Answers & Citations

What gets stored

A visibility check asks each enabled engine your prompt and writes three things in one transaction: the run (score, time), one response per engine, and one citation per tracked brand the answer named.

Answers are stored whole and uncapped. That is deliberate: truncation is irreversible, and the full text is what lets a brand added next month be detected in answers collected today (see scripts/backfill-citations.ts).

How a mention is detected

A brand matches when its name or one of its aliases appears as a standalone token — the boundary rule is “not alphanumeric on either side”, which keeps “Air” out of “Airbnb”. Brands flagged case-sensitive must match exact capitalisation, which is what makes a bare name like Bolt or Lovable trackable at all without counting every ordinary use of the word.

The highlighter and the score share one implementation, so anything shown highlighted in an answer is the same match the visibility percentage counted.

Reading an answer

• Highlights — your brand is tinted in the accent colour, tracked competitors in amber, inside the verbatim text.
• Brand chips — one per brand found, with #rank (order of first appearance) and ×n (repeat count). Being named third in a list of five is a different fact from being named first.
• Mentioned instead — on an answer that did not name you, who did get named, in order. A losing answer with no rivals means a thin market; a losing answer with three rivals means a content gap.
• Sources — the URLs the answer leaned on, with favicons. Extracted from the text at read time, so it works on historic answers too.
• Copy — per answer or all answers, as markdown with the status line and source list, ready to paste into a doc or a chat.
• Formatting — headings, lists, bold and code in the answer are rendered as such (highlights included); a Raw text toggle in the reader shows the untouched characters when you need them.
• Fullscreen & compare — every answer opens in a fullscreen reader, and you can mark two or three answers to read side by side.

Citation snippets

A citation’s snippet is a window centred on the mention, not the first 200 characters of the answer. The opening paragraph of a long answer usually does not contain the brand at all, which made the old preview read as no evidence. The whole answer is still one click away on the query page.

Comparing runs

On a query’s page, Compare runs aligns two runs engine by engine. Turning on Highlight changes shows a word-level diff in the right column; below each row you get the brands that entered or left the answer and the source URLs that were added or dropped. Verdict flips are tinted regardless of the toggle.

Answer gaps

On Visibility, the Answer gaps section is the action list:

• Prompts where you’re invisible — tracked prompts whose answers cite competitors and never you.
• Sources cited when you’re absent — the domains engines lean on in answers that skip you, with your own domains excluded. Earning a mention on those pages is the shortest path into those answers.

Sample data

Answers produced by the built-in simulator (no API key configured) are badged sample data everywhere they appear, and counted separately in data quality. They are never presented as real engine output.