The 16 sources YouTube separates, and the 6 that carry your month
Inside the Reach tab of YouTube Studio there is a report called traffic source types. It answers a question no other number on the dashboard answers: which door the viewer came through. Views do not tell you that. Impressions do not tell you that. Watch time tells you even less. Only this report tells you, and it is the report almost nobody opens.
The official list is longer than the market assumes. Inside YouTube, Studio separates browse features, suggested videos, YouTube search, channel pages, notifications, playlists, end screens, Shorts, sound pages, campaign cards, video cards, remixed video, YouTube advertising and product pages. Outside YouTube it counts external sources and direct or unknown sources. Sixteen boxes, and on a healthy channel six of them carry almost everything.
- Browse features: traffic from the Home feed, subscriptions, Watch Later, Trending and other browsing surfaces. On a mature channel this is the biggest door by a distance.
- Suggested videos: traffic from suggestions that appear next to or after another video, and also from links in video descriptions. Those videos can be yours or somebody else's.
- YouTube search: traffic from search results inside the platform.
- Channel pages: traffic from your own channel page or from another channel's page.
- Notifications: the subscriber bell, personalized notifications, email and digests.
- External sources: websites and apps that embedded or linked your video. Direct or unknown: typed URLs, bookmarks, signed out viewers and apps YouTube cannot identify.
The door you ask for and the door you are handed
Search is the only source where a human typed something. You can aim at it. Title, description, the topic itself, all of it is under your hand, which makes search the most controllable source you have and, at the same time, the one with the lowest ceiling. Its ceiling is the search volume of the phrase, and search volume does not grow because you published more.
Browse features and suggested videos work the other way around. Nobody requests the home feed. YouTube offers your thumbnail to somebody who opened the app for an entirely different reason, and it does that because it believes that person will stay. You do not ask for that door. You become eligible for it, and eligibility is bought with history: session watch time, topic consistency and a publishing rhythm the system can predict.
Put the 2027 bar next to it and the difference stops being philosophical. Anyone joining the Partner Program from 1 February 2027 needs 1,000 subscribers plus 8,000 qualified public watch hours in 365 days, double the old 4,000. Eight thousand hours is 480,000 minutes. A 12 minute video watched at 40 percent delivers 4.8 minutes per view, so you need roughly 100,000 views inside a year. Very few niche keywords carry 100,000 searches in twelve months, and none of them carry it to a new channel. The hours come from the doors you cannot ask for.
That is why the honest reading of this report is not which source is best in the abstract. It is which source is currently in a position to grow, and what it needs from you. If you want the metric layer underneath this, the analytics metrics that matter is the companion read.
What each mix is telling you about your channel
Percentages matter more than absolute numbers here, because the shape of the mix is the diagnosis. Four shapes cover almost every channel, and each one asks for a different next move.
- Search above half of your views: you built a library, not a channel. Views arrive slowly, they arrive forever, and they stop at the search volume of the topic. Great for a tutorial catalogue, structurally slow for watch hours. The fix is not more SEO, it is one series that gives the algorithm a reason to suggest video two after video one.
- Suggested videos in front: the algorithm found your neighbourhood. This is the best position to scale from, and the correct next video is a cousin of the one pulling the traffic, not a new experiment in another topic.
- Browse features in front with low subscriber conversion: you are being offered to cold viewers and they are not staying. The thumbnail is doing its job and the first minutes are not. That is a retention problem, not a discovery problem, and it is exactly where hooks in the first 30 seconds pay for themselves.
- External or direct in front: you are the distribution, not YouTube. Newsletters, forums and your own posts are carrying the video. It counts for watch hours, it does not compound on its own, and it disappears the week you stop posting links.
Why a new channel is almost all search and subscriptions
A new channel has no history, so the system has nothing to predict with. It cannot know who stays because of you, so it will not risk your thumbnail on a stranger's home feed. What is left is the audience that already asked: people typing your topic, and the handful of subscribers who get a notification.
This is also why the first weeks feel like shouting into a well. Impressions stay in the hundreds, and a click rate measured on a sample that size means nothing at all. If your impressions look absurdly low right now, the counting rule explains most of it, and what YouTube actually counts as an impression is worth reading before you conclude anything about your thumbnails.
What opens browse features is boring and expensive: a run of videos on the same topic, published on a rhythm, each one holding people long enough that YouTube gets paid for the offer. Doing that by hand costs between 9.5 and 13.5 hours per finished long video once you count research, script, voice, editing, thumbnail and upload. Twelve weeks of that at four videos a week is 456 to 648 hours of work before the graph even starts to turn. That block is exactly what FalconVid takes off your hands: you approve the calendar once and each video assembles itself before its slot, with the AI specialists running in parallel, so the run exists even in the weeks when you have no free weekend.
Suggested videos: the neighbour nobody talks about
The official definition of suggested videos carries a detail most creators miss: it is traffic from suggestions next to or after another video, and also from links in video descriptions. So the source blends the algorithm's opinion with a link you control, and both land in the same bucket in your report.
You appear next to another channel when you share an audience inside the same session, not because your topics look similar to a human. Somebody watched a 15 minute breakdown of a court case and stayed; your 12 minute breakdown of another case is the natural next offer. The practical levers are unglamorous: same promise, same format, same length band, a series the viewer can continue, and playlists that make continuing the default path. In FalconVid that becomes configuration rather than willpower: the channel DNA locks promise, format and identity, and the calendar is born with the next video on the same topic already scheduled.
End screens and playlists show up as their own separate sources in the same report, which is why they usually look tiny. They are not a growth engine, they are a conversion detail at the end of a video the algorithm already delivered.
What changes when the publishing rhythm stops depending on you
Every lever above is a volume lever wearing a strategy costume. Topic consistency needs many videos on one topic. Session watch time needs a next video to exist. Browse eligibility needs a rhythm the system can trust. The reason most channels never get there is not ignorance, it is that a person cannot hand build four long videos a week for three months.
FalconVid is the machine for exactly that gap. You approve a calendar once and AI specialists work in parallel on each video: a researcher gathering the facts, a scriptwriter, a narrator with premium voices, an editor, sound design and captions, with a finished video in up to 30 minutes and simultaneous generations depending on the plan, from 2 on Starter to 50 on Scale. Publishing goes out to YouTube, Instagram, TikTok, Rumble and Facebook in up to 63 languages, and you can duplicate a project into another language paying only the difference.
A 12 minute video costs 1,008 credits in economy mode, 8,676 in balanced and 26,760 in premium, with engines like Veo 3 with audio and Seedance Pro available on every plan. Nothing in creation is locked behind a tier: what changes with the plan is volume, channels, simultaneous generations, the dedicated server from Pro, the Senior AI Analyst from Pro and support. The Senior AI Analyst matters here in particular, because it is the layer that reads this traffic source report for you every 2 days and tells you which door is open this week. It has its own page at the AI YouTube analyst page, and the full product lives on the FalconVid home page.
The honest arithmetic: about 48 videos before the graph turns
Take the most common plan people describe: four long videos a week for twelve weeks, 48 videos, enough topic mass for the system to have an opinion about you. By hand that is 456 to 648 hours, a full time job with overtime, and it is the reason so many channels die around week six.
Automated, the same 48 videos in economy mode cost 48,384 credits, roughly US$ 152 at Starter credit value. The Pro plan at US$ 97 with 30,000 credits a month, 5 channels and 5 simultaneous generations covers that pace across two months of production, and Business at US$ 297 with 95,000 credits covers it inside a single month with room left for a couple of videos in balanced mode. In FalconVid those 48 videos come out of the same calendar you approved once, with up to 5 simultaneous generations on Pro, and each one finished in up to 30 minutes instead of waiting for your schedule to open.
Both tracks are real, and the difference is not quality, it is whether the calendar survives your bad weeks. The manual track caps at what a person can sustain. The automated track caps at your plan, which on Scale is 317 economy videos a month. The graph does not turn because you tried harder on one video. It turns because forty of them existed.
The 20 minute diagnosis, in the order that saves time
Open Reach, set the window to the last 28 days, and write down the share of each source. Then set it to 90 days and write it down again. The comparison, not the snapshot, is the signal: a channel that went from 20 percent suggested to 40 percent suggested is being placed better, even if total views fell.
Then leave the channel level and go per video. Sort your last ten videos by views, take the top three and the bottom three, and look at where their traffic came from. In most stalled channels the top three are suggested heavy and the bottom three are search heavy, which tells you which of your topics has a neighbourhood and which one is an island.
Finally, act on one door only. If suggested is working, the next video is a cousin of the winner. If browse is delivering but retention is not holding, fix the first minute before you touch anything else. If search is all you have, build the second video that continues the first one. The full sweep of this, with the other 30 checkpoints, is laid out in the channel audit checklist. And if you would rather not run this sweep every week, it is literally the job of the FalconVid Senior AI Analyst, which reads the account every 2 days and writes, in your language, the move to make inside the product.