How the numbers get made

The pipeline

• Weekly pulls: for each customer we run a fixed panel of category questions against AI engines via the DataForSEO LLM-mentions API — the same commercial substrate the enterprise tools use, at a fraction of the retail markup.
• What we measure: mention share (how often your brand appears in answers vs competitors), question winners (which prompts you win/lose), and cited sources (which pages earn the citations — the actionable part).
• Dated + diffed: every report carries its pull date and diffs against your history — "you gained/lost these questions this week," not a static snapshot.
• Real sample: the sample report is an actual pull (2026-07-18) for a real brand, not a mockup — what you see is exactly what subscribers get.

Definitions — exactly how each number is computed

These mirror the engine (engine/citewatch.py); if the code changes, this list changes with it.

• Watched set: your domain plus the competitor domains you name (the sample tracks 3 domains on ChatGPT). Domains are normalized (lower-case, www. stripped) before matching.
• Mention share: for the chosen platform, each watched domain's mention count in the LLM-mentions aggregate, ranked by mentions, then by AI-ask volume as the tie-break. A watched domain absent from the aggregate is listed at 0 — never dropped, so "you don't appear" is a visible row, not a silent gap.
• Question winners: for each of the top questions (by AI-ask volume; 8 per pull in the sample), the "winner" is the first watched domain named in the engine's answer text — or, if none is named in the prose, a watched domain that appears among the answer's cited sources. "Present" lists every watched domain the answer names or cites, so you can see second-place mentions too.
• Cited sources: the top 8 non-watched domains by mention count in the same aggregate — the pages actually earning citations in your category, i.e. the outreach and content targets.
• Trend sparklines: the last 12 months of AI-ask volume for each question, straight from the API's monthly series — no smoothing, no interpolation.
• Diff vs previous pull: per-domain mention deltas against the last stored run for the same customer; the first report has no diff and says so.

What a tracker can't promise

AI answers are non-deterministic and personalized; a tracker measures the trend, not a guarantee. We never promise rankings. If a vendor does, ask for their pull methodology.

Who runs this

Independent and solo-operated. Questions, corrections, or a category you want tracked: hello@answerfootprint.com.