AI mentions vs citations: what should your business measure?

1. Separate the name, the link and the meaning

A mention means the answer names the business or a recognised alias. A citation means a source link points to the domain you are measuring. The exact linked page matters: it could be a service page, an article or a page that is no longer current.

A recommendation is a claim about the answer's meaning and context. A name in a warning, a comparison or a historical example is not automatically an endorsement. Sikwati's mention and citation flags are not a verified recommendation metric.

2. Use the same denominator and allow overlap

For a saved set, count usable answers containing a business mention and usable answers citing its domain independently. An answer can belong to both groups. Do not add the two counts to claim a larger number of unique answers.

In a fictional set of five usable answers, imagine three name your business, two cite your website, and one does both. Mention coverage is 3/5; citation coverage is 2/5; four distinct answers contain at least one signal. None of these numbers describes your share of all AI usage.

Report failed or incomplete checks next to the result. Compare like-for-like question sets and provider settings. A higher percentage from a smaller, easier question set is not necessarily an improvement.

3. Inspect the source behind the count

Open the saved source URL and inspect the relevant passage in the answer. Was your own site linked, or did a third-party directory discuss your company? Both may be useful observations, but a third-party link should not be reported as a citation to your domain.

Watch for names shared by unrelated companies, abbreviations and outdated brand names. Keep an agreed alias list, and manually review unclear matches. A count without the underlying answer is difficult to audit or explain to a client.

• Is this the correct business, not a similarly named company?
• Which domain and exact page were linked?
• What claim does the source appear to support?
• Is the answer favourable, neutral, critical or unclear?
• Was the result produced by the same surface as the earlier comparison?

4. Match the next action to the observation

If you are named but not linked, investigate the sources used; do not assume your robots file is the cause. If your educational article is cited but your business is not named, the page may be contributing useful information without provider discovery. Consider whether that matches the question's intent.

If neither signal appears, inspect the question, answer relevance and the pages supporting other businesses before rewriting your site. Technical readiness and content clarity are separate investigations. No single audit score explains every missing mention.

When you change a page, record the change and keep later results, including unchanged or worse observations. A before-and-after difference is not enough to establish causation.

5. Write a report someone else can verify

Use a statement such as: ‘Across five saved questions checked with this provider on this date, the business was named in three answers and its website was linked in two. One answer contained both signals.’ Then attach the evidence and describe what you will review next.

Sikwati keeps dated OpenAI and Gemini API evidence and competitor comparisons. Those checks do not directly measure consumer ChatGPT/Gemini or Google AI search features. Keep leads, sales and website visits in separate outcome reporting; a citation does not prove that anyone clicked or purchased.