YouTube has signalled more AI and user-built feeds in the months ahead, but the controls that already shape your recommendations exist today. This guide explains how the feed is assembled now and what you can change yourself.
Key takeaways
- Ars Technica reports that YouTube has promised custom feeds and a much larger role for AI later this year, alongside a focus on livestreams.
- The specific feature names, launch dates and countries involved are not established in that report and should not be assumed.
- The YouTube home feed is not chronological: it is assembled from watch history, search history, subscriptions and engagement signals.
- Viewers already have direct controls — history pausing and deletion, “not interested” and “don’t recommend channel” — that change what is shown.
- The subscriptions tab remains the closest thing to a feed you control by hand, and is unaffected by recommendation ranking.
What has YouTube said it will change?
Ars Technica reports that YouTube has promised custom feeds and a substantially larger role for artificial intelligence on the platform later this year, framing the coming period as one that could alter the content viewers are shown. The same report identifies livestreaming as a second area of emphasis.
Beyond that, public detail is thin. The exact names of the features, the account types and countries that will receive them first, and the dates on which they arrive are not known here and should not be inferred. “Custom feeds” is a broad phrase. It could mean user-defined topic feeds sitting alongside the existing home and subscriptions tabs, or tighter controls over the recommendation surface that already exists, or some combination.
What can reasonably be said is the direction of travel: more machine-generated organisation of what you watch, more machine-assisted production tools for the people who make videos, and more weight given to live content. Whether any of it can be switched off, whether parts of it sit behind a paid tier, and how it interacts with existing settings are all unconfirmed. This guide is written on that basis: it covers the mechanics that are already in place and the controls you can use now.
Why this is in the news now
Video platforms tend to announce product direction in batches, and the current round arrives during a period in which nearly every large consumer service is adding generative features. The interesting part is not that AI is mentioned, but where it is being pointed: at the feed itself, which is the part of the product most viewers experience as “YouTube”.
Two structural shifts sit behind this. The first is that a large share of YouTube viewing has moved to television screens in living rooms, where browsing is slower and the ranked feed does more of the work of choosing. A recommendation system that picks well matters more on a remote control than on a phone. The second is live content, which Ars Technica flags as a focus. Live video is harder to catalogue and rank than uploaded video, because there is no finished file to analyse before it is shown, and because its value decays quickly.
Custom feeds address both. A feed a viewer defines is cheaper to get right than one the system must infer, and it gives the platform a place to put content types — live, long-form, short-form — that compete badly against one another in a single ranked list.
How the feed you see today is actually built
A newcomer should understand one thing before changing any settings: the YouTube home page is a ranked recommendation surface, not a list of new videos from channels you follow.
The system draws on several classes of signal. Watch history is the strongest: what you played, how long you stayed, and what you watched next. Search history contributes, as do subscriptions, likes, and explicit negative feedback. Contextual signals such as device, rough location, language and time of day also shape the mix. Videos that hold attention from similar viewers get surfaced more widely, which is why a single long viewing session on an unfamiliar topic can reshape a feed for days.
Separate surfaces behave differently. The subscriptions tab lists uploads from channels you have subscribed to, in reverse-chronological order, and is not reordered by predicted interest in the way the home feed is. Shorts is its own vertical feed with its own signals. Notifications are governed per channel, independently of both.
This split matters for the coming changes. Any “custom feed” is likely to be a new surface layered onto this structure rather than a replacement for it, which means the existing controls keep working even if the home page changes shape.
Who is affected, and how
Casual viewers are affected least in the short term, because the default experience is designed to require no configuration. The change they are most likely to notice is a shift in the mix of content — more live, and more material assembled or surfaced by automated systems.
Heavy viewers with narrow interests are affected more. People who use YouTube as a reference library, for repair guides, music or long-form talks, tend to have carefully trained feeds and are the most disrupted when ranking changes.
Creators face the larger adjustment. Discovery is their distribution, and any reweighting of the feed alters which formats are rewarded. An emphasis on live content favours those who can broadcast on a schedule, which is a different production model from edited uploads. Generative tools may lower production costs, but they also increase the volume of competing material.
Households with children are a distinct case: supervised accounts and the separate children’s app have their own recommendation rules, and parents should not assume settings applied on a main account carry across.
Advertisers and moderation teams are affected indirectly, through changes in what content is eligible for promotion and how synthetic material is labelled.
Where informed people disagree
There is genuine disagreement about whether more user control improves feeds. One view holds that letting people define their own feeds reduces the filter-bubble problem, because the choices are explicit rather than inferred. The opposing view is that self-selected feeds narrow interests faster than algorithmic ones, since recommendation systems at least introduce adjacent material.
A second dispute concerns generative content. Some argue that cheaper production tools broaden who can publish. Others argue that volume is already the binding constraint on attention, and that lowering production cost mainly dilutes the pool. Neither position can be settled with the information currently available.
A third disagreement is about disclosure. YouTube asks creators to disclose realistic synthetic or altered material, and shows a label where it applies. Critics say self-declared labelling is weakly enforced; defenders say automated detection is unreliable enough that self-declaration is the only workable starting point.
What you can do today, step by step
These controls exist now and are independent of any forthcoming changes.
Start with history. In YouTube’s settings, under history and privacy, you can pause watch history, pause search history, clear either, and set activity to delete automatically after a fixed period. Pausing watch history makes the home feed fall back towards generic and popular content, which is a blunt but effective reset. You can also delete individual entries from the history page, which is the precise option when one viewing session has distorted your recommendations.
Next, train the feed directly. On any home-feed thumbnail, the three-dot menu offers “not interested” and “don’t recommend channel”. The second is the stronger signal and works reliably for suppressing a source you never want to see.
Then rebuild around subscriptions. Open the subscriptions tab and use it as your default entry point rather than the home page. Review your subscription list and remove channels you no longer watch, since dormant subscriptions still contribute signal. Set per-channel notifications to “all” only for the handful you genuinely want pushed to you.
Separate your contexts. If you research topics you do not want reflected in recommendations, use a second Google account, a signed-out browser session, or the mobile app’s incognito mode, none of which write to your main history.
Finally, save deliberately. Playlists and Watch Later give you a catalogue you control, which is useful insurance against any reshuffling of the recommendation surface.
What to watch next
Three things are worth monitoring. First, whether custom feeds are additive or replace an existing surface — an additive feed changes little for people who already use subscriptions, while a replacement changes a great deal. Second, whether the new AI features carry an off switch, and whether existing privacy and history settings continue to govern them. Third, how live content is ranked against uploaded video, since that determines whether the emphasis on livestreams is a new tab or a change to the main feed.
Official help documentation is generally updated as features roll out, and is the most reliable place to confirm behaviour once anything ships. Until then, the details above remain unconfirmed.
Frequently asked questions
Can I turn off YouTube’s AI features?
That is not known. No public information establishes whether the AI features described by Ars Technica will include an opt-out, or whether they will be tied to existing account settings. Some current automated features, such as recommendations, can be blunted by pausing watch history, but that is a side effect rather than a dedicated switch. Check official help documentation once anything ships.
How do I stop YouTube recommending a channel?
Open the three-dot menu on any video thumbnail in the home feed and choose “don’t recommend channel”. This suppresses that source from your recommendations. The neighbouring “not interested” option applies to a single video and is a weaker signal. Neither removes the channel from search results, and neither affects channels you are subscribed to, which continue to appear in the subscriptions tab.
What is the difference between the home feed and the subscriptions feed?
The home feed is a ranked list assembled from your viewing signals, and can include channels you have never watched. The subscriptions feed lists recent uploads only from channels you have subscribed to, in reverse-chronological order, without interest-based reordering. If you want a predictable, self-selected experience, the subscriptions tab is the surface to use as your default entry point.
Does pausing watch history make YouTube worse?
It makes recommendations generic rather than absent. With history paused, the home feed falls back towards broadly popular and locally trending content instead of material tailored to you. Search still works normally and subscriptions are unaffected. Many people find this an acceptable trade; others prefer deleting specific history entries, which keeps personalisation intact while removing the sessions that distorted it.
How can I tell whether a video was made with AI?
YouTube asks creators to disclose realistic synthetic or altered content and displays a label where that disclosure has been made. The label depends on self-declaration, so it is not a guarantee. There is no reliable viewer-side detection method for generated video, and confidence should come from the source’s track record rather than from any single visual cue.
Will custom feeds replace the subscriptions tab?
Nothing in the available reporting says so. The subscriptions tab is a long-standing surface with distinct behaviour, and there is no public indication that it is being removed or merged. Until YouTube publishes specifics, the reasonable assumption is that any custom feed is an additional surface. If that changes, official help documentation is where it would be confirmed first.
Sources and further reading
- Ars Technica — the technology publication that reported YouTube’s promise of custom feeds, expanded AI features and a focus on livestreams later this year.
- YouTube Help — the platform’s official documentation covering recommendations, feedback controls, subscription management and synthetic content disclosure.
- Google Account activity controls — the settings documentation describing watch and search history, pausing, deletion and auto-delete intervals.
- Academic and industry research on recommender systems — general literature on how ranked feeds use engagement signals, useful for background rather than platform specifics.
Surfaced from the rss:arstechnica signal “video platform AI features”. AI-assisted draft, editorially reviewed.
