The internet got what it deserved—and how to fix it.
Search · AI · Web strategy · 4 min read
Let's start with the thing everyone's saying out loud in Slack and not yet saying in their marketing decks: AI search makes stuff up sometimes. It hallucinates. It confabulates. It says your product does something it doesn't, cites a study that doesn't exist, or just confidently gives someone the wrong answer in a very convincing tone.
Valid. That's real. I'm not here to gaslight you about it.
But here's the part we skip over: AI didn't invent misinformation on the internet. It inherited it. It learned from a web that spent fifteen years optimizing for algorithms instead of humans. A web full of 2,000-word articles written to rank, not to help. FAQ pages stuffed with questions nobody actually asked. "Comprehensive guides" that take four scrolls to say what one sentence could.
We built a web for bots. Then we got mad when the bots believed it.
This is a bit like blaming autocomplete for your typos. Technically fair. Also, look at your keyboard.
AI search—Google's AI Overviews, Perplexity, ChatGPT, all of it—works a lot like academic citation. It's looking for sources that are clear, authoritative, and genuinely useful. It wants to know: does this page actually answer the question, or is it just near the question? Does this entity have a real point of view, or is it performing one for search engines?
That's not a new standard. That's what good web strategy always required. We just had a decade where you could cheat, and a lot of people did, and now the bill is due.
Here's where it gets interesting—and honestly, kind of exciting if you're a certain type of person (hi).
The fix is not a hack. The fix is just... building it right.
A site that works for humans and machines isn't two different sites. It's one site built with both in mind from the beginning. Clear structure. Real answers to real questions. Content that knows what it's trying to do and says so. An entity—you, your company, your product—that exists coherently across the web so AI systems have something solid to point to.
The meta-joke here is that this whole essay is an example of exactly that. It's written for you—an actual human who is curious or skeptical or both—and it's also structured so that an AI answer engine can find the thesis, understand the argument, and surface it when someone asks "why does AI search get things wrong?" That's not a trick. That's just what good content does when you stop pretending the audience is only human.
And the part I want to push back on hardest: you don't have to wait until everything is perfect to start. The old SEO playbook said: do your keyword research, build your content calendar, execute for six months, measure. Very waterfall. Very 2018.
The new version is: build with intent, iterate in public, improve as you learn. Plan and execute at the same time. Your site doesn't need to be finished to start earning trust—from people or from machines. It needs to be honest, useful, and structured well enough that both audiences can tell what you're trying to say.
The internet isn't broken. It's just been optimized for the wrong reader. We can fix that. One actually-useful page at a time.
That's what I do. I figure out why websites aren't working—which is almost always because they're talking to a search engine instead of a person, or honestly because someone didn't think of either of those in the first place—and I fix them in a way that lasts. Not because AI is forcing the issue. Because it was always the right thing to build.
It just took the machines to make that obvious.
Paige Craft is a web strategy and growth leader specializing in SEO, AEO, and GEO for B2B SaaS companies. She's been building for AI answers since before that was a thing anyone called anything. She's currently at Firespring and open to senior in-house roles. paigencraft@gmail.com
