Every measurement published in the AI Search Index separates three outcomes that are routinely conflated: a source being cited, a brand being mentioned, and a brand being recommended. They are different events with different commercial meanings and different levers, and collapsing them into one “visibility” figure produces claims no one can act on. This page records the decision rules blimpp applies so that any published number can be interpreted, challenged or approximated by a third party.
The three outcomes
Citation. A source is cited when the response presents it as a linked reference used for the answer, in whatever citation format the engine renders. The unit is the domain or URL, not the brand. A citation is recorded whether or not the brand behind the domain is named in the answer text.
Mention. A brand is mentioned when it is named in the answer’s text, regardless of sentiment, context or linking. Negative namings count as mentions. Namings inside comparative lists count once per brand. A mention is recorded whether or not any of the brand’s pages are cited.
Recommendation. A brand is recommended when the answer advises it for the intent of the prompt: it is proposed as a choice, listed among options the user should consider, or endorsed for the stated need. Descriptive namings without advisory force are mentions, not recommendations. A brand named only to warn against it is a mention with negative sentiment, never a recommendation.
Edge-case rules
Hedged advice such as “you could consider X” counts as a recommendation; advisory force does not require enthusiasm. A brand recommended inside a quoted or summarised third-party opinion counts as a recommendation, since the answer chose to carry it. Category examples with no advisory force, such as “brands like X exist at this price point”, count as mentions. Where an answer both recommends and cautions about the same brand, both a recommendation and a negative-sentiment mention are recorded. Ambiguous cases are logged with the response text so classifications can be audited.
Recording rules
Each response is recorded with: engine, date and time, locale, prompt text, run number, full list of cited URLs, all brands mentioned, all brands recommended, and sentiment where relevant. Percentages are always published with their denominators. Outcome rates are never averaged across engines without also being reported per engine, since engines differ enough that a blended figure conceals more than it reveals.
Why the separation matters
The three outcomes respond to different work. Citation problems are usually access, indexation and extractability problems on owned pages. Mention problems are footprint problems across the sources engines read. Recommendation problems are evidence and consensus problems in how those sources talk about the brand. A brand measured on one blended number cannot tell which of the three it has, and typically fixes the wrong one.
Related methodology
Author: Harpal Singh · Last reviewed: 7 August 2026