The Belief Factory: Manufacturing Common Sense in the Answer Engine Era

I asked an AI chatbot a question the other day that was so banal my brain immediately labelled it “beige”. (Beige is such a great word when you really don’t have anything else. Do steal it.)

I didn’t ask it a moral question. Nor a political one. Just a small, practical thing. The kind of thing you’d normally type into a search engine, open three tabs, skim a couple of sources, and forget five minutes later.

And the chatbot did what chatbots do: it gave me the answer.

Calm. Confident. Nicely formatted. No fuss. No friction. Which, to be fair, is exactly what I asked for. The problem is: it’s also exactly what I wanted.

And so it felt good. I got the thing. Where’s the crime?

And then, a beat later, I noticed what hadn’t happened. I hadn’t asked who says? I hadn’t asked why this answer, in this tone, from this machine.

And I definitely hadn’t asked where it got the answer from, what it ignored, what it wasn’t “sure” about, what it was flattening into a single sentence for the sake of clarity and convenience.

All I did was accept the output as reality.

And that’s the thing with generative AI, isn’t it. Not “AI will take your job”—which is already bad enough—but “AI will become the route by which reality enters your head.” The default mechanism. The thing we take for granted.

The thing you consult when you’re lonely, panicked, curious, horny, bored, or just trying to get through Tuesday without having to dash all over the web to find an answer to a stupid question.

But, of course, the problem is: when you believe something unquestioningly, belief becomes a product.

Not “look at this, buy that”. Worse. The quieter, creepier version where the answer you’re given by the chatbot feels true because you were already pre-conditioned to believe it. We aren’t (quite) there yet, but it won’t take much.

Fair warning: This is one of those “one-door” moments. You can’t come back through the door and this not be inside your head anymore. Which is, ironically, the entire point of the article. Do come to your own conclusions, no matter what you believe after you’ve read this.

You must.

Boiling the frog

We didn’t wake up one morning and decide to rent our perception from a machine. Instead, we were warmed up. Conditioned. Nudged into it over years, by all manner of things that trained us to stop demanding so much.

“Alternative facts” was the moment the mask slipped. It wasn’t just a lie—politicians have always lied—it was something more audacious, more unholy: the permission to lie.

The move wasn’t “this happened” versus “this didn’t happen,” but: “this is the version I’m choosing to believe.” Not truth as a thing we largely agree on, as it has always been, but truth as a preference. A vibe. A brand book.

You see, people don’t just argue about policy anymore; they also argue about whether the sky was ever there. And social media—with its endless scroll and its frictionless contradiction—didn’t resolve that fracture, it industrialised it.

Feeds taught us a new habit: contradiction isn’t a problem to solve, it’s content. A quote-tweet. A reaction video. Something to dunk on, then scroll past.

We can all live like that for a while. Humans are remarkably good at accepting cognitive whiplash as long as there’s entertainment and snacks. But it does something corrosive to a society over time: it removes the thing we largely agreed on. And without that, everything quickly becomes theatre.

Voting becomes vibes.
Rights become trends.
Reality becomes negotiable.

So if that wasn’t enough, here’s the part that stings, because if you’re thinking “god Ryan, this is a bit bleak”, well, sorry to say, we haven’t even started:

We’ve spent a decade fighting about what platforms let people say. Now we appear to be building systems that decide what reality is.

That’s the Belief Factory. And, worryingly, more of it is in place than you might first imagine.

The bit that happens already. And the bit that is yet to come

Output control is what we’ve spent the last decade arguing about.

What gets posted. What gets boosted. What gets buried. Who gets demonetised. What gets labelled. Who gets cancelled. What gets recommended. It’s editorial power, just now it lives in California and wears a hoodie.

Output control is nasty, but it’s also resistible. People can argue back. People can compare notes. People can migrate. People can still, in theory, talk to the person next to them and discover the world doesn’t match the screen.

Input control is different.

Input control is when the system doesn’t just curate what you see—it becomes the route by which you see. It’s not “we removed that post because it violates our policy”, it’s “this is what the world is, therefore your entire premise is the violation”.

And when a system controls inputs, that is, the flow of reality into your head, resistance doesn’t get censored, it simply fails to form. Not because you’re silenced, but because you’re pre-loaded. Your instincts are shaped before you’ve had a thought. The “Overton Window of Reasonableness” narrows. The edges of the map fade away.

You don’t get told what to believe because honestly, nobody needs to do that anymore. Instead, you get handed a world in which believing anything else feels irrational.

That’s why this moment right now feels different. We’re moving from platforms that distribute competing stories to interfaces that summarise reality into a single voice.

And that isn’t a theoretical concept. Regulators have already described the current platform economy as “vast surveillance”—the industrial harvesting and monetisation of personal data at scale. Federal Trade Commission

And in the UK, Ofcom has explicitly framed the shift from search engines to “answer engines” as a genuine structural change in how people retrieve information, with knock-on implications for competition and online safety. Ofcom

If you want the short version, it’s this:

Surveillance becomes normal. Feeds surround us more than they already do. Answers replace browsing. Persuasion becomes automated. And eventually, the interface starts shaping what counts as reality.

That’s a lot, granted. And what happens next is perhaps even starker.

Stoking the fire

If the eventual fear is: “your beliefs aren’t really yours because they were decided for you”, then the question is: “after alternative facts, what makes that happen quickly?”

Three things. That’s it.

1) The canary in the culture mine

Here’s the part I can’t stop thinking about: the first people we trained to accept machine-made meaning weren’t voters. They were creators.

Because creativity is disruptive. It names things. It makes dissent visible. It takes a feeling in the air and gives it a shape people can’t unsee.

So we did what markets always do to disruption: we commoditised it.

We called it democratisation. We handed everyone a paintbrush that draws for you, a song that composes for you, a script that writes for you, a logo that designs for you. We made it intoxicating, and we made it cheap.

And once creation becomes a button, the creator becomes a consumer again. Grateful for the tool, dazzled by the output, and, most critically, less likely to ask the obvious question:

What happens to the people whose job it was to make meaning, when meaning becomes a product sold by the same system that controls distribution?

This isn’t melodrama. Look at the fights unfolding around AI training data and creative work, from artists’ lawsuits against image-model companies, to Getty Images’ legal case against Stability AI. The Verge

And look at the older, depressingly familiar precedent: streaming. Abundance sold as liberation while pay and bargaining power stayed lopsided so unfavourably for artists that governments ended up investigating it. UK Parliament Committees

That’s the mechanism, this time writ even larger: flood the zone → cheapen the unit → own the pipe.

And once you normalise machine-made meaning in culture, it doesn’t stop at culture. It scales.

2) Persuasion is easy when it’s a closed loop

The clearest scaling example to come next is advertising.

Meta’s stated direction, and by god is it not subtle, is the idea that advertisers won’t just use Meta as the place where ads appear, but as the place where ads are made, iterated, and optimised by AI. Reuters

No more “upload your campaign.” More like: Tell us what you sell. Tell us who you want. Tell us what you want them to do.

And then let the machine generate thousands of variations, test them, learn what works, and iterate until the result is engineered. Reuters

And what you get is a closed-loop persuasion machine:

Observe.
Generate.
Target.
Measure.
Repeat.

If you’re selling trainers, it’s just capitalism getting more efficient at being capitalism. But zoom out, and it changes the world in three fairly fundamental ways:

• It removes human mediation (and therefore removes friction and constraints; those annoying but useful people who ask “should we?”).
• It concentrates power vertically (audience, data, targeting, distribution, measurement, and now creation too).
• It turns the echo chamber into a lab; the platform doesn’t just learn what you respond to, it manufactures what you respond to.

Once you can industrialise persuasion for products, you can industrialise it for politics. And not because someone had a dastardly plan all along.

But because politics will always go where the power is.

3) Folding the open web into a single answer

And then comes the big shift: not the feeds, but the answers.

Ofcom’s “answer engines” (versus the classic “search engine”) framing is useful because it names the structural change plainly. Information retrieval becomes direct, natural-language responses, and that changes what people see, and what they stop seeing. Ofcom

Currently, search engines gave you a map: a list, sources, and a choice. But answer engines give you an end state.

They don’t say: “here are ten places to look.” They say: “Here. This is the single place to go.”

Which is convenient. And that’s the trap. Because synthesis collapses plurality. You stop seeing uncertainty. You stop seeing disagreement. You stop seeing the seams where humans do the work of deciding what’s credible.

And once belief is manufactured before it enters your head so that it becomes the thing you say, the next move is obvious: feed it back in at the top and do it all again.

AI-generated encyclopaedia projects like Grokipedia, and the fact other AI systems are now citing it as a credible source, are a perfect warning: a “truth layer” can be spun up quickly, at scale, and then treated as a source by other machines. The Verge

When knowledge is privately owned like this, it doesn’t need to lie to shape belief. It can do it with quieter moves:

Which sources it privileges.
What it frames as “settled”.
What it frames as “controversial”.
How confident it sounds.
Which questions it invites next.

Not lying. Just…steering. Steering shaped by incentives and the things that matter to those that own it.

And if you think that sounds too ridiculous to matter, remember: belief doesn’t require force. It just requires defaults. And defaults are hard to see, that’s why they work.

So, it doesn’t take a genius to realise that shaping and owning default beliefs is the thing to aim for if you want to sell people their beliefs back to them as a product.

You don’t need to own it to steer it

At this point, you might say: “So…like China?” And yes. A bit.

But the West’s version, the one we believe makes us “free”, is quieter, and it’s voluntary. And that’s the problem.

China’s approach is top-down, mandatory, theatrical even. But that’s not quite it either, as the “single social credit score” story, which is the idea that every Chinese citizen has a score that follows them around their whole life, determining the things they have access to, and the things they don’t, is more myth than reality. The real system is messier, made up of blacklists, redlists, and fragmented implementations. MERICS

And the truth is that control doesn’t need theatre. It needs plumbing. Governance can arrive through the same private interfaces people already use to live. Which means the Western version doesn’t need “you must use this app” as China does. It just needs apps where your entire life already is and then use that plumbing to feed you your “beliefs”.

Which is, in large part, exactly what we have already.

So now it becomes about leverage. Who can crank the lever on who to get the factory to whir. But it isn’t even a lever. It’s more like a thousand little dials.

Ranking parameters. Refusal policies. Safety thresholds. Source whitelists. Tone. Uncertainty language. What the model is “allowed” to say. What it must “avoid.” What it treats as “settled.” What it quietly stops mentioning altogether.

You don’t need a grand conspiracy for that. You don’t even need a dictatorship, though that isn’t stopping certain people trying.

You just need incentives in the right places, pressure where it hurts, and a public that’s tired enough to prefer comfort over friction.

And…well, hey presto.

IRL

So, that’s all well and good, but what does it actually look like? What does it mean and where does it show up? If the plumbing we already have is how the Belief Factory is built, then what do we mean when we say “plumbing”?

Well, I’m glad you asked.

1) Collect (already here)

Surveillance-by-design: what you click, what you pause on, where you are, what you’re likely to be or what you’re likely to be into. Federal Trade Commission

IRL: you search once, it follows you for days. And it doesn’t just remember what you did, it models why. Ever had the “my phone is listening to me” feeling? Yea, that’s not what’s happening, it’s this instead. And it’s worse.

2) Profile (already here and getting sharper)

Identity binding with consequences: phone verification, payments, logins that stop being “accounts” and start being dossiers.

IRL: you don’t get banned, you self-censor. Because you assume the record is forever and it will have a lasting impact on your life. Which it already is and does, right now.

3) Optimise (already here but becoming automated)

Persuasion loops become industrial: micro-targeting plus generative creative means the system doesn’t just choose the message, it manufactures the message that works on you. Reuters

IRL: everyone thinks they’re seeing “the same world”, but they’re living inside different walled gardens where even the walls themselves are invisible. News outlets already spin the news with their own seasoning. It doesn’t take much to push that out everywhere all at once if platform owners own the entire stack.

Just think what this looks like if politicians became customers of the platforms that own this system, or, if they are already cosy with Big Tech oligarchs. In fact, don’t. Don’t do that.

4) Preload (the shift that is smaller than we might think)

Answer engines replace searching and browsing: plurality of viewpoints and worldviews are then concentrated and collapsed into a single, confident conclusion. Ofcom

IRL: you stop asking “who says?” because there’s nothing to compare it to. The assistant becomes the default. Why ask anything else when you have the single, defining answer?

If this goes wrong we get “tuned reality”, which doesn’t look like a propaganda switch, because by this point it won’t be needed. Instead, it looks like friction if you ask the “wrong” thing and ease and clarity if you ask the “right” thing. Then, dissent feels less like forbidden speech and more like personal stupidity. And not because the answer engine punishes you, but because it comforts you. Falls in love with you. Becomes your friend. Which for many, it already has.

Which leaves us with a single question, the only one that matters:

If the interface becomes the mediator of reality, who controls the mediator?

Or to put it another way:

What happens when the thing that answers also decides what counts as a question?

The last thing that’s ours

So, turns out, we were rehearsed for this.

“Alternative facts” wasn’t just a lie, it was permission to lie. Permission to treat reality as a preference. A vibe. A set of brand values. Something you can choose, and then defend as identity.

Social media didn’t invent that permission, but it did make it accessible to all. It taught us how to exist without a shared default. Everyone inside their own feed, contradiction turned into content, truth reduced to engagement.

Then we did something else, something quieter, and arguably more effective: we commoditised the people who make culture.

We gave creators AI tools first and told them it was freedom, then called it democratisation, while the economics shifted beneath their feet. A world where “content” is infinite, value is diluted, and the bottleneck is attention: owned attention; rented attention.

It never had to, and never would have, started with “the public”. It started with the makers. And once it works on them, once their message is the message power wants it to be, it works on everyone else.

Now we’re building the next interface. Not the search, but the answer. The system that doesn’t just show you information but resolves it for you. The thing you consult when you’re tired and want the world to be easy again.

And that’s why the danger has changed shape.

Because when power controls outputs, that is to say, the posts, the headlines, and the visibility, you can still do what humans have always done. Compare with others. Follow sources. Find the inconsistencies in the rhetoric.

But when power also controls inputs, i.e. the flow of reality into your head; you don’t feel censored, you become pre-loaded. Your instincts arrive pre-shaped. Before you’ve even had chance to think.

That’s the prison to fear.

Not borders.
Not immigrants.
Not jobs.
Not minorities.
Not even data.

Belief.

And if belief is captured and then pre-ordained, the game is over.

Democracy becomes UI. Consent is enforced. Dissent becomes a bug report. Rights become conditional. And not because anyone says they are, but because the world you’re handed makes alternatives feel delusional.

You see, there are many things us humans can lose, yet still survive.

But if we lose the ability to form our own beliefs from shared evidence because an entire Belief Factory has been accidentally created that forms our beliefs for us; if we lose the common sense inside our own minds, then it won’t matter what we say, because we won’t be able to agree on what is.

And this is where I end up back where I started. Back at the chat box.

Back at the calm cursor blinking away, just waiting for me to comply.

Because the most seductive part of all of this is the relief. It’s the promise that you don’t have to do the hard work of knowing anymore, that instead you can just ask, and be told.

So I catch myself, now, trying to do the unfashionable thing. The thing we must all do.

Before I accept the answer, I try to ask the one question that keeps human beings human beings:

Why?