🤖 The Enshittification of ChatGPT Has Begun

🤖 The Enshittification of ChatGPT Has Begun

AI isn't a bubble, but OpenAI's balance sheet sure is looking like one.

Their revenue is growing. Their costs are growing faster. And now they're reaching for ads—something Sam Altman once called 'a sign of desperation.' The prize, if they monetize at the same ARPU as Meta, could be a $57bn ARR business that would see them turn profitable.

But you have to wonder. When the agent that's supposed to act in your interest starts taking money from the companies it recommends? That's a trust problem. For users. For advertisers. And trust is the only thing between ChatGPT and a glorified search engine.

Open AI’s Costs are Growing Linearly with Revenue

OpenAI pulled in $20B annualized revenue last year. Sounds great until you see the math.

The tweet cites OpenAI's CFO Sarah Friar announcing $20B+ ARR. But buried in the announcement is the uncomfortable truth: compute grew ~3x year over year, and revenue followed the same curve.

That's linear scaling. Revenue goes up 3x, costs go up 3x. There's no operating leverage here.

And it gets worse.

Let me translate this thread into plain English:

Zero cost reduction in three years. Every 'AI compute is getting cheaper' narrative just hit a brick wall.

Some analysts now predict OpenAI run out of money by mid 2028. And the Musk lawsuit complicates any IPO timeline.

They need a new revenue engine. Fast. Enter stage left: Advertising.

This analysis by Signull suggests that if OpenAI could monetize at Meta’s average revenue per user (ARPU), they’d get to $57bn revenue per year. That would certainly close the gap!

How do you say no to a revenue line like ads? You don’t.

The enshittification of ChatGPT begins

Ads are coming to ChatGPT. 

Free and Go users in the U.S. will see ads at the bottom of responses starting in weeks. Thankfully, Plus, Pro, and Enterprise are still ad-free.

Here's Sam Altman, in happier times, explaining why ads would be a bad idea:

'If we ever did that, it would be a sign we're desperate.'

Well.

Cory Doctorow has a term for what happens next: enshittification. It's the lifecycle of platforms that prioritize revenue extraction over user value. First they're good to users to attract them. Then they abuse users to benefit business customers (advertisers). Then they abuse those business customers to claw back value for shareholders. 

Remember when Facebook swore they'd never do ads? Remember when you could find actual search results on Google's first page? How’d that go.

OpenAI says they'll never sell user data and will block ads on politics, health, and mental health. Good guardrails—for now. But 'clearly labeled' ads at the bottom of answers still condition users to scroll past noise. And guardrails have a way of eroding when the cash crunch gets worse.

The problem? The other LLMs don't have to follow. Google explicitly said they won't put ads in Gemini. Anthropic hasn't hinted at it. If competitors stay ad-free while ChatGPT doesn't, why stick around?

Especially when the market is already shifting.

ChatGPT is losing market share

12 months ago: ChatGPT had 86.7% market share. Gemini had 5.7%.

Today: ChatGPT has 64.5%. Gemini has 21.5%.

That's a 22-point swing in a year. ChatGPT doesn't have an engagement problem—800 million weekly actives is massive. It has a monetization problem. 95% of users aren't paying.

The bet is that ads fix monetization without killing engagement. If that bet fails, we'll learn exactly where users draw the line between 'helpful AI' and 'monetized feed.'

Will advertisers buy this?

You can imagine some tempting ad products:

Sponsored Agency: Ads target specific purchase-intent prompts instead of keywords. User asks 'Plan a 3-day trip to Paris and book flights,' and an airline bids for a sponsored suggestion the agent presents. ('I see great rates on Expedia—should I check those first?')

Affiliate Model on Steroids: The agent becomes the ultimate affiliate marketer. User asks for running shoe recommendations. The agent returns a curated list. Purchases on partner sites earn ChatGPT commissions on top of ad placement fees.

But will advertisers trust an environment where OpenAI claims the model 'won't be influenced' by ads? That's a bold claim in a world where SEO and SEM have trained everyone to game ranked results.

The contrast with Google and Meta matters: they're known quantities to advertisers, they're not putting ads in their LLMs (Google said so explicitly), and they are using foundation models to improve ad performance in existing channels.

OpenAI has a cold-start problem. No track record. No measurement infrastructure. If it converts and delivers sales, advertisers will come. But that's a big 'if.'

And this is all happening at the same time OpenAI is dabbling in agentic commerce.

The conflict of interest problem

Agentic commerce is where AI acts autonomously on behalf of the user to complete transactions. OpenAI is already building this through partnerships with Etsy and Walmart, letting ChatGPT execute purchase flows.

If ChatGPT successfully integrates ads and agentic commerce, it becomes a hybrid: part ad-supported information tool, part autonomous shopping agent.

For measurement, this is the holy grail. Instead of tracking a click, ChatGPT tracks the entire transaction completed by the agent. Perfect attribution data justifies higher ad spend.

For users, this is where the trust breaks.

Can you trust that the agent is acting in your best interest—finding the best value, highest quality—when its financial incentives (ad revenue, commissions) are tied to specific vendors?

The agent that's supposed to work for you is now being paid by someone else. That's the textbook definition of a conflict of interest.

Doctorow's enshittification pattern predicts where this goes. The platform starts optimizing for revenue extraction. The user experience degrades. Trust erodes. Users leave. The platform dies—or becomes a zombie version of its former self.

ChatGPT isn't there yet. But reaching for ads while building agentic commerce? That's the first step down the path.

The bigger picture

Ads are a gravitational black hole on the internet. They're so wildly lucrative that it's hard to imagine any big tech company ignoring them forever.

OpenAI doing ads now isn't surprising. They have a revenue hole to fill and a cash runway measured in months, not years.

But here's what matters for everyone building in this space:

The agent economy runs on trust. If the agent that books your flights, manages your calendar, and handles your shopping is taking money from the merchants it recommends, the whole premise collapses. We're back to Google circa 2024—a firehose of sponsored results masquerading as helpful answers.

OpenAI is the first to test whether ads and agents can coexist.

They won't be the last.

Not by a long shot.

ST.