Measure mentions, citations and recommendations separately. An assistant can name your business without linking to your website, cite your page without recommending your service, or recommend a provider for one situation while warning against it for another. Combining these events into one score hides information your team needs.
Define the rules before collecting answers. A mention means the answer explicitly names your business. A citation means it links to a page on a domain you have defined as yours. A recommendation means the answer explicitly presents your business as an option to choose for the buyer's stated need. Keep the supporting sentence so another person can review the classification.
For a simple example, imagine an answer saying, “Acme and Beta both offer this type of service.” That is a mention of Acme. If the answer links to Acme's explanation of data migration, it also contains a citation to Acme. If it says, “For the small team and migration requirements you described, consider Acme,” it contains a recommendation under the rule above. Acme and Beta are illustrative names, and the example is invented to explain the categories.
Record the context around each event. A warning that a provider is unsuitable is still a mention, but it should not count as a positive recommendation. An assistant quoting an independent publisher's page about your business is a mention supported by an external source; it is not a citation to your own domain. You can track that external coverage separately when it matters.
Use the same questions for each observation period. Save the service, date, language, relevant location, question wording and whether web search was enabled. Keep branded accuracy questions separate from questions about a buyer's need. ChatGPT Search and Perplexity provide different search experiences; record their answers separately from Google's AI Overviews.
Make the total being measured visible. Suppose you receive 25 usable answers from one service for one fixed question set. If your business appears in 10, links to your domain appear in six, and your business is recommended in four, the respective observed rates are 40%, 24% and 16%. These are hypothetical numbers. An answer can contribute to more than one category, so the rates should not be added together.
Track unavailable results as well. If Google shows an AI Overview for 17 of your 25 searches, report 17 available overviews out of 25 searches. You can then report four recommendations among those 17 observed overviews as a separate figure. Show both totals so the reader can see the difference from 25 usable answers on another service. Do not count a failed request or an absent overview as an answer that rejected your business.
Repeat observations before judging a change. One answer can vary with wording, context and service behavior. A higher observed rate after a page edit may be consistent with improvement without proving that the edit caused it. Record other relevant changes and show the individual answers behind a summary.
Add business outcomes beside these measures. Review identifiable referral visits, landing pages, enquiries and qualified opportunities. Ask customers how they found you when appropriate. Some discovery will remain unattributed, and analytics cannot recover every prompt. A citation is an observation about an answer; an enquiry is a different event in the buying process.
Use the result to choose the next investigation. Repeated mentions with few recommendations can prompt a review of audience fit, useful comparisons and proof. Citations with few enquiries can prompt a review of the page and its next step. Neither pattern identifies the cause by itself. Open the evidence before committing to a change.
Can one answer count in all three categories?
Yes. It can name your business, link to your page and recommend your service. Keep the categories separate even when they overlap.
Does a negative mention count as a recommendation?
No. Preserve the wording and label the context. A recommendation rule should require advice to choose or consider your business for the stated need.
Can WhatBuyersAsk supply these three rates automatically?
This article defines a manual observation method. WhatBuyersAsk's research reports organise saved buyer questions, website findings and priorities. A website answer-coverage or citation score is not automatically the same as the three rates defined here. Confirm a report's scope and definitions before comparing them.
Bring an AI answer to a WhatBuyersAsk demo. Use the buyer question, the linked evidence and the next page decision to make the discussion useful.
