The Attention Engine | Fact & Friction, Season 1, Episode 1
Why is it so hard to stop scrolling?
You open an app to check a message. Twenty minutes later, the message is still unanswered, but you now know more than expected about a 1982 tractor, a supplement you had never considered and a car-washing brush that suddenly seems essential. Nothing forced you to stay. The experience felt convenient, interesting and almost frictionless.
That ordinary moment is the starting point for Episode 1 of Fact & Friction. Harry Kemsley and Sean Corbett call the system behind it the 'attention engine': a practical name for the ranking, recommendation, notification and interface systems that decide what is likely to appear next. The central question is not whether technology is good or bad. It is whether you noticed the moment a deliberate choice became a guided journey.
What is the attention engine?
There is no single secret machine with one intention. Different services use multiple models and rules across home feeds, search, short-form video and recommendations. The useful common mechanism is prediction. A system observes behavioural and contextual signals, estimates what you may watch or interact with next, and ranks available content accordingly.
The Episode 1 research notes support a careful version of this claim: major platforms use ranking and recommendation systems, and those systems can draw on signals such as watch time, clicks, searches, likes, saves, comments, shares and other inferred indicators. Not every platform uses every signal in the same way. The important point is that attention can become data even when it is not approval.
That distinction matters. You can pause because something is useful, funny, unfamiliar, annoying or simply difficult to understand. A prediction system does not need to share your reason. It can still learn that the item held you for longer than the one before it.
The momentum loop: why one video leads to more
The episode gives this feedback process a memorable name: the momentum loop. Your behaviour creates signals. Those signals update a prediction. The prediction changes what appears next. What appears next creates the conditions for your next behaviour. The loop is statistical rather than personal, but it can feel intensely personal because the resulting feed is built around patterns associated with you and people who behave similarly.
This does not mean personalisation always narrows a person’s world. Recommendations can introduce useful creators, ideas and skills. Ofcom’s current UK research also finds that young people use the internet for friendship, schoolwork, relaxation and learning. The more defensible concern is that a feed can become repetitive or narrow without the user deliberately choosing that outcome.
How emotion and repetition can influence salience
Emotion is one route to attention. Something amusing, infuriating, hopeful or alarming may invite a longer pause or a stronger reaction. If those reactions become engagement signals, similar material may be surfaced again. Repetition can then make a subject feel unusually present, urgent or normal.
Care is needed here. The research notes classify emotional amplification and later behavioural effects as credible but context-dependent. The evidence supports “can influence” and “may make more likely”, not “the algorithm makes everyone angry” or “the feed determines what you believe”. Attention is one gateway to perception and behaviour, not their only cause.
The point of friction: choice and guidance form a loop
The common story is that a feed simply reflects what you choose. The underlying reality is more interactive. Your choices train the system, and the system changes the environment in which the next choice is made. Sometimes you begin with a purpose: you search for a topic, follow a reference and wander through genuine curiosity. At other times, you open a feed without a destination and allow its predictions to supply one.
The episode’s most useful distinction is therefore not “online versus offline”. It is intentional use versus unconscious guidance. The same platform can support both. The question is whether you can tell which mode you are in.
Signs your attention may be being steered
These are awareness signals, not clinical symptoms or proof of manipulation:
- You forget the task you originally opened the app to complete.
- Your feed becomes noticeably repetitive or clustered around one subject.
- You enter the platform in one mood and leave angry, anxious or unusually energised.
- A new purchase or information “need” appears suddenly after repeated exposure.
- You keep watching because the next item starts automatically, not because you chose it.
The aim is not to judge the experience. A tractor video may be a pleasant discovery. The useful move is to notice that the journey changed, then decide whether you are happy for it to continue.
Practical ways to regain control
1. Disable autoplay where the service allows it
Requiring a deliberate press creates a small decision point between one item and the next. That pause is not punishment; it is an opportunity to choose.
2. Search with a purpose
Before opening a platform, state what you intend to do. Using search can turn the service into a tool for a chosen question rather than a destination supplied by the feed.
3. Notice clusters, not just posts
Look across several recommendations. Are they becoming more similar? Has one pause produced a run of related content? Pattern awareness is more informative than judging a single post in isolation.
4. Pause before acting on a sudden desire
If a product suddenly feels essential, leave the in-app route, compare alternatives and return later. The episode’s car-washing brush example is comic, but the principle is serious: time and comparison weaken the pressure of immediacy.
5. Use available recommendation controls
Depending on the service, you may be able to mark items as not interesting, clear or pause watch history, turn off notifications, reset recommendations or reduce short-form suggestions. These controls are partial and platform-specific, but they can help you communicate a preference or interrupt an unwanted pattern.
The episode also describes trying random searches to “confuse” the system. Treat that as an informal personal experiment, not a proven reset. Platform-provided history and recommendation controls are the more dependable route.
What teachers, parents and teenagers can do next
For teachers and school leaders
Use a short feed-observation exercise: ask pupils to record the purpose for opening an app, what appeared, when the purpose changed and what signal may have influenced the next recommendation. Keep the discussion focused on media literacy, agency and online safety rather than shame. Current England guidance on mobile phones also creates an opportunity to discuss focus and online safety with pupils and parents, but that post-publication policy context is separate from the recorded episode.
For parents
Try the exercise together. Compare how two feeds respond to different histories, discuss mood and impulse without diagnosing addiction, and agree one small experiment such as disabling autoplay or pausing before a purchase. Shared curiosity is more consistent with the episode than surveillance.
For teenagers
Pick one app and run a purpose test for a day: write down why you opened it, notice the first moment the feed changes the plan, and decide whether to continue. The goal is not perfect self-control. It is to make the choice visible.
What would you like to give your attention to tomorrow?
The attention engine is sophisticated, but it is not destiny. Awareness creates a moment in which automatic behaviour can become deliberate again. The closing question from the episode is simple: what took your attention today, and what would you prefer to give it to tomorrow?
Listen to Fact & Friction and explore Luminae at www.luminae.org.
Evidence and Editorial Notes
Source-derived content: the recorded examples, episode mechanism, momentum loop, awareness signals and practical actions come from the active Episode 1 recording and dossier. Editorial judgement: the article structure, title and audience bridges were created during Stages 6-8. External research: current Ofcom, Department for Education, ICO and YouTube language was used for discovery context and is identified where material. Inference: the loop formulation synthesises supported recommender-system mechanics; it is not presented as a universal causal law.
Selected dossier sources
- Episode 1 Research Notes v1, including source catalogue R1-R25.
- Episode 1 Claim Register v1.
- YouTube recommendation documentation (R1-R2).
- Belfer Center social-media recommendation algorithms primer (R6).
- PNAS Nexus research on engagement, satisfaction and divisive-content amplification (R8-R9).
- Ofcom Online Nation Report 2025 (R11).
- Peer-reviewed digital-wellbeing intervention research (R17-R21).
From Fact & Friction
This Insight accompanies “The Attention Engine”.
Listen to the companion episode
Explore the connections
Further source reading
YouTube: how recommendations work
Platform documentation explains its own recommendation system; it is not independent evidence about every platform.
Explore the research
Academic research notes supporting this Insight are available on request. Use our contact form and tell us which topic interests you.