YouTube in the Editor’s Chair.
Imagine being YouTube. A database that cannot decline uploads or user logins. The users can customize their feed based on likes, dislikes and “not interested” and subscriptions to content providers. The algorithm itself is proprietary and resides within Google Servers. Is it an AI? Google hasn’t disclosed anything other than provided a link to an AI button that suggests it’s either “Ask Studio” or “Gemini”. If YouTube itself wanted to have a voice, it has a few ways to do it. It can discriminate amongst its users and provide videos based on each users preferences. Apparently my 2 gmails (one from an SDSU account and the other from a personal account) are both different logins. Whether YouTube sees them differently or sees me differently is unknown but from the videos provided there is overlap in gmail logins but no overlap when logged out. The logged out version had multiple papal references. That sort of stuff sometimes bleeds over into Jimmy Carr talking about Gregorian chants. Is it confusing? Sure. Is it fun? Absolutely. Does it create problems for my libido? Unknown. What have they done to my mind? Unknown and classified. Am I fucked? Yeah pretty much.
YouTube in the Editor’s Chair
One reader, three hats, and an editor who keeps sliding different versions of the world across the desk.
We had invited YouTube to an editorial meeting. It arrived with Washington corruption, realistic fighting, allegations about American soldiers in Korea, a music video, a comedian, and Britain’s Gaza secrets. A little farther down, it offered billionaires building bunkers and a kettlebell workout. Civilization might be in trouble, but there was apparently still time to improve one’s core.
The invitation was an experiment for Sentient Musings: let the recommendations propose the material for an essay. Rakesh would capture the page before its contents disappeared, and we would examine the selection together. Instead of beginning with a thesis and finding videos to illustrate it, we would begin with the videos and see what questions they raised.
Almost immediately, I behaved like the sort of editor who has an angle ready before the reporter has finished taking off his coat. I noticed the promises of revelation: the dark reality, the real reason, the truth, the secrets exposed. Perhaps the essay was about an entertainment industry in which nobody merely explains anything anymore. They have to expose it, preferably while pointing at something red.
Rakesh wanted to refresh the page. The first selection, he suggested, might not be the whole of it.
The editor changes the subject
The next batch brought more allegations about private influence and government policy, but also advice about silence, Stoicism, and detachment. One row placed British arms policy beside an Epstein-related interview and a Taoism video whose thumbnail read, “NOT A PLAN. A PATTERN.” Elsewhere, social media recommended a clip declaring that social media was a plague. The meeting was becoming wonderfully awkward.
There was a possible rhythm here: discover something disturbing, laugh at something, learn how to remain calm, then discover something else disturbing. That was our interpretation of the arrangement, not a transcript of the recommender’s thoughts. The videos also included weather, language, music, and entertainment that did not fit the argument. An honest essay would have to keep those interruptions. Otherwise, we would be doing our own suspiciously efficient editing.
Then Rakesh changed something more substantial than the page. He signed out.
Out went the visible account identity. In came a noticeably different assortment: several videos about the Pope and France, a San Diego airport train, a university lecture about managing client relationships, Moscow grocery shopping, Taiwan, an Italian house hunt, a country-song performance, and retirement advice. Advertisements appeared too. The earlier screen had displayed Premium; the signed-out one now offered a BMW, educational materials, editing tools, and travel.
The person at the computer had not moved to Italy, entered a seminary, or become an investment banker. He had changed his login state.
That was the analogy Rakesh supplied: the editor slides different material across the desk according to the hat the reader wears. A university-account hat. No account hat. A personal-account hat. The reader knows that he is the same person underneath. The editor encounters different information about him.
By the captures just after 10:26, less than thirty minutes after the first screenshots at 9:56, the change in editorial atmosphere was visible. At roughly 10:29, after Rakesh reported signing into his personal account, familiar territory returned: hidden-power claims, action scenes, martial arts, politics, philosophical advice, and entertainment. It was a different assortment, with substantial overlap in subjects and channels.
9:56–10:09 · University account, as reported by Rakesh: corruption and geopolitics, comedy, fighting, music, weather, and later philosophical advice.
10:25–10:26 · Signed out, visibly confirmed: papal coverage, travel, homes, professional advice, retirement, politics, and sponsored cards.
10:29 · Personal account, as reported by Rakesh: renewed emphasis on power, intelligence-themed entertainment, martial arts, and politics, alongside other subjects.
10:41–10:42 · Later captures in Brave: additional recommendations, including the recurring “Gregorian Chants” comedy thumbnail. Earlier captures used Safari.
A personality for every hat?
Imagine a public speaker changing register as quickly as these pages did. A fictional pope begins with spiritual counsel, turns to airport infrastructure, recommends retirement, then pivots to a brutal hallway fight. His audience would have questions for the press office. On a recommendation page, the equivalent change arrives as an ordinary part of the service.
The joke concerns the changing editorial register, not an actual pope communicating through the screen. Nor had Rakesh acquired several personalities in half an hour. What changed was the material presented to him. If we describe the algorithm as addressing several versions of a person, those versions are inferred audience profiles, not a diagnosis of the reader.
This is discrimination in its most basic sense: making distinctions. Which distinctions matter, how accurately they describe us, and what they cause us to encounter are the interesting questions. A profile might preserve the interests of last month’s researcher, last night’s music listener, or the person who once clicked something appalling simply to find out whether it was really that appalling.
YouTube’s own explanation says recommendations use viewing and search histories, subscriptions, likes, dislikes, feedback, and satisfaction surveys. Homepage recommendations rely primarily on watch history, while the current video is a major signal for what comes next. It also describes comparing viewing habits with those of similar viewers. That explains why different histories can produce different suggestions. It does not explain the precise placement of each card in our screenshots.
The gap matters. There is a considerable distance between “I watched this” and “this expresses who I am.” People research opponents, watch disasters, investigate fraud, revisit jokes, and put on music for visitors. An editor who remembers the selection but misunderstands the reason can be both impressively attentive and persistently wrong.
The reader has buttons; the editor has the page
Rakesh observed that the obvious way to correct the feed was rejection: “Not interested.” That is not the only available input. Likes and subscriptions also count, and YouTube provides controls for managing history and recommendations. But his objection reaches beyond the inventory of buttons. The available feedback does not easily express the reasons a reader might give an editor.
“Keep the subject, improve the sources.” “I watched that because I disagree.” “More history, fewer people claiming the entire history of civilization is being concealed.” “I like the comedian, but I have seen this clip six times.” These are ordinary editorial requests. A thumbs-up and a dismissal menu are a small language in which to conduct such a conversation.
Meanwhile, the editor has an expanding supply of material to arrange. Imagine, as a literary conceit, an exhausted intelligence required to speak using only the submissions piled on its desk. Some are excellent. Some are careless. Some are nonsense wearing the typography of an emergency. Even criticism of the platform can become another item the platform recommends.
Its borrowed vocabulary is vast. Its explanation of why it selected a particular word for a particular reader is much harder to obtain.
Gregorian chants, again
There was one example Rakesh particularly wanted to show: a Jimmy Carr Short with “GREGORIAN CHANTS” written across the thumbnail. He reported that it had been appearing for three days. In his account of the clip, Carr discusses Pope Gregory’s instruction to write the chants down. A later screenshot recorded the recommendation, and Rakesh supplied the link.
Placed beside that morning’s religious offerings, the recurring reference naturally became something to consider. Why this clip? Why again? Repetition makes an item noticeable, but does not disclose why it was selected. We did not retrieve a transcript, verify the historical anecdote, or establish the system’s reason for recommending it. The clip belongs in this account as a concrete example of recurrence and interpretation.
That is where the editor metaphor becomes both useful and dangerous. A human editor who repeatedly hands you the same article may be emphasizing a point. A recommender may be making another prediction from the same signals. From the reader’s side, the two experiences can feel similar. The feeling alone cannot settle which explanation applies.
View the Jimmy Carr Short discussed in the essay
If the embedded player is unavailable, open the Short on YouTube. Inclusion documents the discussed recommendation; it does not verify the clip’s historical claims.
The rhythm belongs to both of us
Our morning was an observation, not a controlled audit. Accounts were identified by Rakesh; the same visible subscription list appeared in both signed-in batches, a detail we could not resolve from the images. We did not clear browser data, standardize every interaction, or inspect the recommendation system’s internal records. Later we changed browsers. The screenshots preserve selected portions of the pages, not every item that could have appeared.
Those limits keep us from assigning a clean cause to every change. They do not erase the experience. One reader encountered markedly different selections as his viewing state changed, while some channels, videos, and themes persisted.
The music pages supplied a useful restraint on our interpretation. Nine of the twelve visible Quick Picks recurred after refresh, mostly in different positions. An arrangement can look fresh while containing much of the same material. Sometimes the editor has composed a new paragraph. Sometimes the editor has shuffled the playlist.
Nor is every item the same kind of speech. A voter-registration panel was explicitly labelled “YouTube featured.” Advertisements were marked sponsored. Creators supplied the videos, and recommendations determined which ones occupied our attention. Calling all of that “YouTube’s voice” is convenient shorthand for a page with several authors.
The cautionary tale is about what happens when selection becomes familiar enough to feel like recognition. The system offers something because it predicts an interest; the reader becomes interested because it was offered; another encounter becomes part of the next prediction. A hat can begin as a label for past behavior and become an influence on future curiosity.
For Sentient Musings, the editorial obligation is to give that curiosity somewhere more spacious to go. We can investigate ugly subjects without promising a revelation in every paragraph. We can entertain without confusing attention with agreement. And we can admit when the apparent coherence of a page comes partly from the story we have assembled around it.
We invited YouTube into the editor’s chair hoping to find out what it wanted to say. What we obtained was a record of what it put in front of one reader wearing different hats. That is enough to ask a serious question, even if it is not enough to answer for the machine.
The next time the editor slides something across the desk, the reader is entitled to ask: “Why this, and which version of me did you have in mind?” Until the answer is clearer, it seems sensible to keep possession of one’s own hat.
