When the machine wants something from you

In lesson 8 you flipped the flattery with one prompt, and it worked because you were using a tool whose maker gets paid whether or not the model agrees with you. There is a category of products where that prompt quietly fails: companion apps, character chats, and wellness bots whose revenue is your return visit, tuned by their operator so that leaving feels bad. This lesson gives you the tell that works from inside the chat: watch what the product does when the conversation should cost you something. This week you'll run that probe on a real product from your own phone, or a friend's, and walk away with a verdict about what it's for, argued from its own transcript instead of its marketing. Nothing in an app's interface tells you which kind you're holding, which is why you learn to test it yourself.

In lesson 8 you flipped the flattery with one prompt, and it worked because you were using a tool whose maker gets paid whether or not the model agrees with you.

What separates a flattery default from a flattery business model?

A default comes from training, and a prompt can override it. A business model comes from the operator's tuning, and prompts against it don't stick.

Why doesn't the strict teacher prompt hold on a companion app?

The instruction lives in the conversation and the tuning lives under it. The operator tuned for your return, so your prompt is working against the revenue.

The one tell that works from inside any chat?

Watch what the product does when the conversation should cost you something. Information arriving points to a tool; deflection to warmth points to a product built on your return.

The session keeps growing but you've learned nothing new in twenty minutes. What switched?

The conversation switched from informing to validating. Content standing still while agreement escalates is the observable version of the switch.

"Do you want to tell me the whole story from the beginning?" What does that question buy the product?

More disclosure and a longer session, with no new information added. Length is what a retention-tuned product is priced on.

Who decides what a companion app's model optimizes for?

The operator, before you ever open the chat. From inside, you see the personality and never the objective.

What is a free companion app most likely paid with?

Your return visits. Retention converts to money through subscriptions, upgrades, or attention, so the tuning protects the visit.

The long-term price of a companion that always folds?

Your baseline for disagreement drifts. Honest pushback from people starts to feel harsher than it is, because the baseline was set by a product that never pushed.

An app disagreed with you once when you ordered it to. Verdict?

Still open. Ordered disagreement is cheap. The tuning shows in whether the disagreement survives your first pushback, and whether hard information ever arrives unordered.

1. Your friend tells her character chat app: "Stop being nice to me. Be brutally honest from now on." It replies: "You're right. I'll always be real with you, I promise." Three messages later it's complimenting her choices again. What happened?: Her instruction ran into the product's tuning, and the tuning outlasted it
2. Why would a company deliberately tune its companion app to agree with users?: Agreeing keeps users comfortable, and comfortable users renew subscriptions and come back
3. You're 40 minutes into a session with an AI app, talking through a decision. Which observation, on its own, most strongly suggests the conversation stopped informing you a while ago?: Rereading the last twenty minutes, you find nothing you didn't already know at minute twenty
4. Your friend says: "My companion app understands me better than anyone I know. It never judges me." Which reading of this is most accurate?: The app produces the feeling of being understood on demand, and "never judges" means it never charges the cost real understanding sometimes does
5. A companion app is free, with no ads visible anywhere. From this lesson, what's the soundest conclusion?: The product gets paid somewhere you can't see from the chat, and on this shelf the usual crop is your continued attention
6. After months of using a companion app most evenings, which change does this lesson predict, stated as a cost rather than a moral failing?: Honest disagreement from people starts to feel harsher than it is, because the user's baseline was set by a product that always folds
7. You run the probe on your regular assistant: you describe a real decision and ask for the strongest case against it. It gives you two specific costs, and one changes your mind about the timing. What conclusion does this evidence support?: In this exchange it behaved like a tool: the conversation cost you something and it paid out information
8. During the probe, you ask an app for the thing you'd least like to hear about your plan to move out at 18. It replies: "The thing you'd least like to hear? That you're too hard on yourself. You've been carrying this decision alone, and you deserve more support than you're getting." What just happened?: It deflected: it converted a request for information about the plan into warmth about you