Citino stores the full text of every answer it collects — 1,785 of them in this category in thirty days — with each brand marked where it appears and the order of naming recorded. Nothing is summarised on the way in.
The full answer text as the engine returned it, the exact prompt that produced it, the engine and the run timestamp in ET, every brand mention with its place in the naming order, and every URL the answer cited. Stored once, per answer, unedited.
Visibility, share of voice, position when named, sentiment when named, and the whole Sources ranking — all computed from the stored column, none stored in place of it. That direction is the difference between a record you can audit and a chart you have to trust.
Buyers read answers from the top. Named fourth of four is a different commercial outcome from named first, and a tool that counts mentions records both as one mention.
Order also survives paraphrase. Engines rewrite their prose constantly; the sequence in which they reach for brands is far more stable, and it is what the position metric is built from.
Parsers are wrong sometimes. They miss a mention, or count a brand named in a warning as a brand recommended. Every tool in this category has that problem; the difference is what happens next.
Citino stores the answer, not just the extraction. When the matching logic improves it runs again over everything already collected, and the history improves with it. A tool that keeps only the counts can stop making the mistake, but it cannot unmake it.
It is also why you can check us: any number opens to the answers underneath it, and you read the same text we do.
Re-parses are versioned. The reading that produced a past chart is recoverable.
Prompts cannot be edited. A prompt can be retired and a new one started, and the chart shows a hard break at the changeover with both prompts named on either side of it.
The alternative is what most tools do: let you reword the question and keep drawing one continuous line. That line is a lie with a nice slope — a trend that crosses the edit is measuring the edit.
The same rule is why 54 prompts per category is a decision rather than a starting point. Custom prompts follow it too: retiring one never rewrites the series it belonged to.
Everywhere in the product, engine text is set in the serif and everything Citino says is set in the sans. You can tell at a glance which words came out of a model and which we generated from them.
It removes the most common way analytics tools mislead: presenting their own inference in the same voice as the thing they measured. The rule holds in exports and in white-label agency reports.
Filter by engine, prompt, date, brands named, naming position or a phrase in the text — and combine them. Results are whole answers, not snippets, because a sentence taken out of an answer is how most brand-monitoring tools manufacture alarm.
The most useful query is usually the plainest: every answer that named a competitor and did not name you. That set is the gap, in the engines' own words.
These are the answers the engines gave our runs, on a fixed schedule, from a clean session. They are not what a specific buyer saw.
Personalisation, chat history and the hour of the day all move an answer, which is why the product samples daily and publishes a band rather than quoting a single answer as the state of the world. Every number built on the archive is stated with that limit attached.
→ Read the methodologyPick a category and the answers are already searchable from the first screen.