Three records that should stay separate
When a local SEO client asks “Did AI recommend us?”, first inspect what was actually saved. A business name, a reference URL and a verified relationship between them are different pieces of evidence.
| Record | What it establishes | What remains unknown |
|---|---|---|
| Business mention | The business was named in this answer. | Whether it was endorsed, seen by a buyer or contacted. |
| Response citation | A source reference was saved with this response. | Whether it supports this particular business or claim. |
| Business-linked source | The saved record explicitly links that source to a mentioned business. | Whether the source is accurate or caused the model’s answer. |
A mention without a citation
An answer can name a roofer without giving a supporting source for that business. Record the appearance if the observation is usable, but keep business-linked support empty. Do not borrow a citation from another repetition or from another business in the same answer.
A name may also occur in a warning, comparison or negative statement. “Mentioned” is a narrower and more defensible label than “recommended” unless the actual wording supports the latter. PresenceCurrent’s visibility metric counts target appearances in usable responses; it is not an endorsement score.
A citation without a business mention
An answer can link a company’s educational page while discussing a general roofing problem, without naming that company as a local provider. That is a source appearance, not automatically a target-business mention.
Conversely, a directory cited in the answer may cover many businesses. Its presence alone does not establish a saved relationship to each name in the response. In the sample inspector, “Response reference only” means no business-specific link was saved. It does not prove the source is irrelevant.
AI answer citations also differ from traditional local SEO citations: a directory listing containing a business’s name, address and phone number is not itself evidence that an AI answer cited that listing.
Evidence coverage is not visibility
The two metrics use different numerators. In the fictional Synthetic Demo V3, target mentions are 20 / 71 usable responses = 28.2% observed visibility. Responses with saved sources are 59 / 71 = 83.1% evidence coverage.
Evidence coverage asks how much of the usable response set has saved source references. It does not mean 83.1% of customers see the business, that all those sources support the target, or that the answers are correct. Excluded attempts remain outside both denominators.
These figures describe the approved demonstration only. The synthetic source domains are fictional. They are not backlinks, real citations earned by PresenceCurrent, or performance evidence from a customer.
How to inspect Citation Context
- Open the synthetic saved-evidence inspector and choose a search question.
- Select a repetition and check whether it is usable or excluded.
- Read the mentioned businesses and each saved response source separately.
- Select View context to inspect the recorded business link, business excerpt and saved response context.
- Read the complete saved answer before describing the finding to a client.
Citation Context displays saved evidence. It does not independently fetch or validate the source page, and it cannot reconstruct why a provider generated an answer. The methodology explains how these limits affect an audit.
Explain the difference in a client report
Use precise statements: “The business was named in this saved response” or “This response includes a source linked to that business in the saved record.” If there is no recorded business relationship, state that rather than turning a response-level reference into proof.
Choose a next step that can be checked: review whether the cited page accurately describes the service and location, inspect a recurring comparable business, or repeat a bounded question set later. Do not promise that adding a source, schema or directory listing will produce a citation.
For a complete process, follow the local AI visibility audit guide. Agencies can use the client review checklist to keep observations and recommendations separate.
Search-system context
Bing’s AI Performance documentation treats visible citation activity as observational data, rather than a ranking or proof of a particular change’s impact. Google’s AI features guidance explains eligibility for supporting links and makes clear that inclusion is not guaranteed.
Those platform reports and a saved prompt audit measure different things. Keep the provider, interface, date and scope beside each result so clients can judge what is comparable.