A personalized feed has a basic problem on day one:
It does not know you yet.
Recommendation systems usually learn from history—what a person watches, clicks, skips, searches, follows, buys, likes, or rejects. A brand-new account has little or none of that. The service still has to decide what the first screen should contain.
That is the cold-start problem.
Possible solutions include popular material, local trends, editor-selected content, demographic or device context, explicit onboarding choices, recommendations based on similar users, or simple search and browse tools until enough history exists.
YouTube currently takes an interesting approach. Its documentation on how recommendations work says the homepage primarily relies on watch history. If watch history is off and there is no significant prior history, the homepage can omit personalized recommendations and instead leave the user with search, navigation, subscriptions, and Explore.
That is important because it proves there is no universal law saying a platform must invent a personalized feed before it knows anything about the person.
The starting view still shapes discovery
Even without personalized recommendations, a new user encounters defaults.
Which categories are easy to reach? Which topics appear in Explore? Which creators dominate search for broad queries? Which onboarding choices are offered? Which regions, languages, and devices affect what is visible?
If another service fills its cold start with globally popular items, the new user begins inside majority taste. If it asks users to pick five interests, those choices become the seed. If it imports contacts or follows, the starting graph comes from social connections.
None of these choices is neutral.
They are practical solutions to missing information.
Early behavior can harden quickly
Cold start is temporary because every action supplies data.
One search becomes a search-history signal. One watch becomes a watch-history signal. A follow or subscription provides stronger evidence. Repeated choices gradually replace the generic starting model with a personalized one.
That means the first few sessions can matter disproportionately. Early recommendations influence what is available to click, and those clicks become evidence for later recommendations.
A fair analysis therefore needs to separate what the platform showed before it knew the user from what the platform learned after the user began interacting.
The Dead Internet Theory question here is not whether the initial feed is fake.
It is whether a person mistakes an engineered starting view for a natural sample of what the internet contains.
A new account begins with almost no personal history.
The screen still has to begin somewhere.