add_prompts
ChatGPTAdd up to 50 prompts to one topic of a project. A prompt is a question RadarKit will ask every tracked AI model on the project's schedule (for example "best CRM for startups"). Prompts that already exist are reported under skipped with their prompt_id, never duplicated, so calling again with the same list is safe. When the plan's prompt limit is reached the ones already added stay and the rest are reported as skipped with reason "limit"; tell the user. Costs 1 API credit per call.
add_topic
ChatGPTCreate a topic in a project. Topics group prompts (for example "Pricing" or "Comparisons"); every prompt belongs to one. Names are unique per project, case-insensitive: if the name exists you get TOPIC_EXISTS with the existing topic_id, so reuse that id instead of retrying. Costs 1 API credit.
create_content_article
ChatGPTHave RadarKit's Content agent write an article for a tracked prompt: it studies the pages AI models cite for that prompt, builds an outline and writes the draft in the project's brand voice. Costs 1 API credit plus one AI content credit from the project owner. Runs in the background for several minutes: poll get_content_run with the returned id until status is completed, then give the user the editor_url or fetch the text with include_body. Calling again for the same prompt, type and keyword returns the existing run instead of writing a second article, unless force is true. The run also appears in the project's Agents → Sheets page on the "API" sheet.
create_export
ChatGPTStart a data export: a gzip CSV or NDJSON file with every row matching the selection (same arguments as estimate_export). Charges the account's Export rows meter exactly the row count, up front; no API credits. Call estimate_export first and confirm with the user. The export runs in the background: poll get_export with the returned id until status is ready, then give the user download_url (valid 30 minutes, file kept 7 days). One export at a time per user.
estimate_export
ChatGPTHow many rows an export would contain and whether the account has enough Export rows to pay for it (an export costs exactly its row count). Free. Always call this before create_export and tell the user the row count. Types: responses (one row per AI answer, with the full answer as Markdown), sources (every source an answer cited or consulted), competitors (every competitor named), rankings (ranked domains with position), query_fanout (search queries the model ran). Dates YYYY-MM-DD, up to 365 days. Scope with topic_ids or prompt_ids, or leave both empty for every prompt.
get_account
ChatGPTWho this API key belongs to and what it can do: key name and permissions, plan, API credits (spent per tool call), Export rows (spent by create_export) and rate limits. Free. Call it to check how many credits or export rows are left, or which permissions the key has.
get_competitors
ChatGPTShare of voice: which brands AI answers recommend for the project's prompts, how often, their share of all recommendations, their average position and their rank, with the project's own brand and its rank called out. Same numbers as the Overview's Share of Voice card. Up to 100 competitors, 31-day window. Costs 2 API credits.
get_content_run
ChatGPTStatus of a content article run: processing (with current_step), completed, failed (with the error) or cancelled, plus the article's title, statuses, word count, scores and editor_url. Free to poll. Set include_body to true to also get the finished article as Markdown; that costs 2 API credits, so do it once, after status is completed.
get_export
ChatGPTStatus of one export by id: queued, running (rows so far), ready (with download_url, valid 30 minutes), failed (with the reason; rows were refunded) or expired. Free, so poll it every few seconds after create_export until status is ready.
get_project
ChatGPTOne project: name, website, region, timezone, the AI models it tracks, active prompt count and next scheduled run. Costs 1 API credit.
get_query_fanout
ChatGPTThe web searches AI models ran behind the scenes to answer a prompt (query fanout): one row per answer that searched, with the list of search queries. Useful to see what the model thinks the question is really about and which phrasings to target. Newest first, at most 25 per call; follow next_cursor. Dates YYYY-MM-DD, at most 31 days. Costs 1 API credit per call.
get_response
ChatGPTOne AI answer in full: the answer as Markdown (cut to max_content_chars), plus everything RadarKit extracted from it — the sources it cited or consulted, the competitors it named, the domains it ranked with their positions, and any shopping tiles. Use a response id from list_responses. Costs 2 API credits.
get_sentiment
ChatGPTHow AI answers talk about the brand when they mention it: counts and shares of positive, mixed and negative answers, overall and per model, for a 31-day window. Same scope as the Overview's Sentiment card (only answers that mention the brand). Costs 2 API credits.
get_visibility
ChatGPTHow visible the brand is in AI answers, day by day: the share of tracked prompt-and-model combinations whose answer mentioned the brand (visibility, 0 to 100), the brand score, and how often the brand appeared at the top, middle or bottom with positive, mixed or negative sentiment. One row per model per day plus an "all" row across every model; summary is the average over the window. Dates are YYYY-MM-DD, up to 366 days per call. Costs 1 API credit.
get_web_search
ChatGPTHow often AI models ran a web search before answering, and what they searched for: the share of answers that searched, then per prompt the distinct search queries with how many times each was seen and the average number of queries per search. Pass prompt_id for one prompt (the Web Search card on its page) or leave it out for every active prompt (the Query Fanout table), most queries first. Dates YYYY-MM-DD, at most 31 days. Costs 2 API credits.
list_exports
ChatGPTThe most recent exports for a project (newest first, at most 50): status, rows, size, who requested it, download_url when ready (valid 30 minutes) and when the file expires. Costs 1 API credit. To check one export use get_export, which is free.
list_projects
ChatGPTEvery project this key can reach: the project id every other tool needs (24 hex characters), the brand website, the AI models tracked, the number of active prompts and the next scheduled run. Call this first. Costs 1 API credit.
list_prompts
ChatGPTThe prompts a project tracks: id, text, topic, when each was last asked and the current status per AI model (completed, pending, processing, hold). Oldest first, at most 25 per call; follow next_cursor while has_more is true. Filter by topic_id to see one topic. Costs 1 API credit per call.
list_rankings
ChatGPTThe brands and sites AI answers recommended, answer by answer: one row per recommended domain per answer with its position in the list (1 = first). Newest first, at most 25 per call; follow next_cursor. Filter by prompt_id or domain (for example the brand's own domain to see where it ranks). Dates YYYY-MM-DD, at most 31 days. For totals use get_competitors. Costs 1 API credit per call.
list_responses
ChatGPTAI answers RadarKit collected for a project: one row per prompt, model and day with brand mention, position, sentiment, whether the model searched the web and which queries it ran, and how many competitors it named. Newest first, at most 25 rows per call; follow next_cursor while has_more is true. Dates are YYYY-MM-DD, at most 31 days per call. Set include_content to true to get each answer as Markdown, cut to max_content_chars; that costs 2 API credits instead of 1. To analyse many full answers use estimate_export and create_export instead of paging.
list_sources
ChatGPTThe web pages AI answers cited or consulted for a project: one row per source per answer with the URL, domain, whether it was cited (linked in the answer) and its citation rank. Newest first, at most 25 per call; follow next_cursor. Filter by prompt_id, domain or cited. Dates YYYY-MM-DD, at most 31 days. For "which sites matter most" use top_source_domains instead. Costs 1 API credit per call.
list_topics
ChatGPTThe topics of a project with how many prompts each one holds. Topics group prompts; you need a topic_id to add prompts. Costs 1 API credit.
retry_content_run
ChatGPTRestart a content article run that failed or was cancelled (get_content_run shows retryable: true). The run keeps its article and spends no new content credit; costs 1 API credit. Refused with RUN_NOT_RETRYABLE for runs that are still processing or already completed.
run_prompt_now
ChatGPTAsk every tracked AI model this prompt right now instead of waiting for the project's next scheduled run. Costs 1 API credit plus 1 realtime credit from the project owner's balance (see realtime_credits_remaining in the result). Refused with PROMPT_PROCESSING while a run for that prompt is in flight, and with INSUFFICIENT_REALTIME_CREDITS when the owner has none — do not retry either. New answers appear in list_responses a few minutes later.
top_source_domains
ChatGPTWhich websites AI answers lean on most for a project: domains ranked by how often they were used, with their share of all source mentions, how many times they were actually cited, how many prompts they appeared for, which models used them, when they were last seen, and whether each is the brand, a configured competitor or another site. Pass cited: true to rank by inline citations only (sources linked right next to a sentence in the answer). At most 50 domains per call, from a 31-day window. Costs 2 API credits.
