A 39-minute conversation on YouTube's own creator channel took apart the idea that a video's fate is settled in its first hour, and the figure left behind is a click-through ceiling that most channels never cross even among the people who subscribed to them.

Creator Insider, the channel Google uses to brief creators on product mechanics, published the interview on September 1, 2026 under the title "YouTube Algorithm 2026 - What Creators Need to Know". The guest is Todd Beaupré, senior director of growth and discovery at YouTube; the host is creator liaison Rene Ritchie. Released as part of the Inside YouTube Podcast series, the video runs roughly 39 minutes and 30 seconds across 13 marked chapters. At the time of capture the page listed 14,060 views and 190 comments, against a channel subscriber count of 909,000.

The page discloses that audio tracks for some languages were generated automatically, which is why the machine transcript renders the participants as Todd Boprey and Renee Richie while the description gives Beaupré and the handle @ReneRitchie. A smaller inconsistency sits in the chapter markers: the description places the second chapter at 0:54, the transcript a few seconds later.

A pull system rather than a push system

According to Beaupré, the vocabulary creators use to describe distribution is wrong in a way that matters. Nothing is dispatched to an audience when a video goes live. "Most of the recommendations are more pulled for each viewer," he said, with notifications the single exception for people who have opted in.

The sequence he described begins on the viewer's side. When somebody opens the application or switches on a television, the recommendation system activates for that session and assembles a ranking in that moment. It draws on watch history, on the current context, on the time of day and on the device in hand, pulls in candidate videos and orders them. "Each viewer gets a different ranking," Beaupré said. The system is not looking at a video and searching for its audience; it is looking at a viewer and searching for content.

The consequence for reading a channel's numbers is direct. If ranking is computed per person per session, an aggregate performance figure summarises many separate decisions rather than a verdict on the video itself.

What the first hour does and does not settle

The most persistent creator belief Beaupré addressed is that an upload is tested on subscribers, and that a weak response there ends its run. Viewers with more history on a channel and a topic do rank it higher, he said, and see it more prominently. The data does not stop there. "We see diverse audiences" in that early window, he said, and "videos are reaching people who've never watched the channel before" inside the first hour.

More consequential is how the system reads the response. It does not compress the first hour into one performance number. It learns separately for different audiences, and can conclude that a video is underperforming with subscribers while performing with people who have never watched the channel. On that point Beaupré was explicit about the fear it addresses: "it's not only your subscribers who will see it," and the ranking is not driven "just based on an overall aggregate opinion." He added a line that reframes the whole question of aggregate taste: "The fact that someone else doesn't like something that I like shouldn't prevent me from seeing it."

Immediate analytics get a similar treatment. "We're not basing decisions just on those initial moments," Beaupré said, describing a continuous flow of impressions the system learns from. Some videos peak on day one. Others accumulate the bulk of their audience a month or more after publication, and evergreen material is not retired from consideration.

No single metric, and no fixed weighting

Asked what the system actually optimises for, Beaupré named a horizon rather than a number. The stated objective is long-term user value, measured by whether people return and feel the platform delivered. Below that, the answer is deliberately plural. "No metric on its own is a good indicator of value," he said.

Click-through rate captures the appeal of a thumbnail and an idea, and falls predictably as impressions widen beyond core fans. Watch time carries its own defect. "We don't want to just assume that all time is valuable," Beaupré said, describing a viewer who stays with a video waiting for a payoff that never lands. Surveys separate time spent from satisfaction. Likes, dislikes and comments each contribute a partial picture.

Then comes the passage that undermines every attempt to reverse-engineer a formula. Asked which signal carries the most weight, Beaupré said the weighting is not constant across people, and not constant for the same person over time. "It actually doesn't even work such that it's the same for you at all times," he said. Different factors carry different value on different surfaces: time spent may matter more on a television, while a like or dislike may carry more information on a phone, where the action is easier to perform.

The system is also built to avoid punishing configuration choices. Beaupré described the integration of comment signals as a case where the team took care not to read a disabled comment section as evidence that nobody engages with the content.

Repetition, and the eight-impression experiment

Survey feedback tells YouTube that viewers see the same recommendation too often. Beaupré said the team has tested hard limits, including an experiment in which "you wouldn't be recommended a video more than eight times". Engagement does sometimes arrive on the later attempt, which complicates the case for a cap.

One anecdote quantifies how invisible repetition can be. Working with a creator whose viewer appeared to be missing recent uploads, and with that viewer's permission, the team examined the logs. The uploads had been recommended 19 times. The viewer had not noticed. Somebody following 50 channels regularly may never scroll past the top of the feed.

The stated direction is contextual rather than numerical, matching a long video to an evening on a television rather than to a moment when it cannot be watched. "We don't want to waste your time," Beaupré said.

Puddles, rain, and who is pulling long videos

The interview's central metaphor arrived in the chapter on channel categories. "One of the biggest breakthroughs I had was realizing that puddles don't cause rain," Beaupré said. Applied to the platform, the causal arrow runs opposite to the industry's assumption: "people think that YouTube is pushing longer videos now, but people are watching on television and they want longer videos and so YouTube is recommending those longer videos."

The same logic covers short-form. Beaupré attributed its success across the market to decision fatigue rather than to platform preference, saying that viewers "sometimes feel overwhelmed with all the choices that they have" and that a swipeable feed "eases the burden of decisionmaking". On living-room screens, he said, "viewers don't want to be picking up the remote every five to 10 minutes to make a new video decision", which is why series formats hold. Growth at both extremes of duration follows from that, in his account, rather than from a thumb on the scale. "We want the data to tell us that answer," he said.

The claim sits awkwardly against documented interface changes. PPC Land reported in December 2025 that home feed changes reduced long-form discovery slots in favour of Shorts rows, and that Shorts had reached revenue parity with long-form per watch hour in the United States. Whether demand or a product decision drives a layout is not separable from outside the company, and the interview offers no data to settle it.

The desktop redesign, assessed after two years

Creators have asked whether the desktop homepage redesign and its larger thumbnails suppressed long-form viewing, on the reasoning that fewer thumbnails fit above the fold. Beaupré dated the change to roughly two years earlier and said the opposite happened: attention and engagement around long-form increased. "Often less is more," he said, describing an earlier assumption that density and choice would maximise the chance of a match. Mobile thumbnails used to be much smaller for that reason. Overwhelmed viewers, he said, sometimes choose nothing at all.

He said the change would not have shipped had it reduced viewership, and that testing preceded it. No figures accompanied any of that. The testing programme was described only as "hundreds of experiments" running at any time, compared to switchbacks on a mountain: reading direction from a single leg misses where the path goes.

Channels as brands, and cross-channel recommendation

On whether a creator changing topics needs a second channel, the stated guidance is to "align channels with audiences". If the same people would watch both, one channel works. If the value propositions differ, separate brands make the expectation legible.

The system is built to absorb the wrong choice either way. Cross-channel tactics route viewers between channels where audience overlap exists. Within one channel, the system distinguishes videos by format, length and topic, so a viewer who consumes only the short material is matched accordingly rather than served everything the channel publishes.

Beaupré also pushed back on treating a channel's best-performing video as the benchmark for everything after it, since different content carries different appeal and a new area takes time for audiences to accept.

Suggested videos outrank the subscriptions feed

The chapter on traffic sources contains the interview's most operationally useful disclosure. Creators inspecting the suggested-video traffic source often find unrelated topics driving views and read that as evidence the system has misclassified their audience. According to Beaupré, "the reason that a video can appear next to another video is not simply just because it's related". The watch page moved years ago from related videos to videos relevant for a range of reasons, most of them personalisation.

Then the figure. "The number one place that people are watching videos from channels they're subscribed to is actually on the watch page in that suggested list," Beaupré said. The subscriptions feed is, in his words, quite small by comparison. Viewers change channels the way they always did, and cross-topic traffic is a feature of that behaviour rather than a symptom of misclassification. "Many viewers watch hundreds of channels and they're not all related," he said.

For new channels with no viewer history, parts of the system operate without it, matching topical preferences against what a video is understood to be about. If no interested audience emerges, impressions stop. Beaupré said YouTube tracks how many videos get seen at all, putting the daily figure at "hundreds of millions of videos that get seen by someone", and whether creators reach sustainable revenue.

The 10 percent ceiling

The final chapter produced the number that gives this interview its weight for anyone pricing a channel. Beaupré directed creators to the subscribers row inside the browse features traffic source, on the grounds that it isolates one audience whose composition does not change between uploads, which makes it a clean read on content rather than on targeting.

What they will find there, he said, is that "probably all your videos have about maybe a 10% or less click-through rate". The implication follows immediately: "even your best videos, 90% of the time, your subscribed audience isn't deciding to watch."

Two conclusions followed. The subscribe action is weaker than assumed, in Beaupré's words it "isn't as strong as a lot of people think". And the pattern is ordinary: "It's normal for an audience that has been interested in your channel to not be interested in every upload," he said, adding that the system is built to find the subset of subscribers likely to want each specific video. His closing instruction on analytics was to avoid overindexing on aggregate statistics.

Ritchie added the audience-composition counterpart from his own analytics: a channel drawing a consistent count per video is not drawing the same people each time, and the share watching most uploads is very small. Beaupré agreed that creators underestimate how many other channels their fans watch.

What the video does not disclose

The session explains mechanics without supporting data. No figure is given for the share of first-hour impressions reaching non-subscribers, the sample behind the 10 percent click-through observation, the outcome of the eight-impression experiment, the size of the long-form gains attributed to the desktop redesign, or the date that redesign shipped. The daily count of videos seen is an order of magnitude. Nothing is tied to a published methodology, and the format allows no challenge: the interviewer is a YouTube employee and the venue is a channel the company operates.

Creator response in the comments

The 190 comments captured on the page do not read as an audience persuaded. One creator with 21,900 subscribers wrote that analytics showed seven people received a notification for a new long-form upload. Another, six years into publishing, described subscribers with notifications set to All receiving nothing and having to search the channel, a problem the commenter said had gone unaddressed for years.

A third account contradicts the interview's central reassurance. Writing under the handle @DOPEFOLKTALES, a creator reported recommendations narrowing to existing subscribers over several months and that "my overall views have dropped by about 65%", with the subscriber share rising to between 80 and 90 percent against roughly 20 percent on older, better-performing uploads. A television viewer complained that the Recommended row repeats for days.

Those accounts are unverified individual reports, and they describe an experience the interview does not map onto.

Why this matters for the marketing community

For anyone buying creator inventory, the operative disclosure is that a channel's aggregate metrics describe an outcome rather than a mechanism. If ranking weights vary by person, device and moment, no benchmark click-through or retention figure is a property of the channel that transfers to a sponsored upload. A subscriber count, on Beaupré's own numbers, predicts a click roughly one time in ten inside the feed built specifically for subscribers.

That lands in a year when the underlying counters have already moved. YouTube began counting views from the first frame across all formats on August 24, 2026, inflating public totals while leaving payment on engaged views, and PPC Land noted the effect on cost-per-thousand pricing for brand deals when a counter rises without a change in attention. Two weeks earlier the company doubled Partner Program entry thresholds to 8,000 qualified watch hours or 20 million qualified Shorts views, effective February 1, 2027.

The transparency question sits above all of it. France's competition authority found in February 2026 that 80 percent of creators hold weak or very weak bargaining power relative to the platforms distributing their work, and issued recommendations on algorithm transparency. X published recommendation source code in January 2026, which analysts read against redacted weighting constants. YouTube's contribution to the same debate is a video interview conducted by its own staff, with metaphors where the numbers would be.

Timeline

Summary

Who: Todd Beaupré, senior director of growth and discovery at YouTube, interviewed by creator liaison Rene Ritchie on the Creator Insider channel, which lists 909,000 subscribers. The audience is creators, channel managers, agencies sourcing creator inventory, and brands pricing sponsorships against channel metrics.

What: A 39-minute interview across 13 chapters setting out how the recommendation system ranks video. The substantive claims are that recommendations are computed per viewer at session time rather than pushed to an audience, that early performance is learned separately for subscriber and non-subscriber cohorts, that no single metric governs ranking and no fixed weighting applies across people, surfaces or moments, and that click-through rate inside the subscribers traffic source typically sits at 10 percent or below.

When: Published September 1, 2026, with 14,060 views and 190 comments recorded at the time of capture.

Where: The Creator Insider channel on YouTube, as part of the Inside YouTube Podcast series, at https://www.youtube.com/watch?v=rHLjxrbXmmY. Distribution is global, with automatically generated audio tracks for some languages.

Why: Creator questions about early-hour testing, subscriber signals, topic changes and traffic sources recur constantly, and the company has faced sustained criticism over algorithmic opacity since documented view drops in August 2025. The interview answers those questions without publishing supporting data, in a venue the company controls, at a point when view counting and monetisation thresholds have both moved.