From the first check to a defensible result.
A practical guide to the live workflow, with explicit definitions for the numbers you see.
1. Run a website readiness audit
Enter a public HTTPS URL, your brand and industry, and a country/language scope. You may include up to two more pages on the same origin. Each page has its own findings. The main readiness score always belongs to the primary URL.
Inspect the evidence beside each finding. A readiness score checks HTML signals and robots.txt rules; it does not query ChatGPT or prove that a crawler visited your site.
2. Save questions and observed answers
Use the prompt workspace to save questions with brand, engine, country and language. In Evidence Lab, enter the question and an answer you actually observed, along with the observation time and any share URL. Keep the original answer and source links intact.
Suggested prompts are editorial templates. They are not query-volume data or a transcript of real customer searches. Submitting evidence records what the collector supplied; HowAICite does not certify the authenticity of a pasted answer.
Metric dictionary · version 1
| Metric | Definition | Boundary |
|---|---|---|
| Readiness score | Passed checks / known checks × 100. | Ten equally weighted diagnostic checks. Unknown checks excluded. |
| Brand mention | The tracked name appears in an answer. | Entity ambiguity requires review; a mention is not a recommendation. |
| Own-domain source | A recorded source URL belongs to the brand's domain. | URL presence alone does not prove that it supports a specific claim. |
| Recommendation rate | Human-reviewed positive selections / eligible reviewed answers. | A research measure; not automatically inferred from mention counts. |
| Source frequency | Distinct observed answers linking to a source domain. | Repeated links in one answer count once; submitted samples are not market share. |
| Comparison scope | Question, engine/surface, country, region, city and language. | Differences in collection method, model or time can explain changes. |
| Coverage | Successful collected answers / planned collection cells. | Failures and missing answers must remain visible. |
Preserve the observation, not just the score
Retain exact prompt text, brand, source URLs, answer, platform, model when exposed, requested market, location provenance, language, collection method and timestamp. OpenAI API runs and ChatGPT consumer searches belong to different cohorts. Model names may be unknown for consumer interfaces; record that explicitly.
Private workspace records stay private. The public research pages use their own explicitly published files. Opting into a future aggregate benchmark does not authorize publishing a customer's raw answer or brand.
Exports and account access
Reports export as JSON or a browser print/PDF. Citation insights and the intelligence ledger provide CSV exports from the workspace. Account data can be exported from account settings. Store your one-time recovery code securely; email resets, team invitations and SSO are not available.
Workspace endpoints require their owner's session. They are not a public API or a machine-to-machine integration contract. Public API keys and MCP integration are planned.
Diagnose an empty report
- No saved observations: record an answer before expecting source trends.
- Provider not active: check coverage; audits remain available.
- No citation found: inspect the raw answer. No linked URL does not imply the brand was not recommended.
- Changed market: compare only matching scopes and keep the original observation.
- Too few samples: report counts and limitations; collect more before making a trend claim.
Start with your own evidence.
Inspect a website, save an observed answer, and decide what to improve.