YouTube Traffic Sources Explained: What Browse, Suggested and Search Are Actually Telling You

The traffic sources report is the most useful screen in YouTube Studio and the most consistently misread. Creators open it, see that Browse features is 62% of views, and conclude either that the algorithm loves them or that they are dangerously dependent on it — and both conclusions are guesses, because the percentage says nothing on its own. The same 62% is a triumph on a channel whose views tripled and a warning on a channel whose search traffic quietly died.

What the report is genuinely good at is narrowing a vague problem to a specific one. Views are the product of three separate things: how often YouTube showed your video, how often somebody clicked it, and how long they stayed. Every traffic source measures that chain in a different context, on a different audience, with different odds. Split the chain by source and a question like "why is this video underperforming" usually resolves into one of about four answers within ten minutes.

This piece works through the whole report: where it lives now that Studio is being redesigned, what each bucket counts, which of them have impressions behind them and which do not, how to read a mix rather than a number, and four conclusions the report cannot support.

Where the report lives, and what it is called this month

In Studio, the path is Analytics → Advanced mode, or Analytics → Reach on a single video, where the panel is titled "How viewers find your video." YouTube's documentation for the Reach report covers the same ground: traffic source types, impressions, impressions click-through rate, views and unique viewers.

Two caveats about what you will see. Since July 2026 YouTube has been staging a redesigned Studio in which the Analytics tab is renamed Insights, charts open into Advanced mode on click, the Trends tab becomes Research, and some accounts get AI summary cards and an "Ask Studio" chat. Two creators comparing screenshots this month may genuinely be looking at different interfaces. The report itself has not changed underneath the rename.

The second caveat is that the friendly labels are a simplification of a longer list. The YouTube Analytics API dimensions reference exposes the raw traffic source types, and reading them is the fastest way to understand what the Studio labels are hiding — most usefully that the bucket Studio calls Browse features is called SUBSCRIBER in the API, and is defined as views referred from feeds on the YouTube homepage or from subscription features. One label, two very different surfaces.

The buckets, and what each one is actually reporting

Here is the practical version: what the source counts, and the question it answers about your channel. The last column is the one worth internalising, because it is the reason to look at the report at all.

Traffic source	What it counts	The question it answers
Browse features	Views from the home feed, the subscriptions feed and similar browsing surfaces	Does your packaging beat everything else competing for an idle viewer?
Suggested videos	Views from the sidebar, the up-next slot and some description links, alongside other videos	Is your video the obvious next thing after something people already chose?
YouTube search	Views from queries typed into YouTube	Does your video answer a question people are actively asking?
Shorts feed	Views from the vertical feed	Does the first second hold, and does the loop survive a thumb?
Channel pages	Views from someone browsing your channel	Does your back catalogue convert a visitor into a second watch?
Playlists	Views from playlists containing your video, yours and other people's	Have you built sequences, or a pile of unconnected uploads?
Notifications	Views from bell and inbox notifications sent to subscribers	How much of your launch is your existing audience?
End screens	Views from end screen elements on other videos	Are you routing your own traffic on purpose?
External	Views from sites and apps off YouTube, including Google search and embeds	Does anything outside YouTube send people to you?
Direct or unknown	Views with no referrer YouTube could attribute — pasted links, bookmarks, some apps	Mostly a measurement residue; rarely a growth signal
Sound page and remixes	Shorts views from a sound's pivot page or from a remix link	Is your audio or your clip being reused by other creators?

Browse features is two surfaces wearing one label

This is the single most consequential thing to understand about the report, because Browse is usually the biggest row and the label conceals a split that changes what you should do next.

Some of that traffic is the home feed: a viewer opened YouTube with no particular plan, and the recommendation system decided your thumbnail was one of the dozen or so things worth putting in front of them. That is cold traffic. The viewer may never have heard of you. Your thumbnail is competing against every other video on the screen, most of which have more subscribers, better production and a face the viewer already recognises.

The rest is the subscriptions feed: people who already chose you, scrolling a list of channels they follow. That traffic has almost nothing to do with algorithmic favour and almost everything to do with whether you have published recently and whether your subscribers still open the feed.

Those two behave in opposite ways, and averaging them produces a number that means nothing. Click into Browse features in Studio to see the sub-source breakdown, and do it before drawing any conclusion about Browse at all. A Browse row that grew because the home feed picked a video up is a distribution win. One that grew because you uploaded three times in a week is a scheduling artefact.

The algorithm starts when a viewer arrives, not when you upload

The mental model most creators carry is that publishing triggers a distribution decision: YouTube looks at the new upload, decides how good it is, and pushes it to some number of people. People on YouTube's growth and discovery team have described it the other way round. The system is triggered by a viewer opening the app, at which point it assembles a feed for that person from everything available — so recommendation is a pull towards a specific viewer, not a push out of a warehouse. Todd Beaupré, who leads that team, has also described collaborative filtering, the machinery behind suggested videos, as automating word of mouth: the system observes that people who watched this also watched that, and behaves like a friend making a recommendation.

Two things follow that are worth more than any tactic. First, "the algorithm didn't push my video" is a statement about demand, not punishment — nobody arrived for whom your video was the best available answer. Second, the number of people your video could reach is not fixed at upload. A video that gets nothing for six weeks and then finds a feed has not been un-suppressed; a different set of viewers turned up.

The size of the pool involved is easy to underestimate. When YouTube last gave a public figure of this kind, its then chief product officer Neal Mohan told CES in 2018 that more than 70% of watch time on the platform came from recommendations rather than search or direct navigation. That figure is old and YouTube has not refreshed it, so treat it as an order of magnitude rather than a current statistic. The direction of travel since has been consistently towards more recommendation, not less: YouTube retired the Trending page in July 2025, explicitly citing years of declining visits and the shift of discovery towards personalised feeds, Shorts and search.

Suggested videos: adjacency, not quality

Suggested is the row creators most want and least understand. It counts views from the sidebar, the up-next slot and end-of-video recommendations — the API name, RELATED_VIDEO, gives away the logic. You are not being recommended because your video is good in the abstract. You are being recommended because somebody is watching something, and your video is the plausible next step for them specifically.

That has a practical consequence that nobody likes: suggested traffic is easier to earn by being adjacent to demand that already exists than by being better. A well-made video on a subject nobody is currently watching has nothing to sit next to. A competent video on a subject with heavy current viewing has a queue of watch pages to appear beside.

If suggested is thin, the questions are about position, not craft. Is there a body of videos yours should be the natural follow-on from? Does the packaging make that connection visible at sidebar size, where the thumbnail is small and stripped of context? Is there a second and third video on the subject, so the system has somewhere to send the people it just sent you? Suggested rewards clusters; single videos on unrelated subjects produce a channel where every upload starts at zero.

Search is the only source that keeps paying

Search is usually a small share of views and almost always the most durable. A video that ranks for a real query earns views on a schedule set by demand for that query, which for most practical subjects is roughly flat for years. Browse and suggested traffic, by contrast, decay with the feed's attention span.

This is why the share number misleads. A channel with 8% of its views from search may be in better shape than one with 40% from browse, if that 8% arrives every month without a new upload. Judge search on views per month per video, and on whether old videos still get any.

One thing has genuinely changed here, and it is worth watching rather than reacting to. Since June 2025 YouTube has been testing an AI-generated results carousel on some searches — initially for Premium subscribers in the US, on shopping, travel and location-type queries — which surfaces clips and summaries above ordinary results, alongside a conversational assistant that has since been expanding to more accounts and, in 2026, to living-room devices. Nobody outside YouTube has reliable data on what this does to click distribution yet. The reasonable response is to keep making videos that answer a specific question completely, and to watch whether your search traffic behaves differently on the query types where the carousel appears. Our YouTube SEO guide covers the ranking side in detail.

The Shorts feed is a separate economy

Shorts views arrive overwhelmingly from the Shorts feed, and that source works nothing like the others. There is no grid of competing thumbnails, so there is no meaningful thumbnail decision at the point of distribution — the video is simply next, and the only question is whether the viewer swipes. What functions as the click decision on long-form is, on Shorts, the first second of playback.

This is why mixing Shorts and long-form in the same channel-level traffic report produces nonsense. Filter by content type before reading anything. A channel-wide CTR that includes a month of Shorts is not a number about your thumbnails. If you publish both, run the two reports separately, always. The thumbnail still matters for Shorts, but in different places — the channel page, search results and the Shorts shelf — which our Shorts thumbnail guide works through.

Impressions do not exist on every surface

Here is the mechanical detail that quietly invalidates a lot of confident analysis. Impressions count how many times your thumbnail was shown on YouTube — but not everywhere it was shown. YouTube's impressions and click-through rate FAQ states that impressions exclude views from external sites, from end screens and from notifications. Thumbnails shown in emails, notifications and similar messages are not counted.

Three consequences follow directly:

• A video can have more views than impressions. If a link goes around a forum or a newsletter, the views arrive with no impressions attached. This is normal, not a bug or a bot attack.
• Impressions CTR describes only the part of your traffic that came through counted surfaces. It is a partial metric by construction, and YouTube says so.
• Sources with no impressions behind them cannot be improved with better packaging. If a third of your views are external and direct, a thumbnail redesign has no lever on that third at all.

A CTR only means something inside one traffic source

The only first-party benchmark that exists is the one in that same FAQ: half of all channels and videos on YouTube have an impressions click-through rate somewhere between 2% and 10%, with a wider spread for new channels and videos with very few views. That is a range containing most of YouTube, which is another way of saying it is not a target.

What the report adds is context. The same thumbnail earns different rates on different surfaces, because the viewer's state differs: somebody who typed a query is looking for a specific thing, somebody scrolling the home feed is looking for anything, and somebody at the end of a video is looking for more of what they just had. Comparing your Browse CTR to a benchmark built mostly from search traffic is comparing two different experiments.

There is a second effect that catches people out on their best videos. YouTube's help documentation on decoding CTR and impressions makes the point explicitly: as a video reaches further and accumulates more impressions, its CTR often falls, because the video is being shown beyond its core audience to broader, less obviously interested viewers. A falling CTR on a rising video is frequently the signature of success, not of decay. The corollary is uncomfortable — the safest way to protect a high CTR is to be shown to fewer people. Our post on what counts as a good CTR goes through the arithmetic of that trade.

You will find tables online giving precise CTR benchmarks per traffic source — search 8–15%, browse 2–5%, and so on. None of them traces back to a first-party source or a disclosed dataset, and the figures differ between sites in ways that suggest estimates being copied around. Use your own history instead: your median Browse CTR across your last twenty videos is a real benchmark for you and takes ten minutes to build.

Reading a mix instead of a number

Patterns in the mix carry more information than any single row. These are the ones that come up most, and what they usually mean.

What the report shows	What it usually means	Where to work
High impressions, low CTR, from Browse	YouTube is testing you with cold viewers and the packaging is not winning the row	Thumbnail and title, tested against each other, not redesigned on instinct
High CTR, low watch time, from Browse or Suggested	The packaging promises something the first minute does not deliver	The opening, and the honesty of the promise
Almost everything from Notifications and subscriptions	The video never left your existing audience	Subject choice — the idea has no pull beyond people who already like you
Suggested near zero on every video	Nothing on the platform your videos naturally follow, or clusters too thin	Series and adjacency: more depth on fewer subjects
Search flat while browse swings wildly	Healthy. This is what a mixed channel looks like month to month	Nothing. Do not fix this
Search decaying across the back catalogue	Rankings lost, or demand for those queries has moved	Refresh the strongest old videos rather than republishing them
A large External row you cannot explain	Usually an embed, an aggregator or a link post; occasionally a single viral share	Find the referrer before celebrating; this traffic rarely subscribes

Three failure modes, three different levers

Underneath all of this there are only three ways a video underperforms, and the traffic source report tells you which one you have.

Nobody was shown it. Impressions are low across every source. This is not a packaging problem — a thumbnail cannot fix a video that is never displayed. It is usually a subject problem: the video has no query behind it, nothing to be suggested beside, and no hook the home feed can test. The fix is in what you choose to make next.

They were shown it and did not click. Impressions are healthy, CTR is below your own median for that source. This is the one case where packaging is genuinely the answer, and the only reliable way to work on it is comparison — running two or three real alternatives against each other rather than trusting a hunch about what looks better. Our thumbnail A/B testing guide covers how to do that without fooling yourself with noise.

They clicked and left. CTR is fine, average view duration is not. Nothing in the thumbnail will help; the video has to keep the promise the packaging made. Worth being precise about which of the two is wrong — an overclaiming thumbnail and a slow opening produce the same chart and need opposite fixes.

How to actually move each source

Browse

Browse responds to broad appeal and to packaging that survives being small and surrounded. The home feed shows your thumbnail inside a grid of competitors, often on a television across a room, which is now the most common way YouTube gets watched in the US — in February 2025 YouTube said the living-room TV had overtaken mobile as the primary device there, with over a billion hours a day watched on TV screens. Design for that: one subject, high contrast, text that reads at a glance, no detail that only survives at full size. It is worth checking a thumbnail at feed scale before publishing rather than after, which is what our thumbnail preview tool is for.

Suggested

Suggested responds to clustering. Make the second video on a subject that worked. Link the two with end screens and a playlist so viewers who finish one have somewhere obvious to go. Keep titles in a family so the connection is legible in a sidebar. None of this is a trick; it is giving the system a coherent thing to recommend.

Search

Search responds to specificity. A video titled for a query somebody actually types, answering it completely and early, with a title that matches the question rather than gestures at the topic. The gains are slow and they persist, which is the opposite trade to browse. Publishing time barely touches any of this, for reasons we went through in the best time to post.

A forty-minute traffic-source audit

1. Filter to long-form only. Run Shorts separately, or the numbers are a blend of two systems.
2. Set the window to the last 90 days, and take your last ten videos individually rather than the channel aggregate.
3. For each video, record absolute views from home feed, subscriptions, search and suggested — four numbers, not four percentages.
4. Record impressions and CTR per source for each. Note your own median CTR per source; that is your benchmark from now on.
5. Flag the videos with low impressions. Ask what they had in common as ideas, not as artwork.
6. Flag the videos where impressions were healthy and CTR was below your median for that source. That set is your packaging backlog.
7. Flag the videos with above-median CTR and below-median retention. That set is more urgent: it teaches the system to stop showing you.
8. Check the back catalogue: how many views did videos older than six months earn in the window, and from where? Close to zero outside subscriptions means no durable traffic, and every month starts from scratch.
9. Write down one change per flagged set. Three changes, not thirty.

Four conclusions the report does not support

"My browse share fell, so I have been suppressed." Shares move when any row moves. A search spike from an old video lowers browse share while browse itself is unchanged. Check absolute numbers before reaching for a theory, which is the first step in our piece on why views drop.

"Suggested traffic proves the algorithm likes my channel." Suggested is per video and per viewing context. It tells you a particular video sits well beside particular other videos. It does not confer standing on the channel, and it can vanish when the neighbouring videos stop being watched.

"External traffic is free growth." External views arrive without impressions, often from viewers with no interest in the channel, and they convert to subscribers poorly. Worth having, but not evidence of anything about your packaging: no thumbnail decision took place.

"Direct or unknown means bots." It usually means YouTube could not attribute a referrer — pasted links, some in-app browsers, some devices. It is a measurement residue. Reading intent into it is the analytics equivalent of reading tea leaves.

The report rewards a specific habit of mind: treat each source as a separate experiment with its own baseline, read absolute numbers before shares, and never compare a rate across two surfaces that put viewers in different states. Do that and the screen stops being a scoreboard and starts being a diagnosis, which is the only thing it could ever do.

Most of what it diagnoses, though, lands in the same place. Two of the three biggest sources — browse and suggested — are decided by whether a thumbnail and title win a row of competing options in a fraction of a second, against viewers who were not looking for you. That is a design problem before it is a strategy problem, and it is the part of the chain you can work on this afternoon. If the audit above hands you a packaging backlog, Thumblore exists to get through it quickly enough that you can test alternatives rather than agonise over one.

For the craft side of that, the 2026 thumbnail playbook covers what wins a cold feed impression, and what counts as a good CTR explains why the number you are chasing is smaller and stranger than the benchmark tables suggest.