So here we are.
This week, OpenAI announced it will begin testing ads inside ChatGPT. Not everywhere. Not for paid users. Lower tiers, at least for now. The company went out of its way to stress guardrails: ads will be clearly labeled, conversations won’t be sold, and personal data won’t be handed over to advertisers.
Measured. Responsible. Adult.
And still, if your first reaction was a quiet sense of disappointment rather than outrage, you’re probably closer to the truth than either the cheerleaders or the doomers.
Because this isn’t about ads.
It’s about gravity.
The Mirror Turns
A few days ago, I wrote that blaming the internet for what it has become is like blaming the mirror because you don’t like how you look. The mirror didn’t do this. It reflected incentives, business models, and human behavior.
AI just walked in front of the same mirror.
For a brief moment, AI felt different. Cleaner. More like a tool than a platform. You asked questions, got answers, and moved on. No feeds. No engagement traps. No dopamine loops. Just utility.
That feeling was real — and temporary.
What we’re seeing now isn’t betrayal. It’s transition.
What OpenAI Actually Said
Before we layer analysis on top, it’s worth grounding this discussion in what OpenAI actually announced.
In its post, “Testing ads in ChatGPT,” OpenAI laid out a cautious experiment. Ads will initially appear only in free and lower-tier plans. They’ll be clearly labeled. Paid plans will remain ad-free. Conversations themselves won’t be sold or directly used for ad targeting.
This wasn’t a sneaky rollout. No pop-ups mid-response. No sponsored answers baked into model output. Compared to how advertising arrived on earlier platforms, this was downright restrained.
But restraint doesn’t change trajectory.
This Was the Worst-Kept Secret in AI
Let’s dispense with the theatrics.
Anyone who thought ChatGPT would remain ad-free forever was either naïve or willfully ignoring the economics. Running frontier AI models costs a staggering amount of money. Compute. Talent. Training runs. Data centers. Power. Cooling. Water. Networking. Marketing. And yes, very expensive GPUs from NVIDIA.
The red ink isn’t theoretical. It’s very real.
Even Anthropic couldn’t resist poking fun at this reality in its recent Super Bowl ad, joking about the idea of advertising creeping into AI assistants.
Funny because it landed. And it landed because everyone knows what’s coming.
Doctorow Was Right (Again)
This is where Cory Doctorow’s concept of enshitification, sharpened by thinkers like Tim Wu, stops being academic theory and starts looking like a timeline.
Platforms follow a familiar arc:
They start by serving users.
Then they serve businesses.
Eventually, they extract value from both.
Not because founders suddenly lose their ethics. Because incentives harden as scale increases. What begins as optional becomes mandatory. What begins as experimentation becomes policy.
We’ve seen this movie before with Google, Meta, and Twitter. Different eras. Same gravity.
AI didn’t escape that gravity. It postponed it.
Ads Aren’t the Point. Extraction Is.
Let’s be clear about something uncomfortable.
Nobody seriously believes OpenAI is going to start selling your prompts to the highest bidder. That would be reckless and radioactive. The real concern isn’t raw data resale.
It’s extraction by inference.
Who you are. What you care about. What persuades you. What frustrates you. What beliefs can be nudged, shaped, or amplified if someone pays enough.
That’s how enshitification actually works. Not through obvious villainy, but through subtle shifts in incentives and optimization goals.
Ads are just the visible edge of that wedge.
Why AI Feels More Fragile Than the Internet
There’s a reason this moment feels heavier than banner ads or promoted tweets ever did.
You don’t scroll AI. You think with it.
AI isn’t just a distribution channel; it’s a cognitive partner. It helps you reason, write, plan, and decide. That intimacy raises the stakes dramatically. When monetization enters that space, trust becomes far more fragile.
This is why the reaction to ads in ChatGPT isn’t just annoyance. It’s grief. People sensed they were losing something that felt different.
And they’re not wrong.
The Anthropic Contrast — And the Reality Check
Anthropic has said it won’t run ads. Its constitution and governance model are explicitly designed to resist certain forms of extraction. I hope that holds. Truly.
But let’s be honest: this isn’t a morality play. It’s a bet.
A bet that enough customers will pay for restraint.
A bet that values can outlast invoices.
A bet that sustainable revenue doesn’t require surveillance.
That bet only works if users and enterprises back it with actual money, not applause.
Someone still has to pay for the chips, the power, the water, and the data centers.
Ideals don’t settle invoices.
The Real Shock Is Still Ahead
Here’s what I think will actually rattle the AI market — and it’s not ads.
It’s the moment when AI moves decisively from “go ahead, the first one’s free” to “this is what it really costs.”
When free tiers shrink.
When usage-based pricing bites.
When enterprises see the true bill.
When consumers realize intelligence at scale isn’t cheap.
At that point, a few ads and some light extraction may start to feel tolerable. That’s how enshitification wins — not through coercion, but through resignation.
Don’t Blame the Mirror
So no, this isn’t about OpenAI losing its way.
And it’s not about AI becoming evil.
It’s about what happens when powerful tools collide with economic reality. The mirror didn’t change. It just turned.
The question now isn’t whether AI will follow the internet’s path. It’s whether we’ll recognize the reflection early enough to make different choices — as builders, leaders, buyers, and users.
Because mirrors don’t force outcomes.
They give us a chance to decide what we do next.
Cue the Music
“And this is the age of extraction
Where the signal gets bent into distraction
Born and raised on free completion
Now we pay for enshitification…”
Yeah. That song’s gonna get a lot of airtime.
Because the mirror’s still there.
And AI is starting to look an awful lot like the internet staring back at us.
