Official measurement contract
How GeoCheckTool measures AI-search visibility
In this product, GEO means generative engine optimization. It does not mean geographic information systems, IP geolocation, geological testing, or geo-blocking. GeoCheckTool observes how explicitly named AI and search surfaces answer real buyer questions about a business, then preserves the evidence and uncertainty needed for diagnosis and later verification.
What one observation records
Each sample belongs to a confirmed business profile, a frozen buyer question, a named provider or search surface, a timestamp, and a method version. GeoCheckTool keeps the raw answer, literal brand mention, stance, factual verification, cited URLs, provider outcome, usage, and retry lineage as separate fields. A reachable URL is not automatically treated as supporting evidence, and a brand mention is not automatically a recommendation.
Current provider, surface, and fidelity boundaries
GeoCheckTool reports the route it actually observed. It does not rename an API or search-result observation as a consumer application result.
| Product mode | Observed route | Fidelity | Boundary |
|---|---|---|---|
| Quick and Formal | Perplexity Sonar through OpenRouter | RAG/API observation with citations | Not the Perplexity consumer application |
| Quick and Formal | Google AI Overview observed through SerpApi | Search-result surface observation | Not a Google API; a valid no-surface result is possible |
| Formal | OpenAI gpt-4o-mini through OpenRouter server web search using Exa | Web-grounded model API observation | Not ChatGPT Search or a signed-in OpenAI consumer interface |
Sampling contract
Quick selects up to 4 confirmed questions and uses n=3 repeated samples per selected question and available surface.
Formal costs 200 credits per batch, covers up to 8 customer-selected questions, and uses n=5 repeated samples per selected question and available surface.
Repetition exposes variation; it does not create a universal score or a stable rank. Sample size, coverage, failures, locale, access mode, and method version remain part of the interpretation.
Outcome states stay separate
Answered
An answer body was returned and preserved for the sample slot.
No surface
The requested optional search surface was legitimately absent.
Not run
The slot was not executed and is not converted into a zero.
Timeout or rate limited
The provider did not complete the slot within its observed boundary.
Invalid or provider failed
The response failed validation or the provider route failed.
From observation to verification
1. Confirm facts and questions
The customer controls the business facts and complete buyer-question bank used by the run.
2. Preserve repeated observations
Every slot keeps its answer, citations, provider outcome, sampling lineage, and comparability boundary.
3. Diagnose evidence gaps
Accepted findings link to the exact question, answer span, citations, and applicable confirmed facts.
4. Prepare reviewable actions
Website repair and source-building packages remain customer-controlled; GeoCheckTool does not auto-publish.
5. Verify public execution
A separate read-only check can record whether an intended public action is actually observable.
6. Run a frozen retest
Later comparisons reuse the same confirmed questions and method when the surfaces remain comparable.
Interpretation and non-guarantee boundary
Publication, crawling, indexing, ranking, AI mention, AI citation, recommendation, conversion, and business improvement are different lifecycle states. GeoCheckTool does not guarantee any of them.
One observation does not establish stable behavior. A later change can support temporal association or mechanism evidence, but it does not by itself prove that one action caused an AI-answer change.
Method FAQ
Is GEO here the same as GIS or IP geolocation?
No. GEO here means generative engine optimization: improving the clarity, evidence, and verifiability that AI answer systems can use.
Does GeoCheckTool reproduce signed-in consumer apps?
No. The current product reports the API or search-result routes listed above and keeps consumer-interface claims separate.
Does a missing optional surface count as zero?
No. No-surface, not-run, provider failure, and answered states remain distinct so partial coverage is not presented as complete.