LLM Pulse

Delete a timeline annotation from a project. Only user-created annotations that belong to your own user can be deleted (system annotations and other members annotations cannot). Never call this speculatively: only when the user explicitly asked to delete this specific annotation. In the in-app chat the user gets a Confirm/Cancel card; external MCP clients delete immediately. Find annotation ids via list_annotations. Available on every plan.

Delete a tag/collection from a project (same as Delete on the Tags page; tags and collections are the same object). IRREVERSIBLE for the tag itself, but the prompts inside it are NOT deleted: only the grouping disappears. Never call this speculatively: only when the user explicitly asked to delete this specific tag, and confirm its name with them first. In the in-app chat the user gets a Confirm/Cancel card; external MCP clients delete immediately. Find tag ids via list_tags / list_collections.

Delete a competitor from a project (same as Delete on the Competitors page). IRREVERSIBLE: the competitor disappears immediately, frees a competitor slot, and its tracked data (mentions, citations, sentiment, share-of-voice history) is purged by a background job. Never call this speculatively: only when the user explicitly asked to remove this specific competitor, and confirm the brand with them first. In the in-app chat the user gets a Confirm/Cancel card; external MCP clients delete immediately. Find competitor ids via list_competitors.

Delete a prompt from a project (same as Delete on the Prompts page). IRREVERSIBLE: the prompt disappears immediately, frees a prompt slot, and its whole history (executions, mentions, citations, sentiment, and any GEO Writer tasks created for this prompt) is purged by a background job. Never call this speculatively: only when the user explicitly asked to delete this specific prompt, and confirm the exact prompt text with them first. In the in-app chat the user gets a Confirm/Cancel card; external MCP clients delete immediately. Find prompt ids via list_prompts.

Delete a webhook subscription by ID. The target URL stops receiving events immediately. Available on Scale and above plans.

Create the real project from a completed draft (same effects as create_project). Idempotent: finalizing an already-finalized draft returns the existing project. Re-validates every plan gate and quota.

Get the plan, subscription window, quota consumption and API rate limits for the authenticated account. Call this BEFORE any tool that spends quota (create_prompts, launch_recommendations, create_technical_geo_report, create_intelligence_task) so you can tell the user what is left instead of discovering the ceiling by hitting it. An unlimited quota returns limit and remaining as null with unlimited=true. The subscription block is only present for callers who may access Billing & Plans.

Get aggregated AI bot traffic for a project from Cloudflare or uploaded server logs. Returns per-day request counts grouped by bot or by company. Available on Scale and above plans.

Get the AI Model Insights position-distribution comparison for the project brand and an optional competitor. For a fair head-to-head, pass brand_kind=non_brand (the in-app AI Model Insights default): brand-focused prompts skew positions toward the brand they name. When brand2 is omitted it defaults to the largest competitor by mentions. Weekly/monthly buckets are aligned to full calendar periods: the first bucket covers the whole week/month containing the range start (so its counts can differ from get_timeseries, which clips partial periods).

Per-AI-model breakdown of mentions, citations, sentiment and weighted visibility for the brand and competitors, in a single call. USE THIS for any question that compares performance across AI models ("which model do I rank best on?", "ChatGPT vs Perplexity vs Gemini", "where am I most visible across models?"). Prefer this over calling get_summary multiple times with different model filters: get_ai_model_summary returns all models at once and avoids hitting the step limit. The in-app AI Model Insights Overview tab defaults to non-brand prompts, so pass brand_kind=non_brand to match its numbers and for fair brand-vs-competitor comparisons. NOTE: mentions_market_share values are the independent mention rate of each actor within a model (mentions divided by the responses of that model); one response can mention several actors, so the values do NOT sum to 100 and must not be charted as a share pie. For a true share partition use get_sov.

Get aggregate Google AI Overview result availability over time plus per-prompt breakdowns.

AI traffic for a project: the human visits arriving from AI assistants (ChatGPT, Perplexity, Gemini, Claude, Copilot, Grok, Mistral, Meta AI, DeepSeek), measured from the connected web analytics provider (Google Analytics 4, Adobe Analytics, PostHog, Plausible or Piano). Returns per-day users, sessions and conversions grouped by AI source, plus totals and conversion rate. Requires a connected web analytics provider. Available on Scale and above plans.

Return one of the LLM Pulse guided-analysis playbooks as text. Use it when the user asks in natural language for an audit, a competitor gap analysis, citation opportunities, a reputation review, an explanation of a visibility change, or content briefs, then follow the returned playbook step by step. Clients with MCP prompt support get the same playbooks via prompts/list.

Get detailed information about a single AI answer/response including full response text, mentions, citations, sentiments, and sources.

Get sanitized cached page content plus page-cache metadata and mention snippets for one cited URL. Text longer than 15,000 characters is truncated (content.text_truncated=true, full length in content.text_total_chars); use max_chars to shrink the slice and offset to page through the rest.

Get detailed URL-level citation intelligence for one cited URL, including page-cache metadata and mention evidence.

Get detailed information about a competitor including matching names, mobile app IDs, and icon URLs.

Get a GEO Writer (formerly Content Intelligence) task by ID, including status and result data when completed.

For responses where a given source domain is cited, returns the share of those responses that mention the brand vs each competitor (brand + competitors sum to 100% per domain). Accepts a list of domains so the whole matrix is one call. Answers questions like "when gmac.com is cited, which brands get mentioned?".

Get detailed information about a project: matching names, industry, business model, primary products, target audience, prompt counts per brand focus (brand / brand_other / non_brand), and data_coverage (the AI models, countries and languages that actually have data). Call this right after list_projects: it replaces separate list_models / list_locales calls and tells you the non_brand prompt pool size before filtered queries.

Read a project draft: state, current step, accumulated data. include_suggestions returns cached suggestions for the current step (never triggers AI).

Get per-prompt aggregated metrics (responses, mentions, citations, mention_rate, citation_rate, avg_mention_position, avg_position) with pagination and sorting. mention_rate (also accepted as visibility) is the percentage of responses mentioning the brand. Returns brand-only metrics broken down by individual prompt. Supports breakdown=model to get per-prompt per-model metrics. Field notes: visibility and mention_rate are the same value (historical alias, both kept for compatibility); avg_mention_position is the average rank of the brand among the brands named WITHIN the answers that mention it, and it is the position behind ai_visibility_score; avg_position is a different thing, the average rank of the brand cited URL among the URLs an answer cited (background citations excluded).

Get full details for one recommendation run, including items, source references, and report data. Each item carries source_refs codes (e.g. P1, C2); resolve them against the report_data indices (prompts_index, citations_index, own_content_index), which are always included once per response.

Get the scores of one reputation report as flat rows: one row per (analyst model, brand, dimension, attribute) with its 0-100 score and the reasoning the model gave. Call list_reputation_reports first to get a report id. Scores come from several analyst models independently, so compare models rather than averaging them blindly, and filter with model= when you want a single view. Requires reputation monitoring to be enabled on the account.

Top Google Search Console landing pages for a project over a date range, ranked by impressions, clicks, CTR or average position, with pagination. Pass filters to narrow the list (for example dimension=query operator=contains expression=pricing), search_type to measure a surface other than web, and data_state=all to include the most recent still incomplete days. For exact property totals use get_search_console_summary. Requires a Search Console connection.

Top Google Search Console search queries for a project over a date range, ranked by impressions, clicks, CTR or average position, with pagination. Pass filters to narrow the list (for example dimension=page operator=contains expression=/blog/), search_type to measure a surface other than web, and data_state=all to include the most recent still incomplete days. Knowingly undercounts anonymized queries; for exact headline numbers use get_search_console_summary. Requires a Search Console connection.

Google Search Console headline totals (impressions, clicks, CTR, average position) for a project over a date range. Pass dimension=country or dimension=device to also get the breakdown aggregated over the range (country values are lowercase ISO alpha-3 as returned by Google, e.g. usa, gbr, plus the zzz unknown-country sentinel). The response data_through field marks the last day with synced data: Google publishes with a 2-3 day lag, so later days are missing, not zero. Requires the project to have a Search Console connection.

Google Search Console property-wide time series (impressions, clicks, CTR, average position) bucketed by day, week or month. The response data_through field marks the last day with synced data: Google publishes with a 2-3 day lag, so trailing days are missing rows, not real zero-drops. Requires the project to have a Search Console connection.

Get Share of Voice metrics - shows the relative share of mentions between the project and competitors over time. Returns current share percentages and breakdown by actor, plus a periods array with the sample size (mentions) and a partial flag per bucket: shares in a bucket with 1-3 mentions read as 100/50/33.33 and should be treated as low-sample noise, and partial buckets are still collecting data. Use group_by=model to get the per-model share comparison in ONE call instead of one filtered call per model. For a fair brand-vs-competitor comparison, set brand_kind=non_brand (the in-app Overview default): brand-focused prompts skew the share toward the brand they name.

Get one custom AI study: its brief, the subjects it compares, the dimensions it scores them on, and its report history. Use the report ids from the `reports` array with get_study_report to read the actual scores. Requires reputation monitoring to be enabled on the account.

Get the scores of one custom-study report as flat rows: one row per (analyst model, subject, dimension, attribute) with its 0-100 score and the reasoning the model gave. Call get_study to list a study reports and pick an id. Scores come from several analyst models independently, so compare models rather than averaging them blindly. Requires reputation monitoring to be enabled on the account.

Get summary statistics for metrics over a time period. Returns one row per metric and actor with total (count metrics are summed; percentage/rate metrics are averaged across periods, never summed), min, max, and last values. Use group_by to get one summary per AI model or per collection in a single call instead of fanning out multiple filtered calls. When comparing the project against competitors 1:1, set brand_kind=non_brand (the in-app Overview default) so brand-focused prompts do not skew the comparison.

Get one technical GEO report by type and id, including its current status and the full result_data once completed. Use agent_readiness for the AI/Agent Readiness report. Poll this tool after create_technical_geo_report; if poll_after_seconds is present, the report is still running. Never launch a duplicate report just to check progress.

Get time series metrics data for a project. Supports metrics like mentions, citations, mention_rate (also accepted as visibility), citation_rate, responses, ai_visibility_score (position-weighted visibility, unbounded: it can exceed 100 when several brand mentions rank high in one period), avg_position, net_sentiment, and sentiment breakdowns. Returns data points over time for the project; competitor series are only included when you pass competitors (ids from list_competitors). With week/month granularity the first bucket covers only the in-window days (partial periods are clipped, not back-filled). When comparing the project against competitors 1:1, set brand_kind=non_brand (the in-app Overview default) so brand-focused prompts do not skew the comparison.

Get top performing source domains for a project. Returns domains ranked by citing responses; avg_visibility (alias avg_mention_rate) is the percentage of all responses in the window that cite the domain at least once (bounded 0-100). It is a property of the DOMAIN, not of the brand: never present it as the brand's visibility "on" or "within" that domain (use get_mentions_by_citing_domain for brand share conditioned on a citing domain).

Get sample webhook payloads for an event type, built from the project's most recent real data (or a static sample when no data exists). Use this to preview the exact payload shape before creating a webhook subscription with create_webhook_subscription. Available on Scale and above plans.

Launch a new AI-powered recommendations generation for a project (same engine as the in-app Recommendations page). This is a REAL, quota-consuming action: it spends the project weekly recommendation-item budget (shared across all types, admins exempted) and runs a 1-3 minute background job. Never call it speculatively: only when the user explicitly asked to (re)generate recommendations, and after telling them it consumes the weekly budget. In the in-app chat the user gets a Confirm/Cancel card first; external MCP clients launch immediately, so get the user approval in YOUR conversation before calling. Track progress via get_recommendation / list_recommendations (status pending -> processing -> completed).

List the paid placements AI answers returned for your tracked prompts. view=advertisers (default) returns one row per advertising domain with its placement count, how many prompts it appeared on, and its average and best position; view=ads returns the individual placements with title, snippet, position and the prompt that triggered them. Every response also carries a totals block (placements, advertisers, your placements and share, average position) matching the KPI cards in the app. Position 1 is the best slot, so a LOWER average position is better. Available on Scale and above plans.

List the catalog of known AI bots (crawlers, assistants, search fetchers) and their parent companies. Use this to discover valid bot slugs and company names for the bot/company filters of get_agent_traffic. Returns each bot's slug, display name, company, category, and description.

List all citations (brand + competitor) for a project with actor_type field. Returns paginated results combining project and competitor citations. The competitors filter narrows which competitor rows are included, but project rows are ALWAYS included. position 0 means unranked/background (real positions are 1-indexed).