Answer Radar
The Action Queue
The Actions tab is what the audits found that is worth doing, as cards. Each card is a short label plus the evidence behind it: the observed facts with their citation ids, impact counts and a confidence value. There is no written brief on a card, because the brief is yours to write from the evidence. This article explains where cards come from, how to work them, and how they move.
Quick Answer
Only two card types are generated today, content_gap (nobody named you for this question) and competitor_won (this competitor is who the engines named), because those are the two claims a measured run can support. Cards start suggested and move through todo, in_progress, done or dismissed. Filter by project, gap, type, status, or open and closed.
- Impact keys: answers without you, intent, evidence, competitor citations, competitor mentions.
- Evidence mode resolves citation ids to the recorded URL, domain and title.
- Groups are shown in lifecycle order and empty groups are dropped.
Where cards come from
Aggregation measures one thing: whether your brand appears in the answers to a prompt, and who appears instead. That evidence supports exactly two finding types. A content gap says nobody named you here. A competitor won says this competitor is who the engines named, keyed on what was observed, a domain or an entity name. Five other card types exist in the vocabulary, untracked page, community opportunity, setup gap, offsite mention and sentiment, and each needs an evidence source a measured run does not have, so none is generated from an absence.
Cards are written by a generator with its own version, stored on every row, so a queue can be read as 'cards from generator N' the way samples are read under a parser version.
What a card holds
A label, bounded at 200 characters, that is a handle rather than a sentence. The observed facts, passed through verbatim from the diagnosis with their citation ids. Impact counts in a fixed order: answers without you, intent score, evidence score, competitor citations, competitor mentions; unknown keys are appended rather than dropped, because hiding evidence is the one failure this surface cannot have. A confidence value. And links to the gap, run, prompt and competitor the card came from.
answer_action_get with mode evidence resolves the citation ids into the recorded sources, URL, domain and title. Ids that no longer resolve are reported rather than quietly dropped.
Working the queue
Statuses are suggested, todo, in_progress, done and dismissed, in lifecycle order. Every generated card starts suggested. In the app, the queue groups cards by status in that order and the buttons on each card are the moves allowed from its current status. Through MCP, answer_actions_list filters by project, gap, type, a single status, or state open (suggested, todo, in_progress) and state closed (done, dismissed), and answer_action_update moves a card.
The lifecycle table the app shows is advisory: the service performs the write and is the only gate, so a stale button produces a refusal the page resolves by refetching, never a bad write.
From card to fix
A content gap card links to its gap. From there, the content gap view says which existing page was supposed to answer the question and what it misses, and answer_fix_draft assembles the evidence skeleton for a fix. Move the card to in_progress while you write, done when the fix is approved, and register a re-verification so the follow-up measures whether anything moved.
- Read the impact counts before the label; they are the measured facts.
- Dismiss cards you will not act on so the open queue stays honest.
- A competitor added as tracked later raises a fresh card; dismiss the old one.
FAQ
Why is there no recommendation text on the card?
By design. The measurement layer writes no prose about why a finding matters; it passes the observed facts through. Your assistant reads the card and explains it, and the editorial rule that keeps observations, hypotheses and actions separate applies.
Why did the same finding appear twice?
Most often because an untracked competitor was later added as tracked. Its evidence then carries a row id and the derived key changes, so the finding is raised once more as a new card while the old one keeps its claim until dismissed.
Can I create my own card?
No. Cards are derived from measured evidence so that every card's credibility is earned. Track your own tasks in your own tool, and use fixes for the work itself.
Related help pages
Answer Radar
The Content Gap View
How Answer Radar scores each tracked page against what the engines' answers required, how a requirement is derived, the three exclusions, and why coverage.
Answer Radar
Drafting a Fix From the Evidence
How a fix goes from evidence skeleton to approved artifact: draft, write, check, save, review. What the faithfulness guard rejects and who can approve.
Answer Radar
Adding Competitors to Answer Radar
How tracked competitors feed share of voice and competitor won cards, how untracked ones are still detected, and why one vendor counts once per answer.
Answer Radar
The Email Brief and Its Preferences
How the recurring Answer Radar brief works: off by default, switched on by you, weekly by default, what it contains, and why it can say nothing meaningful.