When Content Becomes Infinite: Alex Collmer on Creative Intelligence in the AI Advertising Era

Full analysis

Alex Collmer, founder of Vidmob, makes one clean argument: the AI models everyone uses (Claude, ChatGPT, Midjourney) are turning into commodities, so the advantage moves to the proprietary data you feed them. For Vidmob that data is "creative data," the structured record of which ad decisions (logo spot, pacing, word count) drove which results. The decision this teases for anyone buying AI for marketing: do you let Meta and Google run your creative and learning inside their closed systems, or do you keep that learning somewhere you own it? That is easy to undo in the small (you can switch tools), hard to undo in the large (once a platform has trained on years of your performance patterns, you can't claw that back). Nothing sets a hard deadline here. This is a positioning pitch, not a shot clock.

The Skeptic

Every claim here points back to something Vidmob sells. Creative is "50 to 70% of results," measurement tools give it "zero weight," walled gardens are quietly harvesting your edge. Convenient, since Vidmob's whole business is the fix. The 50-to-70 figure is a heuristic Nielsen and others cite with wildly different methods, not a constant. And the "Meta trains on your patterns to help your competitor" line is unproven. It's plausible because Meta's data policies are murky, but nobody has shown cross-advertiser pattern transfer happens the way Collmer implies. He's selling insurance against a fire nobody has confirmed is burning. The commodity thesis itself is fine. The moat conveniently happens to be Vidmob-shaped.

The Researcher

Strip the sales pitch and one observation holds up: creative quality is genuinely absent from most attribution math. MMM (models that split budget across channels) and MTA (methods that assign credit for a conversion) weight time, geo, weather, bid dynamics, and treat the ad itself as a constant. That's a real gap. But "we tag logo placement and pacing with computer vision and correlate to outcomes" is correlation on a small, messy sample per brand. Collmer's own point undercuts him: if LLMs predict the most likely next thing, and everyone optimizes on the same yesterday's-winner signals, you converge on sameness. His mobile-gaming example (2,000 to 3,000 near-identical variations a month) is the proof. Creative data measured this way could accelerate the convergence it claims to escape.

The Builder

The concrete thing shipped in May 2025 is Vidmob360: creative performance data exposed through an API and MCP (the standard that lets an AI assistant pull from an outside data source). You can ask Claude "which assets decayed more than 20% in four days" and get an answer, a brief, a fatigue alert. That's a real, useful pattern, and it's the direction everything is going. What I'd actually check on Tuesday: how fresh is the data, how good are the connectors (they claim ~12 covering 90%+ of North American buys), and does the predictive score survive being wrong a few times before my team stops trusting it. The MCP piece is the durable idea. The rest is a dashboard with a chat box.

The Enterprise Buyer

The one line that should make a CMO sit up isn't about creative at all. It's the contract question. As Meta's Advantage+ and Google's PMax swallow more of the creative-and-media decision, you need written clarity on whether your performance patterns get used across other advertisers. Collmer sits on Meta's Global Creative Council and still raises it, which tells you the answer isn't obvious even to insiders. That's a real procurement action: ask the data-use question, get it in the contract, don't accept "our policies are on the website." The "own vs. rent your intelligence" framing is marketing, but the governance gap under it is worth an hour with your legal team.

Where they part ways

The Researcher and the Skeptic agree the commodity thesis is right and the moat claim is self-serving, but they split on whether creative data even works: the Skeptic says it's unproven insurance, the Researcher says it might actively speed up the sameness problem. The Builder and the Enterprise Buyer split on what's worth paying for: the Builder wants the MCP plumbing and doesn't much care about the sovereignty story, while the Buyer thinks the plumbing is easy to copy and the only durable value is forcing the data-use question with Meta and Google.

What it hinges on

Two beliefs. First, is creative quality measurable well enough per brand to beat the sameness it risks creating? Unproven, and Collmer's own gaming example cuts against it. Second, do walled gardens actually transfer learning across advertisers? Nobody has shown it. The council leans this way: the commodity-models argument is correct and already consensus. Owning your data matters. But "creative data" as Vidmob defines it is a bet, not a fact, and the sovereignty pitch is doing double duty as a sales motion. Before buying anything, run one test: take a brand's real creative data through the API and see if the predictive score beats a simple baseline over a full quarter. And put the cross-advertiser data-use question in writing with Meta and Google now, since that costs nothing and closes real exposure.

Prediction: Neither Meta nor Google will add a contract term, by their next major advertiser terms update, that promises a brand's creative-performance patterns won't inform other advertisers' automated results in Advantage+ or PMax.

Confidence: Medium. The cross-advertiser learning is the product, so they won't fence it off.

Why: Advantage+ and PMax work by pooling signals across advertisers to find what performs, so a promise not to use one brand's patterns for another would weaken the exact machinery that makes those tools convert. Collmer, who sits on Meta's own Global Creative Council, still can't point to such a guarantee, which suggests it doesn't exist and isn't coming. Both platforms compete on automated performance, and voluntarily narrowing what their models can learn from cuts against that. The opposite outcome, a public per-advertiser data-isolation pledge, would mean Meta or Google giving up a measurable performance edge with no regulator forcing them to, and nothing here shows that pressure.

Revisit by 2027-03-27: We're right if neither Meta's nor Google's advertiser terms add a clause isolating one advertiser's creative-performance learnings from others' automated campaigns. We're wrong if either publishes such a clause.