What celebrity art tells you about getting found through AI search
A new study from geoSurge ran nearly 4,000 AI responses across 66 U.S. buyer questions in nine industries, from travel to business software to fitness. It tracked 13,281 fan-out searches the models ran to answer them. The finding: models picked a brand they already knew 3.2 times more often than one they didn’t, 55.7% of the time versus 17.4%. Across brand-specific searches, 63% of the answers came from just five brands per model: whichever five that model already knew best.
Seth Godin posted the same week about celebrity art, and without meaning to, he explained the study. His point: an object doesn’t need to be well-crafted or important to be prized, just widely known, because “celebrity only happens after the network has kicked in.” Swap “collector” for “AI model” and you’ve basically got geoSurge’s finding in one line.
Most AEO advice right now is plumbing: structured data, machine-readable pricing, an ungated docs page so the crawler has something clean to read. Optimize for the model that’s already decided to look at you. The geoSurge numbers point earlier: whether you’re one of the five brands a model already trusts enough to retrieve. No schema tag fixes that. The real lever is how many places on the internet already talk about you: press, review sites, comparison posts you didn’t write, other newsletters (this one included). Familiarity gets built the boring way, same as always, except the audience now includes a model deciding who gets fan-out searched next quarter.
If your GEO/AEO plan is all technical readiness and zero brand-building, you’re optimizing for a callback that only comes after you’re already famous enough to get one.
Worth Your Time
- OpenAI is testing ads that skip your website and open an AI agent instead (Search Engine Land): The format drops users into a ChatGPT conversation with a business agent instead of a landing page, live for a limited group of advertisers. Early read from the field isn’t rosy: one marketer testing OpenAI Ads reported a $47 CPM on a broad campaign and is pulling the plug. Fund a small test, not your Q4 budget.
- This CPO regrets that product management exists (Lenny’s Newsletter): Whatnot’s Tom Verrilli argues fewer, senior PMs doing real IC work beat any ratio-driven org chart: headcount as a proxy for progress instead of a lever for it. If your PM-to-engineer ratio is the number you report up, that’s the tell he’s talking about.
- Google Analytics adds campaign diagnostics for missing aggregate identifiers (Search Engine Land): A new alert flags when GBRAID and gad_ parameters are missing from campaign URLs, the gaps privacy-mode measurement quietly eats without ever throwing an error. Check this before you trust next quarter’s attribution report, not after a channel mysteriously stops converting.
- Vibe Coding for Marketers: A Beginner’s Guide (Ahrefs): Canva gave 5,000 employees a full week to explore AI tools, and most still couldn’t figure out how to start, proof the bottleneck was never access to the tools.
- Why every SEO team now needs a social topical map (Search Engine Land): One YouTube channel pulled 203,000 Google Search impressions in 11 days once someone finally measured it, visibility that had been running the whole time with nobody watching.
- AI is building your digital experience. It’s also making it less accessible. (Search Engine Land, AudioEye): 95.9% of the top million homepages have detectable accessibility failures, and AI tools trained on that web reproduce the same six mistakes at scale. Accessibility lawsuits hit 26,253 in 2025 with ecommerce carrying 78% of them, so this is a legal budget line now, not a nice-to-have.
Where does your brand show up on the internet when you’re not the one posting it?
Wondering where your growth engine is leaking? The Growth Gap Diagnostic is where I’d start. Or just hit reply and tell me what you’re stuck on.
P.S. Curious how AI search engines see your brand? That’s what Smird is for.