Every year someone announces that "the algorithm changed" and every year the underlying logic stays the same: YouTube shows a video to a small group, measures what they do, and shows it to a larger group if the numbers hold. What does change is the weighting — which viewer behaviours count for most — and in 2026 the weighting has moved decisively toward session-level signals: whether your video makes someone start watching, keep watching, and watch something else afterwards.
This article explains the eight signals that matter most this year, how they interact, and what a creator can actually do about each of them. It is written from public YouTube statements, creator-liaison Q&As, and what we see across the channels that run VidCola reports — not from a leaked ranking formula, because there isn't one.
1. Click-through rate in context
CTR still opens the door, but it is measured against the viewer's own norm, not a global one. A 6% CTR from a viewer who clicks 2% of what they see is a strong signal; the same 6% from someone who clicks everything is neutral. This is why chasing raw CTR with louder thumbnails plateaus quickly: you attract the high-clickers and change nothing about distribution.
What to do: package for the viewer you want to keep, not the one most likely to click. Pair the title and thumbnail so they tell one story (a mismatch produces a click and an immediate exit, the worst combination).
2. Average view duration and percentage — together
Neither number alone is the target. YouTube looks at both absolute minutes and percentage because they answer different questions: is the video worth the platform's time, and did it deliver on its promise? A 25-minute video at 35% (8.75 min) and an 8-minute video at 70% (5.6 min) are both healthy; a 25-minute video at 15% is not.
What to do: choose length from the topic, not from a target. Put the first payoff before the 30-second mark and the biggest payoff around the midpoint, not at the end.
3. Session starts
A video that brings someone onto YouTube — from search, an external link, a notification — is worth more than one watched mid-session, because it created a session that would not otherwise exist. Channels with strong search and external traffic are quietly favoured for this reason.
What to do: build searchable evergreen videos alongside browse-driven ones, and drive traffic from outside (newsletter, community, social) on launch day. The First 48 Hours plan in VidCola schedules this.
4. Session continuation
The mirror image: does the viewer keep watching YouTube after your video — ideally your next video? End screens, playlists, series structure and a spoken hand-off all feed this. A video that ends a session (the viewer closes the app) is penalised relative to one that continues it, even at identical retention.
What to do: end with a specific next video, not a generic outro. Series and playlists are the cheapest continuation machine there is.
5. Satisfaction surveys and post-watch behaviour
YouTube periodically asks viewers to rate videos, and it models satisfaction for everyone else from likes, shares, saves, "not interested" clicks and whether the viewer returns to the channel. Clickbait that retains but disappoints shows up here first — it is how the platform separates "watched" from "valued".
What to do: deliver what the title promised, early and completely. The single biggest satisfaction lever is not over-promising.
6. Returning viewers
Studio now shows new versus returning viewers prominently because the platform weighs it. A channel that people come back to is a channel worth recommending to new people. Subscriber count matters less than active returning viewers.
What to do: consistency in topic and cadence. A recognisable format the audience can anticipate beats a string of unrelated hits.
7. Velocity in the first hours
The initial test audience is small and the decision to expand happens fast. Early engagement rate — comments, likes, watch time in the first hours relative to impressions — determines whether the second wave happens. This is not "the first hour decides everything", but it is the cheapest window to influence.
What to do: publish when your audience is active, reply to every early comment, pin a comment that invites a specific response.
8. Topic authority
The system increasingly reasons about channels as topic entities. A channel with twenty videos on home espresso is recommended for a new espresso video more readily than a channel with two, regardless of size. Straying off-topic dilutes this; a second channel is often the right answer for a second topic.
What to do: stay in your lane long enough to own it. Use a Channel Audit to see how tightly clustered your recent uploads are.
How they combine
None of the eight is a lever you can pull alone. High CTR with low satisfaction gets a video shown and then buried. High retention with no session continuation grows watch time but not the channel. The channels that grow in 2026 are the ones that treat these as a system: honest packaging, early payoff, a next step, and a recognisable home to return to.
Tips
- Judge every video by three numbers together: CTR relative to your norm, average view duration, and returning-viewer share.
- Design the ending before the beginning — the hand-off to the next video is a ranking signal.
- Build a searchable evergreen layer so your channel starts sessions rather than only riding them.
- Reply in the first two hours; it is the cheapest engagement-velocity boost available.
Common mistakes
- Optimising CTR in isolation — Louder thumbnails attract high-clickers and change nothing about distribution — and hurt satisfaction if they over-promise.
- Chasing a length target — Length should come from the topic; padding lowers percentage viewed and satisfaction together.
- Generic outros — "Thanks for watching" ends the session. Name the next video.
- Topic drift — Every off-topic upload weakens the channel's entity signal. Start a second channel instead.
Frequently asked questions
Did the YouTube algorithm change in 2026?
The mechanics are the same — small test audience, expand if metrics hold — but the weighting has moved toward session signals: session starts, continuation, returning viewers and satisfaction.
Is CTR still important?
Yes, but relative to the viewer's own click habits and always in combination with what happens after the click.
How long should my videos be in 2026?
As long as the topic needs. YouTube looks at minutes and percentage together; a padded video loses on both.
Does posting time matter for the algorithm?
Indirectly — publishing when your audience is active improves early velocity, which influences whether the video gets a second wave.
Do subscribers still matter?
Less than returning viewers. An engaged returning audience is a stronger recommendation signal than a large dormant subscriber count.