Data coverage
The honest numbers: how often analysis succeeds, what analysis_status means, and what the niche catalogue does and does not carry.
Analysis success rate: ~89–97% per month
Measured across 2026: the share of catalog rows that reach a completed analysis has run between roughly 89% and 97% in any given month — comparable to the ~94% figure the leading competitor in this space publishes. Not every analysis succeeds; the pipeline can fail on a given video (an unreachable URL, a downstream processing error) or leave it pending. analyze_tiktok_video is safe to retry — a video that already holds a catalog slot from a prior failed attempt does not spend a second one.
What analysis_status means
complete— analyzed under the current schema; every field group (hook, timing, scenes, transcript, ...) is present.partial— analyzed under an older schema, before June 2026. Themissingarray names which fields the pipeline didn't yet extract at the time — not a defect specific to that video. See below.pending— a job is still running. Poll again.failed— the pipeline hit an error on this video. Check the accompanying error/failure detail.not_analyzed— nobody has ever requested analysis for this video.
partial and failed are different things and the API is careful to say which — an older schema is not a broken video, and treating the two the same in a client would tell your users something untrue.Scenes, timing and transcript: 100% only since June 2026
Coverage is uneven by vintage. Scene-by-scene breakdown, hook timing and full transcripts are present on effectively 100% of rows analyzed since June 2026, and sparse on almost everything analyzed before it (well under 15% for videos from early 2026) — those fields simply weren't extracted by the pipeline that far back. A request for scene timing on a pre-June-2026 video returns an empty array with analysis_status: "partial" and missing naming what's absent — check that field before assuming the video had no notable scenes.
The niche catalogue carries no per-video analysis
Measured 2026-09-06: 26,745 rows in the niche catalogue, of which only 31 (0.1%) have ever reached full AI analysis. browse_niche_catalog (GET /v1/niches) and GET /v1/niches/{id}/videos return thumbnails, engagement stats and a multiplier ranking — never a hook type, scene breakdown or transcript. To get real analysis on a catalogue video, call analyze_tiktok_video on it directly, which spends a standing catalog slot exactly like any other video (see Rate limits & quotas).
create_niche or browse_niche_catalog to find ranked candidates by multiplier, then analyze_tiktok_video on the best one for the full breakdown. Analyzing every result in a catalogue page can burn 10% of a Creator workspace's lifetime catalog in a single loop — be deliberate about which videos are worth the slot.YouTube Shorts: in the app, not yet in the API
Since September 2026 the web app analyzes YouTube Shorts and the Niches wall mixes viral Shorts in with TikToks. The API and MCP server have not caught up: analysis endpoints accept TikTok links only, and creator tracking, breakouts, trend search and the growth scorecard are TikTok-only everywhere.
GET /v1/niches/{id}/videos (get_niche_videos), GET /v1/videos (search_videos) and GET /v1/videos/{id}/analysis return a platform field: tiktok or youtube. Check it before calling analyze_tiktok_video, which can't analyze a youtube row; use the web app for those. save_video (POST /v1/library) does save Shorts: pass platform, or omit it and a YouTube-shaped id is saved as a Short.