Asked in a real browser. Preserved with a screenshot.
Each platform uses 50 prompts and runs one cycle every 72 hours. Each prompt is executed once per platform by default. We retain the web answer, cited sources and screenshot, then review the evidence before reporting.
Version 1.1 · Last updatedWe reproduce the user path, not a convenient answer
- Freeze the prompt map: confirm 50 prompts, the 44+6 question mix, target-market language, platforms and version.
- Enter the product page: use a real browser and the normal consumer web entry point for the AI product.
- Ask on schedule: create a new batch every 72 hours and execute every prompt once per platform by default. Technical retries retain the original task identity.
- Keep screenshot evidence: retain the full answer, page screenshot, sources and sampling conditions, including failed or anomalous states.
- Review before computing: only observations that meet the report definition enter the denominator.
What exists today, and where it stops
| Item | Status | Boundary |
|---|---|---|
| Prompt scope | Supported | 50 frozen prompts per platform: 44 unbranded discovery/comparison questions and 6 branded fact checks. |
| Cycle sampling | Supported | One cycle every 72 hours from the baseline; each prompt runs once per platform by default, and technical retries do not increase the client quota. |
| Browser collection | Supported | A real browser follows the normal consumer product-page path. A one-off API response is not presented as a web result. |
| Screenshots and raw answers | Supported | Each run retains the answer, screenshot, sources, time, region, language, entry point, prompt version and anomaly state. |
| Human review | Supported | Evidence and report review are separate gates. Unverified claims remain labelled as unverified. |
| Web report and PDF | Supported | The client workspace contains the web report; formal delivery supports PDF. Public samples hide full evidence. |
| CSV | Limited | Structured outputs such as action lists can be included by delivery scope. We do not promise every raw evidence field as CSV. |
| General self-service API | Not currently offered | There is no general client-facing query or bulk-export API today. Integration requests require a separate scope. |
| Multi-user access | Project-scoped | Access is granted by project and account. Unlimited seats are not a default plan promise. |
Screenshots are project evidence, not marketing material
- Client monitoring data is used for that client’s project, reporting, review and support, not as an unauthorised benchmark for another client.
- Screenshots, answers and source records stay within controlled project access. Public samples are de-identified and omit full evidence.
- Source records retain URL, title, necessary excerpt and timestamp. Citation count is never treated as source quality by itself.
- Retention is limited to service delivery, review, legal obligations and dispute handling. Requests are handled under the privacy policy.
Successive cycles show change; they do not manufacture certainty
Answers can change with browser conditions, account, region, language, session history, model version and product UI. Like-for-like 72-hour cycles show observed change; they are not an official platform ranking or a guarantee of indexing, recommendation, traffic or revenue.