You know that sinking feeling. You pour hours into a blog post, a guide, a deep dive. You hit publish, and then… silence. Not because nobody read it. But because a machine hoovered it up, chewed it into training data, and spat out a glossy answer that never mentions your name. No link. No credit. No traffic.
Now here’s the kicker: a new industry is cashing in on that frustration. They call it ‘AI SEO.’ They promise to get your site cited in ChatGPT, Claude, Gemini. The pitch sounds great: ‘Stop being invisible. Get your content into the AI’s answers.’
But here’s the truth they don’t want you to hear: paying for AI citations is like paying a magician to make your wallet disappear. The architecture of LLMs makes it impossible to reliably attribute a response back to a specific crawled source.
Let’s look at the numbers. A recent study of 100,000 websites found that only 8.9% block AI crawlers. Great, right? The rest are open for business. But here’s the punchline: 94.8% of those sites are never cited in any AI answer. The vast majority of web content fuels AI training without ever granting visibility or credit to its creator.
I saw this firsthand. A friend runs a niche technical blog—deep, original work. He signed up for an AI visibility service that promised to ‘optimize his content for AI citations.’ Six months later, zero citations. Zero referral traffic from any AI tool. But his site was still being crawled. His content was still being used. Just never credited.
Why? Because it’s architecturally impossible for an LLM to associate a link or citation that it crawled with a response that comes out the other end. Every link they’re giving you to support their statements is tacked on because it may vaguely match the tokens it just generated. It’s perfectly common to fabricate citations entirely. The system doesn’t remember where it learned something—it just predicts the next word.
So the emerging ‘AI SEO’ industry is selling a phantom metric. Paying to be included in AI training data is essentially paying to be plagiarized without attribution.
This is not a bug. It’s a feature of the architecture. The models don’t have a page rank mechanism. They don’t have a citation database. They have a giant soup of words, and they generate answers by blending—not by citing. Even when a model does append a link, it’s often hallucinated. A study from earlier this year found that 78% of citations generated by GPT-4 were either wrong or non-existent.
And yet, webmasters are shifting from blocking AI crawlers to actively seeking AI citations. The irony is staggering. A few years ago, we were up in arms about being scraped. Now we’re paying to be scraped better.
Let’s be clear: I’m not saying you should block all crawlers. Some AI tools (like Perplexity) have started to experiment with more explicit attribution. But the idea that you can pay a service to ‘get into the training data’ and then see a return in traffic or citations is a fantasy. The data doesn’t support it. The architecture doesn’t support it. The only thing that supports it is the desperation of creators who feel erased.
So here’s my position: stop paying for AI visibility. It’s not a strategy. It’s a tax on your fear of being left behind.
Instead, focus on what you can control. Write for humans. Build an audience that subscribes, not just browses. Use your own channels—newsletters, RSS, communities. The web is not dead. But the idea that AI will be your distribution channel is a lie, and it’s a lie that’s costing you money and time.
The day we stop pretending that AI citations are a meaningful metric is the day we can start building something that actually works. Until then, the machines will keep taking. And the salesmen will keep selling.
FAQ
Q: Is it really impossible for AI to cite sources properly?
A: In current LLM architecture, yes. The model doesn't store a lookup table of where it learned each fact. It generates text by predicting the next token based on statistical patterns. Adding a citation is a separate post-processing step that often hallucinates. Some newer models are trying to ground answers in retrieved documents, but that's still not the same as attribution from training data.
Q: So should I block all AI crawlers?
A: Not necessarily. Blocking crawlers can prevent your content from being used in training, but it also removes any chance of being cited by AI tools that do honor robots.txt. More importantly, most AI tools don't cite anyway. The real question is: do you want your content to be part of the public commons that AI learns from? If not, block. If you're okay with it, don't expect anything in return.
Q: What's the contrarian take on paying for AI SEO?
A: The contrarian view is that early adopters might gain a temporary advantage if AI companies start prioritizing certain sources for attribution. But the evidence so far is thin. The industry is selling a solution to a problem that doesn't exist yet—and may never exist in the way they promise. It's 2025's version of paying for a search engine submission service in 1998.