Yoav Tchelet

Spent years building attribution models. Most of it was theater. Someone catches your billboard at lunch, scrolls past your post that night. Two days later they click a Google ad and convert. The dashboard credits Google for the whole thing. The models are a directional read. They aren't truth. The hours you'd waste asking them to settle arguments they can't settle are better spent on the brand work that makes someone search for you unprompted.

The Burger King Mandalorian campaign is a moulded helmet. Your Whopper ships inside cardboard shaped like Mando's skull, in 70 markets ahead of the May 22 film. Meanwhile the trade press is debating whether the right Meta creative variant count is 200 a month or 2,000. One of these is going home with people.

The volume play used to work in SEO. It doesn't here. Models reading the web are picking up the same thing readers are. This person has nothing to say. They just have a tool that says things.

So you ship a hundred GPT-default pieces a quarter and feel productive. Your domain gets quietly downgraded by the systems your buyers ask first. Your competitor with twelve weird specific posts shows up in the answer instead.

Slop isn't a taste problem. Or it is, but that's not the part that should worry you.

?

Holly Humberstone’s “Beauty Pageant” is a catchy song… It only works so well because it’s lifted straight from The Beatles “Hey Jude”

If your company's brand values include "innovative" and "customer-centric," congratulations. You've described every company on earth. And therefore none of them. Brand is the opinion you hold that makes competitors uncomfortable. If nobody disagrees with your positioning, you don't have any.

Ahrefs studied 150M keywords. 45.7% of Google searches by volume are branded. Separate study of 75K brands: branded web mentions correlated with AI Overview visibility at 0.664. Backlinks sat at 0.10. Your brand work shows up in search data before it shows up anywhere else. If you’re running brand and SEO as separate teams with separate KPIs, you’re measuring the same thing twice in two rooms.

@CyrusShepard analyzed 400+ sites to find what separates Google's winners from losers in 2026. The finding that stuck with me: sites with 4+ of the 5 core features had a 68% win rate. Sites with zero? 13.5%. The 5 features: → Sells a product or service → Lets users complete a task → Owns proprietary assets → Tight topical focus → Strong brand The uncomfortable part: first-hand experience and UGC didn't correlate. Not because Google ignores them. Because they're table stakes now. Content alone doesn't win anymore. You need to be the destination, not the explainer. Full research: signal.zyppy.com/p/winning-goog…

"Replaced my marketing team with AI" vs "AI will never replace humans." Both misreading the data. Marketing roles up 14% YoY. AI adoption 91%. Only 6% fully embedded. A Stanford study found early-career sales and marketing roles down ~20%. The effect shrinks with seniority. A mid-market marketing team needs 1,100+ decisions per week. They have capacity for maybe 300. AI handles that gap and it should. But the 300 that actually matter require positioning judgment, competitive context, and calls made with incomplete information. You can't wire that into a workflow because the context shifts weekly and depends on signals that aren't in any system. The teams struggling now aren't ignoring AI. They automated everything without knowing which decisions were worth protecting.

There's a type of competitor that never leads. They watch. They wait. Then they produce a slightly worse version of whatever you did last month. It'd be flattering if it weren't so boring. Original thinking leaves fingerprints. Copies don't. Your audience can tell even when you think they can't.

Define your terms. Lead with your core message in the first two lines. Use consistent category language. Post frequently enough that the system has something to find. The vector math is moving in your favor if you do the work.

The infrastructure underneath AI search keeps getting better at rewarding specificity and punishing vagueness. Two years ago you could get away with approximate positioning. The retrieval math wasn't precise enough to care. That window is closing.

@semrush analyzed 89K LinkedIn URLs cited by ChatGPT, Perplexity, and Google AI Mode. A Google Research paper (TurboQuant) explains the vector math underneath those citations. I read both. The implications for your content strategy are worth 2 minutes. 🧵