A YouTube channel can hold views flat and nearly double watch time in the same 28-day window. This note decomposes one such shift on the PushLeads channel (a local-SEO and contractor-marketing channel), separates what is real from what is a counting artifact, and lays out how the traffic-source mix connects to citation behavior in Google’s AI surfaces. It builds on two earlier pieces: How to Measure YouTube Citations in AI Overviews and Time-to-Citation for YouTube Videos in AI Search.
Headline result
For the 28 days ending November 15, 2026, YouTube Studio reported 3,613 views (+0.33% versus the prior 28 days), 69.2 hours of watch time (+90.02%), 20 net subscribers (+150%) and 172 published videos (+11.69%). YouTube search supplied 34.5% of views. External sources supplied 39.9% of watch time.
The useful number is the ratio. Watch time per view rose from roughly 0.61 minutes to roughly 1.15 minutes. Volume did not change. Depth of attention per session nearly doubled.
Data
All channel figures come from the YouTube Studio 28-day summary for the channel, read on November 17, 2026. Studio excludes the last two days of uploads from the published-video count because of processing delays, so the window is effectively closed on November 15.
Supporting context comes from two internal sources already described in earlier notes: the weekly AI-answer snapshot table (one row per prompt, engine and date, with full cited URLs) and a crawl of pushleads.com run on November 17, 2026, which found an embedded YouTube player on 1,356 of 1,648 crawled pages.
No new collection was run for this analysis beyond reading the Studio summary.
Definitions
- View (public). Since August 24, 2026, YouTube registers a public view when playback begins, on every format. Shorts moved to this rule in March 2025 (OutlierKit, 2026).
- Engaged view. What YouTube previously called a view. YouTube states that click-through rate, average view duration and retention are still calculated on engaged viewing, and that recommendations did not change (OutlierKit, 2026).
- Co-views. On November 16, 2026, YouTube Creator Insider announced that estimated co-views would be folded into public view counts across Shorts, long-form and live, with historical counts unchanged (OnlineMarketing.de, 2026).
- Watch time. Estimated hours watched, summed across all sources and formats.
- Watch time per view (WT/V). Watch time divided by views. A blended engagement-depth proxy, not a per-video retention metric.
Decomposition
Back out the prior period from the reported deltas:
- Prior views: 3,613 / 1.0033 ≈ 3,601
- Prior watch time: 69.2 / 1.9002 ≈ 36.4 hours
- Prior WT/V: 36.4 × 60 / 3,601 ≈ 0.61 minutes (about 36 seconds)
- Current WT/V: 69.2 × 60 / 3,613 ≈ 1.15 minutes (about 69 seconds)
Total watch time is views multiplied by WT/V. With views effectively constant, essentially all of the +90% sits in WT/V. There are three mechanisms that can move a blended WT/V without moving volume:
- Format mix. A lower Shorts share of views raises blended WT/V, because a Short caps out at seconds and long-form sessions run minutes. Publishing more long-form, or a long-form video catching search demand, pushes the blend up.
- Source mix. Sources differ in session depth. Search and embedded plays on topical pages carry intent. Feed-driven sources carry curiosity. Shifting share toward intent sources raises WT/V.
- Within-video retention. Same videos, same sources, people watch longer. This is the hardest to produce and the least likely sole driver of a 90% change in four weeks.
The Studio summary attributes the top viewership to an evergreen long-form pricing video (How Much Does Mold Remediation Cost? Real 2026 Pricing) and a recent Short on review removal. That combination, plus 39.9% of watch time from external sources, points to mechanisms 1 and 2. Treat mechanism 3 as unproven until per-video retention curves are compared across the two windows.
Is the view number contaminated by counting changes?
Mostly no, for this comparison. Both 28-day windows sit entirely after the August 24 first-frame change, so the definition of a public view is constant across them. The co-view change landed November 16, which is outside the window once the two-day processing exclusion is applied. Studio’s own summary flags the update as a possible small contributor to recent counts, which is consistent with a boundary effect rather than a material one.
The larger risk is cross-window comparisons that straddle August 24. A channel comparing September to July will see view inflation that has nothing to do with content, and the effect is not uniform: plays that bounce in the first seconds now count, so formats and sources with high early abandonment inflate most. Independent tracking also shows public counts are not monotonic. One study of 215,023 consecutive public-view readings across 111,625 videos found 0.27% of readings decreased, with Shorts showing more decreases than long-form (OverseerOS, 2026).
Rule: after August 24, 2026, use watch time, engaged views and average view duration for trend work. Use public views only for same-regime comparisons, and annotate November 16 on any chart that crosses it.
Traffic-source analysis
YouTube search: 34.5% of views
Search-sourced views are query-matched. The viewer typed a question and chose a result whose title matched it. On this channel the titles are question-shaped by design (“How much does X cost”, “Why won’t Google remove Y”), so search share is effectively a measure of how well titles map to real query demand. A rising search share is also a leading indicator for Google web surfaces, because the same question phrasing that wins YouTube search is the phrasing that appears in AI Overview fan-out queries.
External: 39.9% of watch time
External watch time is plays outside youtube.com, mostly embedded players. With an embed on 1,356 of 1,648 site pages, the channel’s own website is the most likely majority source, though the Studio summary does not break external traffic down by referrer. Embedded plays tend to run long because the viewer arrived on a page about the same topic and pressed play deliberately. The share gap is the signature: external sources contributing a much larger share of watch time than of views means external sessions are longer than average.
Two implementation details decide whether embeds contribute at all. First, the embed has to be on the page whose topic matches the video, not a generic channel trailer. Second, the page needs to render the player without blocking the main content, typically with a lightweight facade that loads the iframe on interaction. A player that never loads produces no watch time.
Shorts feed
The recent review-removal Short drew quick views from search and the Shorts feed. Shorts inflate view volume and depress WT/V. In our citation tracking they are also nearly invisible to AI answers: in the October 23 to November 8 cohort, Shorts were cited once in 48, against 39 of 73 long-form videos cited by November 15. A Short’s job in this system is routing, not ranking.
Why this matters for AI search visibility
Watch time is a YouTube metric, but the content that earns it is the content AI answers cite. The external evidence on YouTube’s role in generative search is consistent across vendors:
- Ahrefs Brand Radar data shows YouTube as the most cited domain in AI Overviews, up 34% over six months (Ahrefs, 2026).
- Among AI Overview citations that did not rank in Google’s top 100 for the same keyword, 18.2% were YouTube URLs (Ahrefs, 2026). That is a citation path that does not depend on classic organic rank.
- BrightEdge measured a 29.5% YouTube citation share in AI Overviews and 16.6% in AI Mode, against 0.2% in ChatGPT (MediaPost, 2026).
- Silverback Strategies found that 65% of the YouTube citations in its sample occurred in AI Overviews, versus 7% in Gemini and 1% in ChatGPT (Silverback Strategies).
Our own snapshots match the engine split: Google AI Overviews and AI Mode account for almost all of the channel’s video citations, Perplexity for a handful, and ChatGPT and Claude for none so far. The practical reading is that a YouTube strategy is a Google AI strategy first. The same properties that drove watch time here (one specific question, a concrete answer early, long-form depth) are the properties that make a video extractable as a cited source.
Mechanically, AI Overviews split a query into related sub-queries and cite pages that recur across those sub-query results (Search Engine Journal, 2026). A channel with many narrowly scoped, question-titled videos has more chances to match a sub-query than a channel with a few broad ones. That is the structural argument for publishing one video per question rather than one video per topic.
Comment-mined topics
The Studio summary surfaced two comment threads with clear topic value. One flagged Google Ads’ default “Presence or interest” location setting as a quiet budget leak for local advertisers. That is accurate to the product: by default, location targeting includes both physical presence and locations of interest (Google Ads Help, 2026), so a service-area business can pay for clicks from people who are far away but typed its city name. The other, from a commenter who works on brand visibility in ChatGPT and Gemini, noted that strong content on a third-party listing does nothing for the owner’s domain, and that a page with concrete distances and explicit rules is the kind of page assistants quote.
Both are entity-level observations. The first is about matching spend to a real service area. The second is about where the citable facts live. Each maps to a 5 to 10 minute video and a matching page on the owner’s domain, which is the pair that maximizes both watch time and citation surface.
Pipeline implications
- Measure depth, not volume. Track WT/V and average view duration per format and per source. A flat view count with rising WT/V is a win, not a plateau.
- Bridge Shorts to long-form explicitly. Every Short gets a pinned comment and description link to the long-form video answering the full question. Measure the bridge with the related-video and end-screen sources on the destination video.
- Pair every long-form video with a page. One question, one video, one page on the owner’s domain with the embed, a transcript excerpt and VideoObject markup. The page earns external watch time; the video earns AI Overview citations; both reinforce the same entity.
- Prioritize cost and diagnosis questions. In this vertical, pricing and “why is this happening” videos carried the window. They match pre-purchase intent on YouTube, on Google and in AI answers.
- Re-angle, do not wait. A long-form video uncited after about four weeks of weekly snapshots should be retitled or re-scoped to a narrower question, consistent with the time-to-citation findings.
- Annotate counting regime changes. Mark August 24 and November 16, 2026 on every view chart and store engaged views alongside public views where the API exposes them.
Limitations
- One channel, one vertical (marketing for contractors and home-service businesses), one 28-day comparison.
- The Studio summary does not break external traffic down by referrer, so attributing external watch time to on-site embeds is an inference from embed coverage, not a measured split.
- Format mix (Shorts versus long-form share of views) for each window was not available in the summary, so the decomposition into mechanisms is directional.
- Watch time per view is a blended ratio. It cannot distinguish longer sessions from fewer short ones without per-video data.
- The co-view change falls at the edge of the window. Its effect on the reported views is assumed small, as Studio itself suggests, but is unmeasured.
Practical implications
- After August 24, 2026, public YouTube views are a weak trend metric. Watch time and engaged views are the stable ones.
- A large gap between a source’s share of watch time and its share of views identifies where your most engaged sessions come from. Here it was external embeds.
- Question-titled long-form videos that answer cost and diagnosis queries win YouTube search, hold attention and are the format Google’s AI surfaces cite. Shorts are a routing layer.
- Comments are a cheap topic source. When two viewers raise the same specific mechanism, it is a video and a page.
A plain-English version of this breakdown for local business owners is on PushLeads.