A name and a link answer different questions.
A brand mention means the answer names a brand. A citation is a source link supplied with the answer. A recommendation is a judgement about suitability. An answer can contain any combination of the three.
This distinction matters when someone reports that an AI platform “featured” a business. Ask to see the answer and its source list. Being named in a warning is different from being recommended. Being cited as a source is different from being discussed as a supplier.
Heading’s visibility calculation checks answer text for the configured brand name, aliases and domain. A source URL appearing only in the source list does not, by itself, raise that mention-based score. Read the methodology before comparing it with a metric from another tool.
Four examples to classify before you count.
These are invented examples for a fictional business, Northstar Reporting. They are not captured AI answers or customer results. Assume the business uses northstar.example and that its configured aliases are known.
Keep the source list separate from answer text in your review. A domain in the answer can match Heading’s detection even when the brand’s display name is absent. A domain present only in the source list is a citation, not a text match.
| Answer or source evidence | Mention | Own-site citation | Interpretation |
|---|---|---|---|
| Answer: “Northstar Reporting is one option.” No link. | Yes | No | Presence; no recommendation is established. |
| Answer: “Check data retention before signing.” Source: northstar.example/security. No name or domain in answer text. | No | Yes | The website is cited; the brand is not named. |
| Answer: “Northstar Reporting supports this workflow.” Source: northstar.example/workflow. | Yes | Yes | Both signals; verify the claim against the source. |
| Answer: “Northstar Reporting may not fit this requirement.” Source: an independent review site. | Yes | No | A caution and a third-party citation, not an endorsement. |
Follow the link and check what it supports.
Open the exact cited URL. Check who published it, when it was updated and whether the relevant statement is still present. A homepage citation gives you less specific evidence than a link to a page that addresses the requirement directly.
Compare the answer’s claim with the page. Does the source actually support it? Does the answer omit a limit or turn a conditional statement into a promise? Record a mismatch rather than assuming the citation makes the claim correct.
A third-party page about your company is not a citation of your own domain. It may still be useful evidence of how the brand is described. Keep those categories separate so the report can explain what changed.
Do not infer a purchase or even a click from a source link. Citation evidence belongs alongside referral and conversion evidence, with a clear distinction between them.
Count responses before you count links.
One answer can mention the same brand five times and cite three pages from its website. That is one answer with a detected mention, not five visible answers. A count of source links is a different unit from the percentage of responses with a match.
For an illustrative ten-response sample, suppose six answers contain a brand match and four cite the website. Two contain both. The mention-based visibility is 60%. The share with an own-site citation is 40%. Eight contain either signal, but calling that 80% “Heading visibility” would use the wrong definition.
Keep the prompt set, platform, market and dates with these numbers. A collection failure changes coverage. A platform that returns no AI answer has produced a different observation. Combining the two hides a limitation in the sample.
Turn a mismatch into a task with evidence.
Imagine an answer says your product only exports PDFs, but your documentation now explains a live dashboard. Save the answer, prompt, platform, collection date and cited URL. Confirm that the live dashboard is available to the plan and audience being discussed.
Your task could be: “Check whether the reporting page clearly distinguishes PDF export from live dashboard access. Add the current plan limits and link to the setup guide.” Assign an owner and a review date. That is a testable content task; “improve AI authority” is not.
After the edit, review the same questions over several runs. Record changes to answers and sources without claiming the edit caused them. Another source, a platform update or normal answer variation could explain the difference.
- Preserve the original answer and source URL.
- Check the claim against current product evidence.
- Name the content gap and the reader it affects.
- Assign one action and a review date.
- Report observed changes separately from your explanation.
Know what these signals leave out.
A literal brand match can be ambiguous, especially when a name is an ordinary word. An unconfigured alias can be missed. A source can be out of date or only weakly related to the claim. No single count resolves those questions.
Use mentions to locate answers worth reading. Use citations to locate sources worth checking. Use the combined record to make a precise decision about what to investigate next.
For your next review, classify five answers using the four example patterns above. If two reviewers disagree, resolve the definition before aggregating a score.
