How to read an AI visibility result
Read mentions, recommendations, citations, and missing answers as separate observations.
On this page
Read the evidence before the headline. Start with the scope, inspect the returned answers, then choose one action the record supports.
This order avoids treating a mention as proof of expertise or a future recommendation. It also avoids treating a missing or failed answer as evidence that an engine rejected you.
Gistful currently measures saved questions across ChatGPT, Gemini, Claude, and Perplexity. The engine names appear in the result because an answer from one channel should not be blended into an unnamed claim about “AI.” Google AI Overviews are not included in new Gistful scans.
1. Read the measurement label
Before interpreting a result, check these fields:
- the exact question and whether it is a discovery question or a name-recognition question
- the engines that were included
- the collection date and requested locale
- the profile scope and topics used for the scan
- the status of each check: returned, missing, failed, or awaiting review
These fields tell you what the result covers. A name-recognition question can show whether an engine associates a name with something, but it should not inflate discovery coverage. A missing or failed answer is unmeasured. It does not belong in the denominator for a missed mention.
For a fuller explanation of the measurement boundaries, see Gistful’s engines and methodology.
2. Open the answer behind every meaningful result
An answer card can summarize the state, but it cannot replace the text. Read the actual response when a result names you, omits you, returns a different person, or asks for more context.
Check three things in order:
- Identity. Does the name refer to you rather than another person with a similar name?
- Relevance. Does the surrounding text connect you to the expertise, audience, and geography in the saved question?
- Evidence. Does a cited page support the narrow point you are relying on?
A cited page may establish who someone is without confirming the services, results, or credentials the answer attributes to them. OpenAI’s instructions for ChatGPT search tell readers to check whether cited sources support the answer. OpenAI’s guidance describes how to inspect the sources attached to a response.
3. Keep four result states separate
Use plain labels when you review a scan:
| Result state | What it means | What to do next |
|---|---|---|
| Identity-supported mention | The returned answer appears to identify you in relevant context | Open the answer and cited pages. Confirm your public information is current. |
| Mention needing review | The text may refer to you, but the identity or context is unclear | Do not count it yet. Inspect the wording and source. |
| Returned answer without a supported mention | The engine returned an answer, but the record does not support identifying you in it | Read the question and response before deciding whether any public information needs work. |
| Missing or failed check | The check failed or no answer was returned for this test | Keep it unmeasured. Retry missing checks when appropriate. |
This table is a reading tool, not a promise that every answer fits neatly into one box. An answer can also list a directory, answer a different question, or contain a name that needs human review.
4. Inspect citations for the claim you need
A citation is a link worth checking. It does not prove that every sentence in an answer is correct. Open it and ask what the page supports. Is it current? Does it identify the right person? Does it substantiate the service that matters to the client question?
This habit also improves your own public pages. If a cited profile is old, correct the old claim. If it is thin, add factual context that a reader can verify. Do not add a claim you cannot support just to make a page sound more impressive.
Google’s current guidance for generative search emphasizes unique, useful content for people and warns against creating scaled pages merely to manipulate results. Its AI search guidance applies to Google Search features, not as a formula for every engine. It still offers a sensible editorial standard: publish material that helps the person who lands on it.
5. Choose one evidence-led action
Use the following decision note after each scan:
| Field | Record |
|---|---|
| Question reviewed | Exact saved question |
| Engine and date | The named engine and collection date |
| Returned answer | A short factual description of what it said |
| Cited page | The narrow point the page supports |
| Remaining uncertainty | What the answer or source does not establish |
| Next public action | One fact to correct, add, or leave alone |
| Reader reason | Why the action improves the information a reader can use |
The note forces a useful limit. “Publish more content” is not an action supported by an answer. “Correct the service list on my profile because the cited page still lists a service I stopped offering” is specific, reviewable, and good for a reader whether or not a later answer changes.
The published 60-answer baseline study shows the same discipline. It preserves the question set, engines, dates, answer states, and limitations, then reports what the archive supports. It does not turn an absence into a verdict or a parser correction into a visibility gain.
Inspect the evidence, fix what is demonstrably inaccurate or incomplete, and keep the same scope when you need a comparable rescan.