Younger SEOs I consult with keep forwarding me ChatGPT’s take on their sites and asking what I think. Usually I agree with most of it, and then I hit the part where it flags a 25% stat on one page against a 27% stat on another, or a 4.8 rating that’s really a 4.7. No customer has ever walked away over that. It’s artificial intelligence, not artificial wisdom. I said so on Unscripted SEO to Josh Shankowsky of Snap SEO, an Edmonton SEO agency, and he gave me a better name for the problem.
They pull from the same pool
Josh calls AI tools consensus engines. They report what most people are saying, and if that consensus is two years old, so is the advice. My version: the models are probabilistically pulling from the same pool. Give different chatbots a creative-writing prompt built to produce wildly different human answers, and they drift to the same anecdotes. You and I draw from different pools, and that difference is the only thing that makes content worth ranking.
So when a client asks me to “just have AI write it,” here’s what I tell them. The AI can help with the draft. The value has to come from you: your founder story, your customers’ actual questions, the thing every buyer asks that your site never answers. Josh’s agency runs a full “cultural onboarding” to capture exactly that before writing a word.
Attribution madness
The other thing I keep coming back to is how little Search Console tells you. For most of my clients, if Search Console reports 100 clicks, you get query detail for maybe 40 to 60 of them, and often far less. That’s less than a coin flip, and people make big decisions off average position for a third of their queries.
What I do instead is triangulate. I connect the Search Console, GA4 and Microsoft Clarity APIs to Claude, match a query’s timestamp to a session, and let Clarity tell me which sessions were bots. Google’s Direct bucket carries a big share of what Clarity identifies as bot traffic. I wrote up how far the numbers drift in a Search Console vs Clarity check across 4 sites.
Josh pushed me to add a fourth source: Bing Webmaster Tools. It has always given more detail than Search Console, and it shows Copilot searches. He also uses Bing results to find the directories and best-of lists ChatGPT cites, then gets clients listed there with a review. I’m adding both to my process.
“It depends” isn’t a dodge
I’m tired of saying “it depends,” and I can’t stop. A client who put review schema on every page probably tripped a penalty, but I can’t swear that’s the only cause when they might also have redirected a dozen domains or been hacked. Calling one fix “the problem” would be malpractice. Google runs constant tests, and some of its systems exist to confuse anyone reading ranking moves. Even Googlers can’t always explain a result.
Josh’s client version is the honest one: “You’re going to hate me, but it depends.” Then he explains why, and he tells them to hold course.
Hear the whole conversation, including Josh’s Google Discover notes and his nonprofit sponsorship play, in the Josh Shankowsky episode of Unscripted SEO. If you want this kind of triangulation on your own site, here’s how I work.