ANALYTICS4YOU
FeaturesHow it worksGuidesTutorialsPricingContact
Menu
FeaturesHow it worksGuidesTutorialsPricingContactLog in
Log inGet started
HomeFeaturesGuidesTutorialsPricingContactAboutDocsllms.txtYouTubevs vidIQvs TubeBuddyvs ViewStatsvs Monitor YTLog inSign upTermsPrivacyRefundsDelivery
ANALYTICS4YOU

Competitive intelligence for YouTube creators. Track every channel in your niche and see what changes, as it changes.

India

support@analytics4.you

© 2026 Analytics4You. All rights reserved. Prices are in USD.

Analytics4You is an independent service. It is not affiliated with, endorsed by, or sponsored by YouTube or Google LLC.

// docs

How detection works

An A/B test and a title change look alike from one viewer. The difference is in when observations disagree, and this page explains how that is measured.

Markdown version · llms.txt

docs

  • Getting started
  • Event types
  • How detection works
  • Limits

on this page

  • Two kinds of difference
  • Independent identities
  • The Atom feed
  • Publish times
  • Counts
  • Derived events

Two kinds of difference

  • Within the same short window, between viewers. Different people are shown different titles or thumbnails right now. That is an A/B test.
  • Between moments, for everyone. Every viewer now sees something different from before. That is the creator changing the video.

A single observer can never see the first kind: one page load cannot disagree with itself.

Independent identities

For each recent upload, a probe cycle loads the watch page several times, one after another, each from a fresh, cold browser identity that shares no session and no cookies with the others. If they disagree, the video is under test.

  • The identities are never batched into one session. When capacity is short, cycles run less often. The number of identities never goes down.
  • Probing covers uploads younger than 21 days.
  • A video that has shown no test slows down: half as often after 7 days, a quarter as often after 14. Any test sighting keeps it at full speed.
  • The uploads tab is a second detector. A title or thumbnail that comes back to a version shown recently is recorded as an A/B test, not as two changes.
  • Thumbnails are compared as pictures, by perceptual hash. A distance of 6 or less is the same image, so re-encoding is not a change.

The Atom feed

YouTube publishes a public Atom feed of each channel's newest uploads. It is plain HTTP, not the YouTube Data API, and it is read at most once every 5 minutes per channel.

  • It finds new uploads and gives their exact publish time, to the second.
  • It is never used as an A/B signal, and it never sets the title or thumbnail the product shows. Those come only from what viewers are actually served. A video first found in the feed gets its served title from the first sweep or probe.
  • It lists only about the 15 newest uploads, so older videos are dated another way (below).

Publish times

  • From the feed: exact.
  • Otherwise: derived from YouTube's relative text ("3 days ago"), which is not precise enough for an hour of day. So an upload's hour is shown only if the monitor first saw it within 6 hours of going live, and a gap between two uploads only if both were seen within 7 days. Everything else is shown as unknown rather than estimated.

Counts

  • The uploads tab shows views in YouTube's short form (1.2M), so sweep counts are rounded.
  • Probes read the exact count from the watch page.

Derived events

  • Breakout: an upload 8 to 72 hours old gaining views at least 3× as fast as the channel's own uploads did at the same age, and already doing so 6 hours earlier. At least 1,000 views.
  • Milestone: a subscriber count crossing a round number, dated at the crossing.
  • A/B result: reported 72 hours after a test was last seen, and only if the video's current title or thumbnail matches a tested version. A test that resumes is not reported twice.
  • Rename and new avatar: reported after two agreeing observations in a row. An empty or missing header is never read as a change.