How to Read Mention, Position, Citation, and Sentiment Results

A CiteCue prompt result is a small evidence record: the exact question, the AI engine, the full answer, whether your brand was mentioned, where it appeared, which sources got cited, and how the mention reads. No single field tells the whole story, which is why reading them together beats skimming any one number. These prompt-level fields also feed the 0-100 visibility score, so understanding them here pays off everywhere else.

Step 1: Start with the prompt and full answer

Open a result in Prompts Monitoring and read the exact prompt before you look at any metric. Note who it's for, the decision stage, the use case, and any constraints. Then read the full answer.

Ask yourself:

• Did the engine answer the question that was asked?
• Is the answer a ranked list, a comparison, a narrative, or a refusal?
• Is your brand being recommended, merely named, or warned against?
• Do the cited pages actually support the statements they sit next to?

Metrics summarize this record; they don't replace it. CiteCue's Prompts Monitoring keeps the full answer attached to every result so a number can always be checked against its context.

Step 2: Interpret a mention

A mention means the recorded answer named your brand. It's the broadest visibility signal, and on its own the least informative.

Classify the context:

• Recommended: the brand is presented as a suitable option
• Compared: the brand appears beside alternatives, with trade-offs
• Referenced: the name shows up as an example or a factual reference
• Qualified or criticized: the brand appears in a limitation, warning, or negative assessment

Don't chalk up every mention as a win. A blunt warning and an enthusiastic first-place recommendation both count as "mentioned", but they call for very different responses.

Check your brand aliases too. A product name, parent company, former name, or common misspelling needs to be represented accurately in AI Context, or the monitoring can miss appearances that were really about you.

Step 3: Interpret brand position

Brand position records where your brand lands in an ordered or rank-like answer, when that concept applies at all. Earlier usually means more prominent in that result, but position needs the full answer next to it.

Position means less when:

• The answer isn't actually ordered
• Every option is framed as best for a different use case
• Your brand appears first only because the prompt named it first
• The engine wrote a prose discussion rather than a shortlist
• A top placement carries a serious negative qualification

Compare position for the same prompt and engine over time. And resist averaging unlike answer formats into a claim like "we rank #2 in AI". Without a defined prompt set and method, that sentence means nothing.

Step 4: Separate citations from direct citations

A citation is any source the AI answer displays or links as support. CiteCue records your pages, competitor pages, and third-party sources like publishers, review platforms, and forums.

A direct citation is a citation to your own domain. The distinction matters:

• Mentioned, not directly cited: the answer names you but leans on other sources, or shows no source for the mention
• Directly cited, weakly mentioned: your page supplies evidence, but the answer doesn't prominently recommend you
• Mentioned and directly cited: your brand and your first-party content both appear
• Neither: another brand or source owns the result

Open the cited URL. Work out which fact or section likely supported the answer, whether the page is current, and whether a better internal page should own that topic. Citations and Competitors aggregates these source patterns across your whole prompt set, and if you want the background on why engines pick the sources they do, we've written up how AI engines choose citations.

Step 5: Interpret sentiment

Sentiment describes how positively, neutrally, or negatively the answer talks about your brand when it mentions you. Treat it as a cue to go read the text, not a verdict.

Read the actual wording when:

• Sentiment shifts between scans
• The answer mixes praise and criticism
• A factual limitation could be commercially important
• The wording is sarcastic, conditional, or attributed to someone else
• The answer contains an inaccurate or outdated claim

A negative statement can be accurate and useful. An answer that notes a real plan limitation is doing its job, and the right response may be improving the product or clarifying the audience rather than suppressing the statement. If the claim is false or outdated, log it in Sentiment and Brand Risk and trace the source that needs correcting.

Step 6: Compare engines and prompt groups

Build a compact comparison table for the question you're investigating:

When one engine disagrees with the rest, that usually points to a source or retrieval issue on that platform. A gap across every engine more often signals a broader content, entity, or third-party evidence problem. That's a diagnostic pattern, not a certainty, so inspect the cited pages before you commit to a fix.

Step 7: Record one evidence-backed next action

Finish each review with one concrete action tied to the result:

• Clarify an answer on an existing page
• Correct an outdated first- or third-party source
• Add primary evidence or a transparent comparison (our guide to writing citation-ready content covers what that looks like)
• Fix crawler or index access
• Seek legitimate reviews for an under-evidenced use case
• Keep the prompt unchanged and verify on the next scan

Write the hypothesis down: "If we add current plan limits and link the policy source on the pricing comparison page, direct citations for these three prompts may improve." The next scan tests that hypothesis. It doesn't guarantee the result.