New: AI extraction now runs with concurrency 6 across your project queue.Learn more
AI-powered YouTube research for teams

Turn Any YouTube Channel Into Structured Research Data

Import channels, transcribe every video with Gemini 2.5-flash, and extract Q&As, facts, sponsors and studies automatically. Your team searches one database, not fifty playlists.

2,000+
Channels tracked
500K+
Videos analyzed
1M+
AI insights extracted
99.9%
Uptime across 10+ cron jobs

Research at YouTube scale breaks human workflows

  • Pulling 500 channel stats by hand takes 14 hours and the numbers go stale the same day.

  • Watching a 90-minute interview to find one study citation wastes 89 minutes you will never get back.

  • Your team's research lives in four Notion docs, three spreadsheets and one person's browser history.

You need one database where every video is already transcribed, tagged and extractable. That is what this does.

Core capabilities built for research

Six primitives that turn raw YouTube channels into queryable data.

Channel import at scale

Import any public channel by URL. We pull metadata, videos, thumbnails and stats, then keep them fresh on adaptive schedules.

AI video extraction

Transcribe with Gemini 2.5-flash, then extract Q&As, facts, testimonials, studies, sponsors, people and products with GPT-4o.

Trigram full-text search

Search millions of titles and descriptions with PostgreSQL trigram GIN indexes. Results return in milliseconds, not seconds.

Viral score detection

Surface videos outperforming their channel baseline. Sort by view velocity, engagement ratio or custom median-duration filters.

Mind maps with RBAC

Visually connect insights, people and studies. Share boards with OWNER, EDITOR, COMMENTER and VIEWER roles.

Team workspaces

Group channels into projects. Share research across teammates with multi-user permissions, audit logs and real-time activity.

AI extraction that reads every video for you

Drop a channel into a project and every video gets a full Gemini 2.5-flash transcript within minutes. GPT-4o then reads the transcript and pulls structured entities into your database. You never open the video unless you want to verify a claim.

  • Extract Q&As, facts, testimonials, studies, sponsors, people, products per video
  • Store transcripts and summaries in PostgreSQL for cross-channel queries
  • Classification jobs run in background with concurrency 6 and retry logic

So you can cite a specific claim from 200 videos without scrubbing a single timeline.

Screenshot: AI extraction

Channel intelligence that updates itself

Channels import with an "auto" refresh frequency. If a creator posts daily the system checks every 12 hours. If they post monthly it checks every 14 days. YouTube API quota gets spent where it matters.

  • Adaptive update frequency from 12 hours to 14 days based on publishing cadence
  • Per-user quota rotation across multiple API keys with automatic fallback
  • Quota snapshots every 2 hours feed a full analytics dashboard

So you can track 2,000 channels without burning through a 10,000-unit daily quota by 10am.

Screenshot: Channel intelligence

Screen page for filtered discovery

The /screen page is the research surface. 23 column filters, three query paths (tag, keyword, standard) and a CTE-based search strategy that narrows 500K videos before running ILIKE. Built for power users who need fast answers.

  • 14 channel filters and 9 video filters stack on top of sidebar controls
  • Median duration filters separate shorts, mids and long-form on one grid
  • Trigram indexes make %keyword% searches 50-200x faster than naive scans

So you can isolate "5 minute interview clips from fitness channels mentioning creatine" in under a second.

Screenshot: Screen page

Every feature, across six categories

From import to extraction to team sharing — the complete research stack.

Channel Intelligence

Every channel you import stays fresh, scored and comparable against its own baseline.

  • Import by URL or search
  • Adaptive update frequency (12h–14d)
  • Subscriber, view, like, comment tracking (BigInt-safe)
  • madeForKids detection from YouTube API flag
  • Thumbnail archiving
  • Upload cadence analysis
  • Channel-level engagement ratios
  • Historical stats snapshots every 2 hours
  • Managed playlists per channel
  • Channel grouping via projects

AI Content Extraction

Gemini transcribes, GPT-4o structures, PostgreSQL stores – per video, per channel, per project.

  • Gemini 2.5-flash transcription (65K token output)
  • GPT-4o structured extraction
  • Q&As per video
  • Facts and claims
  • Testimonials with attribution
  • Study and research citations
  • Sponsor and affiliate detection
  • People, products, books, podcasts mentioned
  • Summaries per video
  • Smart JSON repair on malformed model output
  • Classification jobs with concurrency 6
  • Re-run extraction with updated prompts

Discovery & Research

Query 500K+ videos across keywords, tags, duration buckets and engagement thresholds.

  • Trigram GIN indexes on titles and descriptions
  • 23 stackable column filters on /screen
  • Keyword + tag combined queries
  • Median duration bucketing
  • Viral score sorting
  • Project-scoped or global search
  • Date range filters
  • Channel-set filtering
  • View/like/comment thresholds
  • Saved column presets

Knowledge Management

Turn video insights into a linked research graph your team can cite from.

  • Mind maps with nodes and edges
  • TipTap rich text notes
  • Q&A library across channels
  • Facts database
  • Studies and sources library
  • People profiles with enrichment
  • Product catalog
  • Books, podcasts, magazines registries
  • Glossary and topics
  • Watch later and watch history
  • Affiliate link tracking
  • Graveyard for removed content

Automation & Alerts

Cron-driven scheduler runs 10+ jobs with database locking, quota awareness and audit logs.

  • Host cron + scheduler endpoint
  • Channel-updates job (adaptive)
  • Video import job
  • AI classification job
  • Quota snapshot job
  • RSS feed collection
  • Email notifications via Resend
  • Push notifications
  • Error log with 10d/30d retention
  • Admin quota priority mode (pause crons)
  • Real-time socket.io activity
  • Fire-and-forget audit logging

Team & Enterprise

Built for teams that share channels, projects and research across multiple researchers.

  • Multi-user permission model via ProjectChannel
  • RBAC roles: OWNER, EDITOR, COMMENTER, VIEWER
  • Team workspaces
  • Audit log per action
  • Real-time activity feed
  • Google OAuth + credentials auth
  • Admin dashboard at /admin
  • User quota management
  • Database maintenance UI (VACUUM, ANALYZE)
  • Error log viewer with resolution workflow
  • Cron log monitoring
  • Role promotion controls

Built for six kinds of researchers

If your job requires citing what people said on YouTube, you are the target.

Content strategy agencies

Track every channel in a client's niche and surface trending hooks weekly. Extract claims competitors make so strategists can counter-position without rewatching videos.

  • Weekly trend reports in 15 minutes
  • Competitor claim library per client
  • Hook patterns across 200 channels

Academic researchers

Build citation databases from YouTube interviews and lectures. Pull study references and speaker claims into structured rows ready for qualitative coding.

  • Citation extraction across 500 lectures
  • Speaker attribution on every quote
  • Export to CSV for NVivo or Atlas.ti

Journalists and investigators

Monitor public figures and channels for claims, sponsors and contradictions. Get alerted when a tracked channel publishes and surface the transcript within minutes.

  • Claim timeline per public figure
  • Sponsor disclosure tracking
  • Fast transcript access under deadline

Content creators studying competitors

Load 50 competitor channels into a project and sort by viral score. See which topics outperform each creator's baseline and mine hooks that worked.

  • Viral outlier detection per channel
  • Hook and title pattern library
  • Topic gap analysis across niche

Marketing analysts

Extract sponsor mentions, affiliate links and product placements across industries. Report on which creators drive which brand categories.

  • Sponsor mention database per vertical
  • Product placement frequency tables
  • Brand-creator affinity mapping

Internal research teams

Share one channel library across researchers with role-based access. Prevent duplicate imports and keep notes, tags and extractions in one workspace.

  • One source of truth per team
  • Role-scoped access to sensitive projects
  • Audit trail for compliance reviews

How it works

01

Import channels

Paste a YouTube channel URL or search by name. We pull metadata, subscriber counts, every public video and thumbnails in one background job.

02

Auto-transcribe

Gemini 2.5-flash reads each video directly from its YouTube URL and produces a transcript inside your PostgreSQL database, no manual upload required.

03

Extract insights

GPT-4o parses transcripts for Q&As, facts, studies, sponsors, testimonials, people and products, writing each as a structured row.

04

Search and share

Query across channels with trigram full-text search, visualize connections in mind maps, and share projects with your team via role-based access.

Frequently Asked Questions

How does CreatorInsights extract data from YouTube videos?

CreatorInsights sends each video URL directly to Gemini 2.5-flash, which generates a transcript with up to 65,536 output tokens. GPT-4o then reads that transcript and returns structured JSON containing Q&As, facts, testimonials, studies, sponsors, people and products. Every extraction is stored as rows in PostgreSQL so your team can query across every video in a channel, project, or entire workspace.

What AI models power the video analysis?

Two models run the pipeline. Gemini 2.5-flash handles transcription because it accepts YouTube URLs natively and returns long-form output cheaply. GPT-4o handles structured extraction because its JSON mode is reliable across entity types. The codebase locks both model choices to avoid drift, and smart JSON repair logic catches malformed model outputs before they hit the database.

Can I track channels my team also researches?

Yes. Channels are global objects in CreatorInsights, not user-owned. You add a channel to a project, and every teammate with access to that project sees the same videos, transcripts and extractions. This prevents duplicate imports and avoids burning API quota on channels someone else already loaded. Access is enforced through the ProjectChannel model with four RBAC roles: OWNER, EDITOR, COMMENTER and VIEWER.

How much does CreatorInsights cost?

Pricing is tiered. A free tier lets individuals import a limited number of channels and run AI extraction on their own YouTube API quota. Paid plans unlock higher channel limits, team workspaces, shared quota pools, admin controls and priority classification. Check the /pricing page for current numbers or start on the free tier and upgrade when you hit a limit.

How does the platform handle YouTube API quota limits?

Every user gets a 10,000-unit daily quota from YouTube that resets at 00:00 Pacific time. CreatorInsights rotates across user keys automatically: your key is tried first, then other users' keys, then a system default. A soft limit kicks in at 9,900 units to preserve buffer for personal operations. Quota usage is tracked per call in the YouTubeApiCall table, and admins monitor consumption at /admin/quota-analytics with 2-hour snapshots.

What does the viral score detect?

Viral score compares a video's view velocity and engagement ratio against its own channel's baseline, not against all YouTube. A mid-size channel hitting 10x its normal views scores higher than a mega-channel hitting 2x. You can sort any video list by viral score, stack it with median-duration filters, and isolate outliers inside a single niche.

Can I export research out of CreatorInsights?

Yes. Videos, transcripts, insights and channel lists export to CSV and JSON. Mind maps export to PNG and JSON. The underlying PostgreSQL schema is documented in db_structure/ so teams running their own analytics pipelines can connect directly if they have database access on their plan.

Is CreatorInsights for creators or for researchers?

Both, but the feature weight favors researchers and analysts. Creators use the viral score and trend dashboards. Researchers, journalists and marketing analysts use the AI extraction, citation libraries, mind maps and team workspaces. If you only want thumbnail optimization, this is the wrong tool. If you want structured data out of 500 videos, this is exactly the tool.

How fresh is the channel data?

Channels refresh on an adaptive schedule between 12 hours and 14 days. A creator publishing daily gets checked every 12 hours. A creator publishing monthly gets checked every 14 days. This keeps API quota spent where it actually matters and avoids wasting 2,000 calls on dormant channels. Admins can override frequency per channel.

What happens to my data if I cancel?

Your imported channels, transcripts, insights and mind maps stay in your workspace during the cancellation grace period. You can export everything to CSV or JSON before your workspace is archived. Channels imported by other users remain in the shared pool since they are global objects, but your private projects, notes and team settings are removed on final deletion.

Stop scrubbing.
Start querying YouTube.

Import your first channel, get transcripts and extracted insights in under ten minutes. Free to start.