Our Methodology

Transparency matters. This page explains exactly how videos enter our directory, how they get classified, and what determines whether a hub page is published. No black boxes.

Step 1: Channel discovery

Our pipeline discovers ASMR channels through multiple strategies. We run targeted searches for specific trigger and intent combinations (like "rain asmr sleep" or "tapping asmr tingles"), prioritized by search volume. We also monitor trending ASMR content weekly to catch new creators as they emerge. Every channel we discover gets queued for full library ingestion.

Step 2: Video ingestion

For each discovered channel, we walk through their entire uploads playlist and pull metadata for every video: title, description, duration, view count, and publish date. We use a short-circuit mechanism — if we encounter three consecutive videos we have already processed, we stop paginating that channel. This keeps the system efficient while ensuring we catch new uploads.

Step 3: AI classification

Every ingested video is analyzed by an AI model that examines the title, channel context, duration, and description. The model determines:

• Primary trigger — the dominant ASMR trigger (tapping, whispering, rain, etc.)
• Secondary triggers — up to five additional triggers present in the video
• Intents — what the video is useful for (sleep, study, relaxation, tingles, etc.)
• Intensity — rated 1 to 5, how strong the ASMR response typically is
• Format — whether the video is ambient, roleplay, or another format
• Duration bucket — categorized from 5 minutes to 10+ hours
• Language — detected and normalized
• Confidence score — how certain the model is about its classification

Step 4: Validation

Raw AI output goes through strict validation before anything gets published. Every trigger must match our approved taxonomy of 36 trigger types. Every intent must be from our 13 recognized intents. Intensity must be a whole number between 1 and 5. Duration buckets must match our standard set. Roleplay videos must specify a recognized scenario.

Videos that fail validation get flagged with a specific failure reason and held for retry. Only videos with a confidence score of 0.7 or higher and a positive ASMR classification get published to the directory.

Step 5: Hub intelligence

Once videos are classified, we compute aggregate statistics for each intent-trigger combination. A hub like "Sleep / Rain" gets metrics computed from every published rain video tagged for sleep: average intensity, percentage over one hour, percentage that are no-talking, duration distribution, and the top contributing creators.

A hub page only becomes eligible for our sitemap when it reaches 15 or more verified videos. This threshold exists to ensure every page we put in front of search engines offers genuine value — not a half-empty collection.

Step 6: Ongoing maintenance

The directory is not a one-time snapshot. Every day, our pipeline:

1. Searches for new videos matching our taxonomy
2. Ingests new uploads from known channels
3. Classifies pending videos
4. Checks for dead or removed videos and removes them
5. Recomputes hub statistics with fresh data
6. Fills gaps in underfilled collections with targeted searches

We also monitor for semantic drift — if the average confidence score of our classifications drops significantly over a rolling window, the pipeline automatically pauses to prevent quality degradation.

What we do not do

• We do not accept paid placements or sponsored rankings
• We do not manually override classification results
• We do not create pages for trigger-intent combinations that do not make practical sense (no "tapping for tinnitus" pages)

Questions

If you have questions about how a specific video was classified or why a particular collection exists, reach out through our contact page.