The questions an audit asks the AI surfaces. Generated from the brand profile, then reviewed and selected. Beta: these shapes may change; see the changelog.
The prompt set with its selection state. ready is true when the audit is waiting for prompt review, and maxQueries is the plan’s ceiling on how many may be selected — selecting more than that is rejected when the run starts, not here.
Adds one prompt you wrote (3–500 characters). It is selected when the plan has room and lands unselected otherwise — an add is never refused for capacity. It starts under the CUSTOM angle and is moved to a better-fitting angle in the background, so a later read may show a different angleId.
selectedIds is optional: send it to replace the selection of existing prompts in the same transaction as the add — the way to make room on a plan whose minimum and maximum selection are equal, where a separate deselect would be refused. Omit it to leave the selection alone.
Takes the prompt out of the set. A prompt that has already been answered is archived rather than deleted, so its history stays in past snapshots; removed says which happened. Either way it no longer appears in the list.
Removes a prompt found in search results (mined: true) and remembers the dismissal, so the next run does not find it again. Other prompts are removed with DELETE /audits/{auditId}/prompts/{promptId}.
Changes a prompt’s text. An unanswered prompt is edited in place. An answered one is archived with its history and a new prompt takes its place, keeping its selection: the response’s id is then the new prompt and parentId the old one, otherwise null.
Changes the language a prompt is asked in. A prompt that has been answered is replaced rather than changed, so its history stays in past snapshots: the response’s id is then the new prompt and replacedPromptId the old one.
Overrides the classifier’s branded/unbranded call. promptTypeSource becomes user, so a later reclassification leaves it alone. RECALL is fixed: no prompt moves into or out of it.
Applies one action — select, deselect, remove, restore — to up to 500 prompts, and reports what happened to each. remove archives a prompt that has been answered and deletes one that has not; removed counts both.
Adds up to 100 prompts in one request, one per entry in texts. Each line is cleaned the way the dashboard’s paste box cleans it (list markers, numbering and surrounding quotes are stripped). If any line is out of bounds the whole request is a 400 naming every bad line, and nothing is added. Duplicates — within the list or of a prompt already on the audit — are skipped and counted in skippedDuplicates, so sending the same list twice is safe.
New prompts are selected while the plan has room. selectedIds is optional: send it to replace the selection of existing prompts in the same transaction as the add — the way to make room on a plan whose minimum and maximum selection are equal, where a separate deselect would be refused. Omit it to leave the selection alone. selectedIds: [] deselects every existing prompt.
Send returnPrompts: false to receive only the counts: the full list can run to tens of kilobytes, and an automation rarely needs it.
Every prompt, active and archived, as a CSV download whose headers the import accepts. Past 5000 rows the file ends with a # truncated line; page the rest with GET /audits/{auditId}/prompts/manage.
Generates the prompts for this audit and returns the resulting set, leaving the audit awaiting prompt review. This call costs money — it runs an LLM generation — so treat it as a one-time step per audit rather than something to retry casually.
Both body fields are optional. Sending a non-empty topicIds selects the deterministic template path; omitting it generates over the audit’s selected angles. languages overrides the audit’s stored targeting languages for this generation only.
Requires status to be AWAITING_CONFIG or AWAITING_PROMPTS, and answers 409 otherwise. A newly created audit is PENDING: call POST /audits/{auditId}/prepare first and wait for it to reach AWAITING_CONFIG.
Tops the set back up: each selected angle short of its share of the plan’s prompt limit gets new LLM-generated prompts, skipping texts the audit already has. Existing prompts are kept. This call costs money — it runs an LLM generation.
Requires status to be AWAITING_PROMPTS (after POST /audits/{auditId}/prompts/generate) and an organization plan; answers 409 otherwise, and when every angle is already at its limit. A 400 means the audit has no brand profile or no selected angles to generate from.
Adds prompts from a CSV file sent as text (at most 1000 rows and 5 MB). A prompt column is required and a topic column files each row, creating up to 50 new topics; other columns are reported in ignoredColumns. A file from the export re-imports as-is.
Send dryRun: true first: it answers 200 with the same per-row report and writes nothing. A real import answers 201. Rows over the plan’s capacity are added unselected (over_capacity), never refused.
One page of prompts, filtered and sorted, with the plan’s capacity. Unlike GET /audits/{auditId}/prompts, archived prompts can be listed too. Pass nextCursor back as cursor for the next page; it is null on the last one.
Renders the prompts the template path would generate from the profile, competitors and topics in the body — not the audit’s stored ones — so an editor can show them before saving. Deterministic and free: no LLM runs and nothing is written. Every body field is optional.
Replaces the selection: the prompts named in selectedIds are selected and every other prompt is not. Ids that are not live prompts of this audit are ignored. On a paid plan the selection has a floor, so a run stays meaningful; going under it is a 400 that names the minimum.