Know what was measured. Know what it means.
Every conclusion should lead back to a real observation. Citestra keeps prompts, provider differences, citations, and limits visible so you can act on evidence with confidence.
Explore our methodologyHow we turn observations into useful evidence
Our rules make the scope and limits of each measurement explicit.
What we measure
Citestra observes responses to approved buyer-intent prompts through configured OpenAI, Google Gemini, Perplexity, and Anthropic Claude APIs. Each observation records its provider, model, methodology, prompt version, locale, and time. API observations may differ from the providers’ consumer applications.
Prompts and repetition
We crawl public pages, draft a brand profile, and propose category, problem, use-case, feature, comparison, alternative, trust, and location prompts. You review prompts before scanning. We run scheduled repetitions because answers can vary; a single response is limited evidence.
What the rates mean
Mention Rate measures explicit brand or confirmed alias presence. Recommendation Rate requires positive recommendation evidence under the current analysis rules. Citation Rate counts provider-native citations to confirmed first-party domains. A URL only in generated prose does not count as a native citation. Each rate uses eligible successful observations and shows its sample size; failed and skipped measurements stay visible separately.
Why providers differ
Providers use different models, search tools, grounding metadata, and citation formats. Search being configured does not prove it ran for every answer. We preserve provider-level results and source semantics rather than treating all URLs or responses as identical.
From evidence to Fixes
Citestra combines successful observations, cited sources, competitor appearances, and site crawl findings into diagnostic cases. Fixes identify a target, evidence, proposed action, and metric to watch. When evidence is too weak for a specific action, no Fix is generated.
Comparing periods
Change Sets compare repeated cohorts after implementation. Prompt, provider, model, methodology, and locale changes can make cohorts only partly comparable or incomparable. A change in an observed rate is an association, not proof that a Fix caused it. Citestra does not guarantee placement.
Technical guidance
Crawler access is tested by user agent. We distinguish OAI-SearchBot from other crawlers. llms.txt is optional and experimental; accurate Schema.org markup should describe the real page. No special markup guarantees an AI citation.
Follow the evidence in an illustrative workflow.
Explore the demo workspace or start with a bounded public site audit.