Solutions / PR and brand teams
How AI assistants describe your brand, which sources they lean on, and what moved this week
Communications teams type their questions into our search results in full sentences: a weekly report on how assistants describe us, which sources the engines cite about our history, whether the answers turned negative. Here is what Answer Radar gives you for each, and the one thing it does not.
Updated 2026-09-04
What a brand team gets
Four asks we see in our own Search Console, answered with what exists.
Answer Radar measures what ChatGPT, Gemini, Perplexity and Claude say when asked the questions you approve about your brand and category, stores the answer text and every cited source per sample, and reports per engine: whether you were named, whether you were recommended, who else was, which domains were cited and flagged as yours or a competitor's, and a judged sentiment mix. A brief compares one window with the last on the cadence you choose. The fix loop carries a correction through a source guard, human approval and a re-measurement. It does not poll engines for real-time alerts, and this page says so rather than implying it.
Our own brief, pulled on 4 September 2026
The payload the weekly email is rendered from, on our account. One project measured once, so movement is not_comparable, which is the honest answer for a first week.
answer_digest_get({ days: 7 })Through the Linkeddit MCP connector. The email renders the same fields as fixed sentences; the narrative is yours or your assistant's.
answer_digest_get, days 7, abridged
- window
- 29 August to 5 September 2026 against 22 to 29 August.
- movement
- not_comparable on all five engine segments, reason absent_in_baseline: one run in the window, none before it. Not a fall to zero.
- actions
- 9 new, 9 open, 0 completed, 0 dismissed. Top card: Not represented for the prompt about competitor monitoring software for pricing changes and product launches.
- nothing_meaningful_changed
- false, because nine cards were raised inside the window. It would be true only if movement held still or could not be compared, no card was raised and nothing was closed.
- confidence
- null. None is computed, so none is reported.
From the response's own notes: no sentence explaining what any of this means is written. The facts are yours to narrate, which for a comms team is the correct division of labour.
Four asks from comms managers, and the honest answer to each
The left column is how people phrased it in our Search Console this summer. The right is what exists.
| The ask | What Answer Radar gives you |
|---|---|
| A weekly report on how AI assistants describe our brand | The brief. Weekly or monthly, compares the last window with the one before across every active project, per engine, and says nothing_meaningful_changed with the reason when the week was quiet. Template-thin by design: your assistant writes the narrative from the same payload. |
| Which sources the engines cite about us | Citation share, per engine, each cited domain counted once per answer and flagged as yours, a tracked competitor's or neither. Plus the fan-out view: the searches the engine ran before it answered, which is where a stale article gets picked up. |
| Whether the answers are negative or inaccurate | Sentiment mix, as judged by the mention extractor: the share of a brand's mentions labelled positive, neutral or negative. Accuracy is not scored; the answer text and citations are stored per sample so a person can read what was actually said and against which source. |
| Alerts the moment an answer turns | Not offered. Each sample costs a credit, so engines are measured on a cadence you set, not polled. A vendor promising real-time alerts on AI answers is either polling at someone's expense or describing something else. We would rather tell you the cadence. |
Sentiment mix, as the metric layer defines it
Copied from the answer_metrics_get response, version visibility-metrics-v1, so the page and the product cannot drift.
“The distribution of judged sentiment (positive, neutral, negative) across a vendor's mentions in the segment, as a share of that vendor's mentions carrying a sentiment label.” On our own run on 4 September 2026 Semrush carried 9 labelled mentions on Claude, 89% positive and 11% neutral, and 18 on Gemini, 61% positive. Nobody in our category carried a negative label on that run. A brand team reads this next to the stored answer text, not instead of it.
What it does not do for a communications team
Each is a thing the category's marketing tends to imply.
- No real-time alerts. Engines are measured on your cadence, because each sample costs a credit and a continuously polled engine is somebody's bill.
- No accuracy score. Sentiment is a judged label; whether an answer is factually wrong is read by a person from the stored answer text and its cited source.
- No attribution. A re-measurement after a correction reports an observed delta with a threshold and control prompts. There is no field in which a cause can be recorded.
- No consumer-app reproduction. Measurement runs through official provider APIs with grounded web search, a controlled proxy, and every number is stamped with the engine, instrument and prompt configuration behind it.
- No editing of third-party pages. When the cited source is a news article or a forum thread, the fix is a correction through that publisher's process, and the evidence store tells you which page to take it to.
Frequently asked questions
Is there a tool that reports weekly on how AI assistants describe our brand?+
Yes. Answer Radar's brief compares the last window with the equal window before it across every active project, per engine, and lists what moved, what actions are outstanding or new, and what was closed. It is off until you turn it on, weekly by default, and it says nothing_meaningful_changed with the reason when the week was quiet rather than manufacturing a highlight. The same payload is answer_digest_get through the connector, so an assistant can write the narrative for your stakeholders from it.
Which sources do AI engines cite about our company, and how do we correct the record?+
The citation share metric lists every cited domain per engine, counted once per answer, flagged as yours or a competitor's. The fan-out view shows the searches the engine ran to get there. When a stale or wrong page is doing the work, the fix loop drafts a correction against the recorded evidence, a guard refuses any source the engines did not actually return, a person approves, and a re-measurement after the indexing delay reports whether the answers moved. It reports movement, not cause.
Can it alert us when AI answers about the company turn negative or inaccurate?+
Not in real time, and we will not describe the brief as if it did. Engines are measured on the schedule you set, weekly or monthly, because every measured sample costs a credit. What you get is the sentiment mix per brand per engine on each run and the stored answer text to read. If you need a same-day signal on a live incident, that is a different product and probably a person.
Does it measure sentiment?+
It reports a sentiment mix: the distribution of judged sentiment, positive, neutral or negative, across a brand's mentions in a segment, as a share of the mentions carrying a label. It is a judged label from the mention extractor, versioned so it stays comparable, and it is one input to read next to the answer text, not a reputation score.
We are not a software company. Does the measurement still apply?+
The instrument does. Every number we publish today is from B2B software projects, because that is who has run public projects with us; the metric definitions, the fix loop and the brief are the same for any brand with a name an engine can say. Write the prompts the way your stakeholders and customers ask, and the run measures those.
Can we share a result with an executive who has no seat?+
Yes. A share link opens one gap, one run, one fix or a project's metrics to someone without an account. Summary only, 30 days by default, revocable in one call. The answer text and sources stay inside the account.
Read next
Fix and verify
The recorded source check, human approval, and six movement outcomes after the indexing delay.
Read For the executiveShare links
One result to someone with no seat, summary only, 30 days, revocable.
Read DefinitionsMethodology
Every metric's definition, including sentiment mix, copied from the measurement layer.
ReadKnow what the engines say about you before the board asks
Answer Radar is included with the Compete plan, alongside the weekly competitor brief, in the browser and through the MCP connector.