Algorithmic Rabbit Hole

A simple explanation

A question arrives — small, ordinary, answerable in two minutes. You open a video. The video is good. The next video the platform offers is on a slightly more specific aspect of the same question. You open it. The next one is more specific still. Three hours later, you are watching a fifty-minute breakdown of a sub-sub-topic you did not know existed at the start of the evening, and the original question — the one you opened the app to answer — is somewhere in the rear-view mirror.

This is the algorithmic rabbit hole. It is not aimless scrolling. Every step inside it feels chosen, intelligent, related. The descent has the texture of curiosity. What it does not have, when measured the next morning, is the deposit curiosity is supposed to leave.

An everyday example

You wanted to know how to season a cast-iron pan. You opened a five-minute video. The video was clear. The platform offered you, next, a comparison of three brands of cast-iron. You watched. The next suggestion was a short documentary on the history of cast-iron foundries in the American South. You watched. The next was a long-form video on the political economy of nineteenth-century iron production. You watched.

It is now two hours past when you meant to stop, and you still have not seasoned the pan. The hole was, in the strict sense, well-recommended. Each video was higher-quality than most things you might have chosen. The recommender did not mislead you. It selected, with great precision, the exact next item that would hold your attention without ever returning you to the question you originally had.

Why do rabbit holes feel productive?

Because the Reward System has learned to count engaged attention as useful learning. The two are correlated for shallow purposes and uncorrelated for any deep purpose. The hole is engaging — that is what the algorithm optimises for — but engagement and learning diverge almost immediately once a recommender is involved.

The hole also feels productive because each step is a small choice — yes, I do want to know that next — and choice produces a small felt-event of agency. The System credits agency as wisdom. The body, the next morning, knows the difference: the cast-iron pan is still unseasoned, the kitchen is the same, the supposedly learned material has dissolved into a vague sense of having watched some stuff.

The behavioral loop

A loop that hides because it disguises itself as inquiry:

1. Genuine question — a small, real curiosity arrives.
2. Initial satisfaction — the first piece of content actually answers the question reasonably well.
3. Side-door suggestion — the recommender offers an adjacent piece that is more interesting than the original answer was useful.
4. Topic drift — each next suggestion drifts a little further from the original question, in a direction that increases engagement.
5. Auto-credit — the System logs each choice as deepening understanding, even when the choices are pulling sideways.
6. Late-stage loss of thread — by the third or fourth turn, the original question is no longer recoverable without effort, and the user is committed to the descent.
7. Soft exit — the session ends from exhaustion, scheduling pressure, or a single piece of content sufficiently dull to break the spell.
8. Aftermath — the loop-runner walks away with a vague sense of having learned, and a specific awareness of having lost the time.

Emotional drivers

Three feelings that keep the descent running:

• A reward-shaped curiosity that the algorithm has learned to optimise against.
• A faint pride in being the kind of person who follows interesting ideas, even when the ideas are being chosen for you.
• A low-grade resistance to the original task — the pan, the email, the conversation — that the hole offers a graceful exit from.

What your nervous system does

The dopamine circuitry responds to information gain at a roughly logarithmic rate — small amounts of new information produce disproportionately large signals. Recommender systems are tuned, implicitly or explicitly, to maximise this signal. They feed you a tight stream of small information-gain hits, each just novel enough to fire the reward, each just adjacent enough to feel chosen.

Over months and years, the system's baseline shifts. Real reading, real conversation, real practice — all of which produce information gain at a slower, more uneven rate — begin to feel underwhelming. The loop-runner often notices they have less patience for books or long conversations and attributes it to age, busyness, or stress, when the more accurate reading is that the recommender has retuned the reward floor.

The DojoWell interpretation

The algorithmic rabbit hole is a clean instance of effort without deposit. The original system is curiosity — the impulse, evolved to drive exploration, to learn the shape of the world. The Reward System, asked to serve curiosity, supplies the substitute: a felt-event of learning. The substitute and the original share a surface — both involve attention to novel material — and they are opposite on the inside.

Real curiosity loops produce structures: a question is asked, an answer is found, the answer is integrated, the next question is enabled by the previous one. The rabbit hole produces drift: the question is replaced rather than answered, the integration step is skipped, the next question is supplied by the recommender. The deposit is near-zero. The effort, however, is large — three hours of close attention is not nothing, even when the attention was steered.

This is also why the closure_pattern is substituted. Something does close at the end of a rabbit hole — the I am done with this topic for now felt-event arrives. The closure is real. The original need — to be in honest contact with a question that mattered — was not met. The pan is still unseasoned, and the question that opened the evening is the same shape it was when the evening began.

How do I stop going down rabbit holes?

You do not stop being curious. You stop letting the recommender be the curator of your curiosity. The System will still credit each interesting step; what is workable is whether the curator is you or the algorithm.

Three moves, in order of difficulty:

1. Name the original question on entry. Before opening the platform, say or write the actual question you came in with. The named question is the only anchor against drift.
2. At each suggestion, check the anchor. Does the next piece answer the original question, or has the question quietly been replaced? Replacement is fine if you decide; replacement is the hole if it decides.
3. Stop on resolution, not on exhaustion. The original question being answered is the closure. Continuing past that point is the rabbit hole. The discipline is leaving while still interested.

Practical steps

1. Open a single tab, not the home page. The home page is the entry to a recommender; a single search result is the entry to a question.
2. Disable autoplay on every platform where you can. The pause between videos is where the question can be re-asked.
3. Set a maximum stack depth. Allow yourself two suggestions deep from the original answer; on the third, leave or take notes.
4. Take a sentence of notes per session. The act of writing one sentence at the end is the integration step the recommender removed. The sentence is the deposit.
5. Revisit one rabbit hole a week later. Almost nothing will have stayed. The forgetting is the data.

Reflection questions

• What was the original question of your last rabbit hole, and is it still unanswered?
• Which platforms most reliably substitute drift for inquiry for you?
• When did you last leave a long content session feeling more in contact with a question, rather than further from it?
• Where has the rabbit hole begun to displace the slower forms of learning — books, conversations, practice — that produce the deposits you actually want?

Frequently Asked Questions