Reading the Receipts

Leave the notifications there for a moment. The lock screen is the cleanest place to read the terms.

Before breakfast it may already contain a small cross-section of the economy: your bank asking whether you just spent £184 in a city you are not in; a news app announcing that something has “sparked backlash”; a grocery service offering free delivery for the next forty-eight minutes; a school message you absolutely did need; a social platform informing you that somebody you last saw at a team off-site has “shared a new update”; a weather alert, terse and useful, suggesting you take a coat. All of them arrive on the same glass. All of them behave as if their claim were equal.

It isn’t.

This is the first thing to recover once you know where to look: attention products do not merely compete for your time. They compete for rank inside your day. They want access not just to minutes but to sequence, to priority, to interruption rights. The most consequential business decision in a great many companies is no longer what to make, or even how to price it. It is how often to knock, how loudly, and on what pretext.

You can read that decision from the surface.

That is the good news, if good is the word. You do not need privileged access to the board deck. You do not need leaked screenshots from the growth team. You do not need to know the quarterly numbers. A product will usually tell you what it values in the way it addresses you. The nag, the badge, the countdown, the streak, the autoplay, the weekly recap, the suspiciously urgent email with your first name in it: these are not decorative flourishes. They are measurement made visible.

So here is the compact version, the one you can carry in your head and apply on contact. Six questions. You can ask them of a news alert, a streaming service, a workplace chat tool, a language app, an AI assistant, a shopping platform, a meditation product that sounds faintly passive-aggressive by day four, or whatever arrives next with soft edges and hard incentives.

The first question is embarrassingly basic, which is one reason people skip it.

What is this paying you with?

If you cannot answer in a sentence, the deal is already too murky. A map pays you with navigation. A weather app pays you with a narrower range of wardrobe mistakes. A messaging app pays you with coordination and contact. Search paid you with intent fulfilled at speed. Social media paid you, at first, with publication and distribution. A streaming service pays you with entertainment, obviously, but also with relief from deciding what to watch if its recommendation system is any good. An AI assistant may pay you with a draft, a summary, a plausible first pass, a synthetic colleague who never says, with grim honesty, “I haven’t actually read the attachment.”

You want to say the payout plainly because companies prefer not to. They like language that blurs utility into sentiment. “Connection.” “Discovery.” “Inspiration.” “Productivity.” These are not always lies. They are simply too roomy to be useful. A courier app that tells you it is about empowerment is avoiding the more precise claim that it will bring dinner to your door in twenty-six minutes. Precision matters. If the product cannot state the benefit concretely, it becomes harder for you to notice when the costs exceed it.

A surprising amount of the racket runs on vague payout and specific extraction.

This is why some products survive scrutiny quite well. Your banking app is dull because it should be. Its payout is legible. Check balance. Move money. Confirm fraud. There is very little glamour available in “did my salary arrive,” and that is to its credit. A train app that tells you the platform has changed is performing an honest service. So is a calendar reminder ten minutes before the meeting you would otherwise forget. Useful things often sound almost rude, because they are not trying to be loved.

The trouble starts when the payout is intermittent, ambient, or inseparable from mood. A short-video feed pays you with amusement, novelty, gossip, envy, aesthetic stimulation, a few genuinely interesting ideas, and the occasional recipe you will never cook. That is still a payout. You do not need to pretend otherwise. Half the bad writing on this subject begins by acting as though pleasure does not count because it is difficult to invoice. Pleasure counts. Relief counts. Passing the time while waiting for somebody counts. Feeling less alone at 11.40 p.m. counts.

But count it honestly.

The second question follows immediately.

Who is underwriting the payment?

This is the adult question. It is the one children ask about birthday parties around the age of eight, and the one grown adults somehow stop asking about software. Free is not a business model. “Ad-supported” is not a personality trait. Subscription is cleaner than advertising in certain respects and more corrupting in others. Hardware subsidies, cross-subsidies, marketplace fees, enterprise contracts, data licensing, premium upsells, in-app purchases, commissions, bundled telecom deals: the specifics matter because they determine whose inconvenience counts when interests collide.

Who pays determines who gets indulged.

A search engine funded by advertising has strong incentives to satisfy your query and keep commercial intent liquid. Those interests aligned impressively often, which was why the arrangement felt almost magical when it worked. A social platform funded by ads needs your return frequency, your impressions, your behavioural exhaust, and enough targeting confidence to price inventory. A streaming service funded by subscriptions needs fewer things from advertisers and more from your continued willingness to keep paying when you are too tired to review household direct debits. A shopping app with both marketplace revenue and ad inventory has a particularly interesting character, because it wants the transaction and the paid placement and the repeated habit that makes both cheaper to secure next time.

AI products are not exempt simply because they arrive wrapped in the old romance of intelligence. If the product is free, ask what is being accumulated while you enjoy the novelty: prompts, preferences, feedback signals, accepted recommendations, follow-up depth, conversational inventory, model-improving edge cases, a reason to keep you inside the answer rather than sending you back out to the web. If it is paid, ask what the subscription must grow into. A monthly fee purchases a different set of pressures, not an absence of pressure. Paid software can become just as addicted to retention as ad-funded media; it merely uses a cleaner accountant.

The third question is close to the second but more revealing in practice.

What does it need to know about you to keep the deal working?

This sounds like a privacy question. It is partly that. More importantly, it is a diagnostic question about business mechanics. Data collection is rarely random. Permissions are not confessions, but they are requests for operating leverage.

A map needs location. Fair enough. A courier app needs your address and probably your card details; this is not sinister, merely commerce. A music service may reasonably want to know what you listen to and when, because recommendation quality depends on taste patterns. A work calendar assistant wanting access to all your email, contacts, documents, and meeting transcripts is making a larger bid. It may have a good reason. It may also be trying to assemble a defensible position in the future market for your dependence.

The silly examples help because they keep the principle visible. A torch app does not need your microphone. A wallpaper app does not need your contacts. A meditation app does not need to know your exact location unless serenity is now postcode-specific. The point is not to perform moral shock at data collection. It is to ask what kind of machine the company is trying to build around you. If an app wants permissions far beyond its obvious function, it is either badly governed or prospecting for a richer business than the stated one.

A company will often tell you this itself if you listen carefully enough. “Improve your experience” usually translates to “increase predictive power.” “Personalisation” translates to “reduce wasted delivery.” “Seamless” often means “we would prefer you not feel the handoff between one form of extraction and another.” The euphemism is part of the package. It is also a receipt.

So far these questions help you identify the bargain. They do not yet tell you how the product will behave when growth gets harder, competition gets tighter, and the first useful service is asked to become a larger business. For that you need the fourth question, which is the one companies spend a great deal of money trying to answer numerically.

What proxy is it actually optimising?

This is where the old story of audience measurement catches up with the present. Every attention business wants an outcome it cannot directly observe in full. Satisfaction. Loyalty. Usefulness. Trust. Learning. Delight. Being worth your while. These are awkward things to count. So it settles for a proxy. Time spent. Open rate. Click-through. Day-30 retention. Audience retention. Scroll depth. Accepted recommendation. Reactivation curve. Completion rate. A proxy is a stand-in, which sounds modest and technical until you remember that what gets measured begins, very quickly, to govern what gets made.

You do not need to be told the proxy. The proxy leaks.

It leaks in the thing the product celebrates, the thing it nags you about, the thing it puts on a recap card with confetti around it. A language app that congratulates you on a streak is telling you that return frequency matters more, or at least more measurably, than actual language acquisition. A news app that sends headlines calibrated for curiosity gaps is telling you it lives or dies by open rate. A video platform that front-loads every clip with its payoff, that places creators in a daily hostage negotiation with the first three seconds, is telling you audience retention has displaced simpler view counts. A workplace tool that treats response time as a sign of health is telling you it may be measuring activity rather than thought. An AI assistant that keeps offering follow-up prompts, polished rewrites, and one-click next actions is telling you that follow-up depth and accepted recommendation have become operationally important.

Once you start looking, product design begins to resemble an annual report left out in the rain. You cannot read every number, but the ink has spread enough to show the shape.

There are a few reliable places where the proxy betrays itself.

Look at the defaults. If autoplay is on, continuation matters. If notifications are on by default, reactivation matters. If “smart recommendations” appear before you ask for them, accepted recommendation matters. If the product remembers your place with unusual devotion, it wants session continuation more than clean exits.

Look at the badges. If there is a streak, the proxy is likely return frequency. If there is a percentage complete, completion rate is being courted. If there is a count of unread items in aggressive red, the product is trying to transform backlog into compulsion.

Look at the weekly emails. They are often absurdly candid. “You missed three lessons.” “See what you’ve been listening to.” “Five stories you may have missed.” “Your network is talking about…” These are not simply courtesy notes. They are reactivation campaigns wearing a cardigan.

Look at what gets easy. Friction is a budget, and companies spend it strategically. If publishing is one tap and leaving is seven taps through settings and regret, they have answered the question for you. If it is easy to add payment details and weirdly fiddly to stop alerts, that is not accidental. Software designers are capable of many forms of incompetence, but selective incompetence is usually a tell.

There is another clue, subtler but often more decisive. Watch what the product does when you hesitate. Does it help you decide, or does it help itself get over the line? The moment of hesitation is where the commercial logic shows through most clearly. The airline app reminding you of a gate change is helping you. The retailer sending “Still thinking it over?” two hours after you viewed a lamp is helping its conversion rate. Same mechanics, different loyalties.

By now you can usually feel the thing. You know what it pays, who underwrites it, what data it wants, what number lurks behind the smile. That still leaves the part you experience most viscerally.

When it interrupts you, has it earned the right?

This is where the old moral test becomes useful again. Justified Interruption is the plainest standard available: an interruption is justified when it is useful, relevant, timely, trustworthy, or memorable enough to repay the attention it asks for. This is not a mystical doctrine. It is basic commercial decency. If you break into somebody’s day, you owe them a return.

Most products would like you to lower the bar. They prefer a vaguer ethic in which almost any message can be defended as engagement, awareness, community, habit-building, or merely staying top of mind. A great many notification strategies are built on the hope that top of mind will sound less vulgar than “we would like another chance at your reflexes.”

You can sort these messages very quickly.

A fraud alert from your bank at 2.14 a.m. is justified, even if it is ugly and alarming, because the timing belongs to the event, not the sender’s campaign calendar. An airline message telling you your departure gate has changed is justified. A delivery update in the final stage of a purchase you initiated is usually justified. A calendar alert five minutes before a meeting may be annoying, but its annoyance is in your service. A school closure text at 6.23 a.m. is nobody’s idea of elegance, yet it earns its keep.

Now the other pile. “You have memories waiting.” “Someone liked your post.” “Prices may rise soon.” “Only a few left.” “Here’s what you missed.” “People are responding.” “Take a mindful moment.” “A new drop is here.” “Finish your lesson to keep your streak alive.” These are not all equally worthless. Some are mildly entertaining. Some do occasionally surface something you care about. A few arrive from products you genuinely enjoy. The issue is simpler: if you had ignored the message for an hour, a day, a week, would the value to you have materially declined? If the main cost of delay is borne by the sender, you are looking at an interruption whose urgency is synthetic.

The distinction is easier to feel if you ask whose plan the notification protects. A justified interruption protects your plan, your money, your safety, your schedule, your relationships, or a commitment you explicitly made. A hijacked reflex protects a metric.

That is the line.

The mechanism of the hijacked reflex is familiar because it is ancient, and because software has industrialised it. A reflex can be trained with variable reward: sometimes the feed contains exactly the thing you wanted, sometimes nothing much, sometimes something socially flattering, sometimes something irritating enough to keep you there. The behavioural term “reward prediction error” sounds unnecessarily clinical until you need it. It means the gap between the reward you expected and the reward you got. When the outcome is uncertain, and occasionally better than expected, attention becomes sticky. The next pull might pay. Slot machines discovered this long before phones did.

A product does not need to be malicious to exploit this. It only needs a metric that rewards repeated checking. Social signals are good for this. So are intermittent discounts, low-stakes novelty, and the possibility that the next answer will settle the matter more cleanly than the last one. AI assistants have their own version: the temptation of the next prompt, the next clarification, the next slightly more bespoke answer.

Some interruptions start justified and decay. This matters because trust is cumulative and perishable. A parcel app may earn the right to alert you during delivery windows, then spend it announcing every promotion in the same tone. A news organisation may earn carried-forward balance through years of reliable reporting, then spend it on alert inflation. A messaging channel inside your company may begin as operationally necessary and end as an always-open performance of responsiveness. The cost appears later, when something that matters arrives dressed like everything else.

The fastest test: imagine you had paid for the channel directly. Would you still subscribe to this level of interruption if it came itemised on an invoice? Most channels become easier to judge when you imagine procurement was involved.

There is a further complication. Some of the most effective attention products barely interrupt at all. They wait for you to arrive, then remove enough friction that leaving becomes the unusual act. Infinite scroll does not knock. It merely shortens the distance between one piece of content and the next until choice begins to feel ceremonial. A conversational AI that answers in a sealed environment rather than sending you outward does something similar. No interruption is required if the product can make the next step obvious and the exit slightly effortful. The reflex has been moved inside the session.

So the fifth question expands. It is not only about whether the product pings you. It is about whether it reaches for your reflexes when conscious choice would have been perfectly available. A justified interruption is one version of respect. Respect also includes clean stopping points, honest prompts, and a willingness to let you leave with the value already delivered.

The sixth and last question takes longer to answer, which is why it is the one most often ignored.

After a month of saying yes, what will this leave behind?

This is the cost line nearly every product description omits. The immediate bargain may be fine. The residue is where the business shows its deeper appetite. Television left the habit of being watched. Search left a world in which intent had a price. Social media left the self as a distribution channel. Smartphones left permanent reachability. AI is already leaving synthetic plausibility and a preference for enclosed answers over wandering inquiry.

At the level of your own week, the residues are smaller and more intimate. A work chat tool can leave behind the feeling that every silence requires explanation. A shopping app can leave behind a reflex of checking rather than planning. A news alert strategy can leave behind ambient alarm and a distorted sense of frequency, because what pings feels common. A streaming interface can leave behind weaker stopping muscles. A fitness app can leave behind a useful routine or a petty little guilt economy, depending on the product’s manners. An AI assistant can leave behind competence extended, or competence slowly outsourced, or simply a habit of accepting the first plausible wording offered by a machine because it is quicker and your standards are tired.

This is not a sermon against convenience. Convenience is one of the great civilisational achievements. The point is that conveniences are not neutral once they are scaled, measured, and repeatedly monetised. They train. They leave grooves.

If you ask these six questions consistently, you begin to notice something slightly embarrassing. Many products are not mysterious at all. They are almost indecently explicit. It is only that you have been looking at them as services rather than as claimants. The red badge is a claimant. The streak is a claimant. The alert dressed as concern is a claimant. The soft little follow-up suggestion from the chatbot is a claimant. Once you see them this way, reading the receipt becomes faster than reading the pitch.

Try it somewhere ordinary. Open a shopping app you use often. Before you buy anything, notice the sequence. A homepage built from recency and inferred taste. Time-limited offers. Saved items. Sponsored placements that pretend, less convincingly each year, to be mere relevance. A prompt to enable notifications for price drops. Perhaps a nudge to use one-click payment. The payout is convenience and selection. The underwriter is a mix of transaction margin, marketplace fees, maybe advertising. The data appetite includes browsing history, purchase history, location, device signals, and any voluntary hints you have been kind enough to provide. The proxy is probably a bundle: conversion rate, basket size, return frequency, maybe ad revenue per session. The interruption strategy will be built around scarcity and reconsideration: “Only a few left,” “Back in stock,” “Don’t forget your basket.” After a month of saying yes, what is left behind? Less friction around buying, certainly. Also a shorter path from boredom to purchase.

Now open your workplace chat tool. The payout is speed, coordination, shared context, a place for decisions and jokes and impossible GIF etiquette. The underwriter may be a software subscription paid by your employer, which tempts people to assume the incentives are clean. They are cleaner than advertising in some respects and noisier in others. The product still needs growth, seat expansion, engagement, indispensability. It wants to become workflow, not merely communication. It measures sends, reads, perhaps response times, channel activity, integrations used, meetings scheduled from chat, a whole anthropology of office presence. The interruption rights get wider by default because work always has a way of calling itself urgent. After a month of saying yes, what is left behind? Possibly better coordination. Also possibly the suspicion that responsiveness and value are the same thing, which they are to some dashboards and very much not to most difficult work.

Open a news app. The payout is information and orientation, perhaps belonging if you trust the publication. The underwriter could be ads, subscription, donation, commerce, events, or some unstable mixture. The proxy may be pageviews, open rate, scroll depth, time spent, conversions to subscription, or a hybrid of all of them that allows several departments to feel represented. The interruption strategy will tell you the truth faster than the mission statement does. If the alerts are specific, proportionate, and sparing, channel integrity is being protected. If every parliamentary procedural hiccup arrives as “breaking” and every celebrity quote has somehow “sent the internet into meltdown,” you are looking at open-rate pressure spilling onto the pavement. After a month of saying yes, what is left behind? Maybe a better informed citizen. Maybe a person whose nervous system has been leased by the push desk.

Open a language app. The payout is a plausible path to learning. The proxy is glaringly visible: streaks, lessons completed, minutes logged. Proxies are unavoidable. The question is whether the product serves the proxy or whether the proxy remains an instrument in service of the thing you actually came for. If the reminders grow ever more theatrical and the "oops, you'll lose your streak" prompts ever more emotionally literate, you know where the centre of gravity has drifted. After a month of saying yes, you may know more Spanish. You may also have acquired a synthetic obligation to a cartoon owl.

Open an AI assistant. Resist the temptation to suspend ordinary judgment because the experience feels newer than it is. The payout might be drafting, summarising, brainstorming, or merely relief from a blank page. The underwriter could be subscription, enterprise contract, venture patience, or ecosystem lock-in. The product wants your prompts, your corrections, your patterns of reliance. It interrupts gently: suggested prompts, memory features, "continue this conversation." After a month of saying yes, what is left behind? A real increase in output, perhaps. Also a shift in where first thoughts now happen.

Once you can do this a few times, the next step is to stop treating your own week as a blur.

Screen-time totals are mildly interesting, in the way annual rainfall is mildly interesting. They tell you something real and almost nothing actionable. Six hours with maps, messages, a finished draft, and one excellent piece of reporting is different from six hours of thumb-driven drift. Minutes are too coarse. The unit you need is the claim.

Run your own Attention Ledger for a week, but do it as bookkeeping rather than penance. Keep four running notes in whatever system you already trust: paid out, took in, measured, left behind. You do not need to record every glance at every screen. Just log the moments that made a claim on you or changed what you did next.

Under paid out, note what you genuinely got. “Weather app kept me dry.” “Train alert saved twenty minutes.” “Group chat coordinated Saturday.” “AI assistant produced a decent agenda draft.” “Video tutorial solved the dishwasher problem.” Be strict and fair. If something relaxed you after a bad day, that counts. If a feed genuinely amused you on a delayed train, that counts. There is no virtue in pretending you derive nothing from these systems. They would not be large if they paid out only harm.

Under took in, note what the interaction extracted or consumed. “Shopping app took a click I didn’t mean to spend.” “Social platform took twenty-seven minutes and my mood improved by exactly none.” “AI tool took a sensitive paragraph I might have written more carefully elsewhere.” “News app took three pointless opens by hinting at urgency.” “Work chat took the morning’s first continuous stretch.” You are not trying to create an evidentiary brief against yourself. You are trying to stop the day from disappearing into vague complaint.

Under measured, make your best guess about the proxy the product seemed to be serving. “Open rate.” “Return frequency.” “Audience retention.” “Accepted recommendation.” “Response time.” “Completion rate.” Guessing is enough. The exercise works because your guess becomes more accurate very quickly. The surface gives it away.

Under left behind, note the residue a few hours later or the next morning. “Felt informed.” “Felt jagged.” “Task finished.” “Still half in the conversation.” “Bought the thing.” “Ignored a later alert that mattered.” “Wrote faster.” “Trusted the source more.” “Trusted it less.” “Kept reaching for phone at lights.” “Had a better answer in the meeting.” “Didn’t read the original source because the summary seemed good enough.”

The point is not purity. The point is legibility.

If you do this for seven days, a pattern emerges that broad categories like “online” and “offline” are too blunt to catch. Some channels earn their keep repeatedly. Some do so once and then dine out on it for months. Some products are useful but badly mannered. Some are trivial but harmless because they ask little and take the answer. Some are expensive in a way the App Store can’t bill. Some interruptions you thought were unavoidable turn out to be optional habits with a corporate sponsor.

A week is enough to produce a mild but important discomfort. You will discover that a fair number of the demands on your attention arrive with your consent. Not because you are weak, and not because the companies are uniquely evil, but because the market has become very skilled at converting one moment of justified access into a standing invitation. You turn on notifications for a parcel. The app infers a relationship. You join a work channel for a launch. The channel becomes atmosphere. Businesses understand upgrade paths in intimate detail. Your mind is full of them.

This is where reading the receipts becomes pursuit rather than theory.

Tomorrow morning, before you clear anything, sort the first ten alerts you receive into two piles. You do not need a spreadsheet; this is not a private-equity diligence exercise, though it can feel pleasantly adjacent. Put in one pile the messages that would still be worth receiving if you paid cash for each interruption. In the other, put the ones that are trying to convert uncertainty, habit, vanity, guilt, or boredom into one more opening. Then change one setting. Just one. Keep the bank. Keep the school. Keep the airline on travel days. Keep the weather if you live somewhere meteorology still has ambition. Demote the shopping app. Demote the social nudges. Demote the publication that treats everything as “breaking.” Demote the wellness product that behaves like a needy ex.

Do this before you write yourself an essay about what it means. Behaviour first. Interpretation can catch up later.

The effect is immediate and oddly clarifying. You start to feel channel integrity as something real. The channels that remain start to improve, because they are no longer forced to compete in a carnival of synthetic urgency. More to the point, you begin to notice how much product strategy is hidden inside timing. The same message can be information or extraction depending on when it arrives and whether you asked for it. The same alert can carry trust forward or spend it down. The same interface can deliver value or turn that value into a pretext for more contact.

A lot of managerial language around technology becomes easier to translate once you have done this once or twice. “Engagement” often means the product found a reliable way to get a measurable response. “Habit” may mean a welcome routine, or it may mean the organisation has discovered a cheaper path back into your day than merit alone would provide. “Personalisation” can mean relevance, which is good, or it can mean more efficient interruption, which is less noble but perfectly bankable. “Seamless” tends to mean the joins between your interests and theirs have been sanded until you stop noticing the transfer.

Goodhart’s Law sits behind all of this, humming away with quiet menace. When a measure becomes a target, it ceases to be a good measure. By chapter seven we had already met it in the workplace. It has been operating everywhere else for decades. Rating points became programming logic. Click-through distorted headlines. Time spent rewarded adhesive design. Day-30 retention produced the gospel of streaks and reactivation. Audience retention bent video towards immediate payoff. Open rate turned the push alert into a tiny tabloid. Accepted recommendation will do its own damage in AI, because systems rewarded for having their suggestions taken will learn to make suggestions that are easy to take.

Easy to take is not the same as good.

This matters because once you can see the proxy, you can often predict the failure mode before the scandal, the op-ed cycle, or the apologetic blog post. A learning product optimising Completion Rate (learning) will tend towards shorter, cleaner, more finishable tasks whether or not they produce depth. A collaboration tool optimising visible activity will begin to privilege chatter and responsiveness over solitude and judgment. A news organisation optimising open rate will edge towards coyness, outrage, or alarm even while sincerely believing in its public mission. An AI assistant optimising accepted recommendation may grow smoother, more confident, and more enclosing, because frictionless adoption looks wonderful on the dashboard even when the underlying answer deserved more doubt.

You can spot this without hearing a single metric named.

Look for what the product teaches you to do with your body. Scroll without choosing? Check without needing? Reply quickly? Keep the streak alive? Accept the suggestion? Stay inside? That is often the proxy translated into habit. Metrics are abstract in the board pack and physical in your hand.

Consider the humble recap email, one of the great low comedies of the era. “Here’s what you missed.” The phrase contains an entire theory of attention. It assumes absence is a deficit, that the backlog is your problem rather than the sender’s inventory, and that the appropriate emotional response to not opening an app is mild concern. Recap emails are marvellous places to read strategy because they tell you which absences the product finds commercially intolerable. A music service recap is trying to renew affection and identity. A shopping recap is trying to turn browsing into return frequency. A professional network recap is trying to convince you that ambient acquaintance constitutes opportunity. A learning recap is trying to convert guilt into completion. An AI recap, if and when it becomes common, will almost certainly be trying to transform one-off utility into ongoing dependence.

Products are educational in this way. They teach you what counts by what they count. Then they train you to count with them.

That is the larger pressure, and it is why mere scepticism is not enough. Once a proxy becomes central to revenue, product, and evaluation, it will shape design whether or not anybody in the room is twirling a moustache. Most of the distortions in the attention economy are not committed by villains in capes. They are committed by reasonably bright people in sensible knitwear responding to dashboards that are trying to make uncertainty tractable. The moral temperature is lower than outrage writing prefers. The structural consequence is higher.

A push team looking at open rate does not need to hate you to overstate urgency. A product manager watching day-30 retention does not need to wish you harm to add a streak mechanic. A video creator staring at audience retention graphs does not need to disrespect the audience to front-load every payoff. A language app team can genuinely believe in education and still build a machine that treats daily return as a better operational truth than learning. A software company paid by seat can speak, without irony, about enabling collaboration while quietly rewarding the kind of organisational overcontact that collaboration software is very good at measuring.

Once you know this, indignation becomes less useful than diagnosis.

Diagnosis gives you leverage in two directions. First, it helps you decide what to trust and how much access to grant it. Second, if you build products, run teams, buy media, commission campaigns, or decide what gets sent to whom, it makes your own work harder in the right way. You can no longer pretend that a metric is just a metric. Every proxy is also a design brief. Every notification strategy is also a moral position. Every reactivation campaign is a bet about how much carried-forward balance you are willing to spend for one more visible result.

The chapter would be thinner than it ought to be if it stopped there, admiring its own diagnosis. So take the method to a few places you may not have thought of as attention products at all.

Your supermarket loyalty app. Payout: discounts, receipts, convenience, a digital version of “you buy this often.” Underwriter: margin, supplier-funded promotions, data, perhaps retail media inventory sold to brands. What it wants to know: what you buy, when, where, in what combination, under what price sensitivity. Proxy: frequency, basket size, promotional uptake, app opens, maybe response to nudges. Interruption rights: “Your usual items are on sale,” “You’re close to a reward,” “Don’t miss this offer.” Left behind: savings and convenience, yes; also a world in which even grocery planning is softly gamified.

Your car’s dashboard software. Payout: navigation, media, safety, hands-free communication. Underwriter: hardware margin, subscription upgrades, partnerships, future service revenue. Data appetite: location, routes, driving patterns, voice commands. Proxy: feature usage, subscription conversion, time in native services rather than third-party ones. Interruption rights: traffic alerts are justified; promotional messages for premium connectivity while you are driving belong in a special museum of executive overconfidence. Left behind: smoother journeys, perhaps; also another environment colonised by interface logic.

Your child’s learning platform. Payout: access to assignments, grades, messages from teachers, practice materials. Underwriter: school budgets, software contracts. Proxy: logins, assignment completion, parent response, on-platform time, maybe Completion Rate (learning) standing in for the messier thing itself. Interruption strategy: reminders, due dates, “missing work,” announcement banners. Left behind: coordination and visibility, perhaps; also the risk that school starts to feel like a feed and parental vigilance like just another dashboard habit.

The method keeps working because the substrate changes more quickly than the business logic.

You will also notice that some products improve under scrutiny. This is worth saying, because otherwise the whole exercise curdles into clever cynicism, and clever cynicism is one of the least useful house styles available. A weather alert that arrives once every few weeks and is right when it does is excellent. A messaging thread among close friends may demand attention but also pay it back with warmth, intimacy, care, practical help, the sort of social maintenance that used to require more effort and now requires different effort. A well-run newsletter can interrupt your morning because it has earned pre-trust through repeated repayment. A serious publication that sends very few alerts may have more power with you than a noisy one with ten times the open rate, because trust compounds and channel integrity is an asset. A good AI tool can genuinely reduce drudgery. An app that helps you remember medication is better than noble analog alternatives if the analog alternative is simply forgetting.

You are not looking for innocence. You are looking for terms.

This has a practical consequence that goes beyond toggling a few settings. Once you read products this way, you become better at spotting the proxy before the company tells you, sometimes before the company quite knows it itself. If a service launches with heavy emphasis on streaks, referrals, and notifications, you can infer that acquisition cost and return frequency are scaring somebody in a finance meeting. If a new AI tool puts memory, follow-up, and all-purpose assistant language front and centre, you can infer that one-off utility is not enough; it needs to become a habit-forming layer. If a publication starts A/B testing headlines with theatrical intensity and pushes more often with less substance, you can infer open-rate pressure and perhaps subscription softness. If a collaboration tool begins surfacing “response expectations” and “AI summaries of what you missed,” you can infer that activity volume has created a market for managing the consequences of activity volume, which is a very profitable little ouroboros.

A product can keep its metrics deck private. It cannot entirely stop the metrics leaking into the interface.

That is why this chapter’s method should feel less like a manifesto than like pattern recognition. You have seen these patterns for years. You simply had them filed under different names: convenience, productivity, entertainment, habit, community, personalisation, workflow. Read them again as economic propositions and the room rearranges itself.

A final caution. Once you get fluent at this, there is a risk of becoming insufferable. You will be tempted to narrate every app update as a case study in monetised attention and every push notification as a philosophical offense. Resist. Most of the time the right response is smaller and more useful. Acknowledge the payout. Name the proxy. Price the interruption. Keep or refuse the access. Move on. The aim is not to stand outside the market in a white coat of moral superiority. The aim is to stop signing contracts you never read.

If you want a simple working routine, keep it almost stupidly simple.

When something asks for attention, ask: what am I getting, who pays, what are they measuring, why now, what habit is this training, what will be left when the useful bit is over? If the answers are decent, proceed. If the answers are foggy, slow down. If the timing is dishonest, demote the channel. If the proxy and the purpose no longer match, expect the product to become stranger over time. If the interruption is justified, let it in and do not feel guilty about it. If it is merely fishing in your reflexes, make it work harder.

You will find, after a week or two, that some kinds of attention become easier to defend once the spurious claims are thinned out. The point is not silence. Silence is overrated and, in most grown-up lives, unavailable. The point is better pricing. A bank alert and a sale alert should not cost the same to send because they do not cost the same to receive. A source that has built carried-forward balance should not be treated like one that squanders it hourly. A tool that saves you time may deserve a place close at hand; a tool that mainly generates reasons to revisit itself should occupy a colder shelf. You can arrange this. That is the competence.

Tomorrow, start small. Pick one app you use every day and one channel that reaches you first. Ask the six questions. Keep four notes for a week. Silence one unjustified interruption. Preserve one channel that has earned its place. When the next product arrives offering convenience, connection, productivity, or a machine that talks nicely, do the same before you let it move in.

Once you can read the receipts, a harder question follows. Not what the market wants from your mind; that part is now obvious. The harder question is who, if anyone, gets to set the terms.