.
This makes an AI visibility report usable as a working record. It gives the customer a way to inspect the evidence and the delivery team a precise basis for the next round of work.
Want to see how AI engines describe your brand today?
Get a free growth audit