We’ve seen this movie before. First there was the web. Then there were search engines. Then an entire industry sprung up to game those search engines, and we called it SEO. The internet got worse.

Now we’re doing it again, except faster and dumber.

This week brought two data points that fit together perfectly. First, a new study found that roughly a third of web pages published since ChatGPT’s launch show signs of AI authorship. Second, ChatGPT’s search feature now uses the site: operator at scale, automatically filtering searches to specific domains based on what it thinks you want.

And where there’s a new algorithmic gatekeeper, there’s already a new optimization industry. Meet GEO: Generative Engine Optimization. It’s SEO for the AI era, and companies like Promptwatch are already selling it hard.

The pitch is straightforward. Just like you once needed to rank high in Google to get traffic, you now need to show up in ChatGPT’s answers, Claude’s responses, and Gemini’s summaries. Promptwatch offers “automation to track responses to prompts across end-user chat products” and consulting to help your content surface in AI outputs. It’s early days, but the playbook is forming.

Here’s what makes this particularly absurd: we’re optimizing AI-generated content to be consumed by AI systems that will summarize it for humans. It’s a snake eating its own tail, except the snake is made of garbage and we’re all paying the hosting costs.

The feedback loop from hell

Think about what’s actually happening here. A company uses ChatGPT to generate blog posts about, say, “best wireless routers 2026.” Those posts are written specifically to rank well (sorry, to “surface well”) in AI search results. ChatGPT then ingests that content when someone asks “what’s a good wireless router?” and regurgitates a summary based partly on content that ChatGPT itself created.

We already saw how this corrupted traditional search. SEO turned huge swaths of the web into keyword-stuffed spam designed to trick algorithms rather than inform humans. “How to boil an egg” became a 2,000-word journey through someone’s childhood memories of their grandmother’s kitchen before grudgingly revealing that yes, you put the egg in boiling water.

At least with traditional SEO, there was a human at the end of the chain reading the garbage. Now we’ve got AI writing for AI, with humans as a distant afterthought.

The economics make it inevitable

I wish I could say this won’t catch on, but the incentives are too strong. If ChatGPT is routing search queries using the site: operator to privilege certain domains, you’d be crazy not to optimize for that. If a meaningful chunk of your traffic comes from people asking AI assistants for recommendations, you need to make sure your product shows up in those answers.

Google’s response has been telling. They just announced a new feature letting publishers add a “follow” button so readers can make them a “preferred source” across Search, Discover, and Google News. It’s a defensive move, trying to let publishers build direct relationships before AI search cuts them out entirely. It probably won’t work, but points for trying.

The real problem isn’t that companies are optimizing for AI search. It’s that AI search is fundamentally built on ingesting and repackaging other people’s content without reliably sending traffic back. That creates an incentive to produce the cheapest possible content that might get picked up, which means AI slop. Lots of it.

We’re speedrunning the same mistakes

With traditional SEO, it took years for the ecosystem to fully rot. Google launched in 1998. By the mid-2000s, SEO was an established industry. By the 2010s, content farms and algorithm gaming had degraded search quality enough that Google was running regular updates (Panda, Penguin, etc.) trying to fight back.

We’re doing the same thing with AI search, except compressed into maybe 18 months. A third of new web pages are already AI-generated. GEO is already being sold as a service. The feedback loops are already forming.

And here’s the kicker: it’s going to work, at least for a while. Some companies will successfully optimize their AI-generated content to show up in AI search results. They’ll get traffic, make money, and encourage everyone else to do the same thing. The web will get worse, AI training data will get worse, and the outputs will get worse, but that’s a tomorrow problem.

What actually needs to happen

The fix isn’t complicated in theory. AI companies could commit to citing sources clearly and sending traffic back to publishers. They could develop ways to compensate or partner with content creators instead of just hoovering up their work. They could build systems that privilege original reporting and primary sources over AI-generated summaries of summaries.

But that would require leaving money on the table in the short term, and we know how well that goes.

So instead we’re getting GEO. We’re getting a new layer of optimization consultants. We’re getting an internet where AI writes content for AI to read, and somehow we’re all pretending this is progress.

The smarter play for anyone who actually creates original content is probably to opt out entirely. Robots.txt already blocks most AI crawlers, though enforcing it is another question. Some publishers are going to decide that being summarized without compensation isn’t worth the “exposure.”

But most won’t make that choice. The fear of being invisible in AI search will override the knowledge that the whole system is extractive. So they’ll hire GEO consultants, generate AI content optimized for AI consumption, and we’ll all pretend the web isn’t slowly drowning in synthetic noise.

I’ve seen this movie before. I know how it ends. But apparently we’re all buying tickets again anyway.