Every marketing team is being told to point AI at its market. It is a reasonable instruction, and almost nobody has measured what happens when you follow it. So we did, against ourselves, and we are publishing the parts we lost.
The test
We took two real brands and asked one question of each, on Meta, LinkedIn and Google: what creative angles are actually winning in this market right now, who is running them, and what's the evidence?
Three contenders answered. MessCube. A frontier AI model (Claude Fable 5) given the same access to the same ad libraries we use — deliberately, because if it wins on equal data, we have no product and we would rather know. And the same model with no data at all, answering from what it already knows, which is what most "AI for marketing" actually is.
We wrote down what we expected before we ran it, published every answer verbatim, and checked every claim against the underlying ads. Seven market cells. Over ten thousand ads read.
Finding one: on the first question, the AI tied us. And cost less.
Given the same data access, the frontier model produced a legitimate, well-evidenced answer. It found the same market structure we did. On the first question it was cheaper to run than we were.
If your team asks one question about your market per quarter, buy the AI. We mean that.
But nobody asks one question. You ask what's working, then who's running it, then what changed since last month, then what your new competitor is doing, then the same thing again for the other platform.
Finding two: the second question is where the money is
The AI has no memory of your market. Every question re-buys the data and re-reads it from scratch. We keep a screened, analysed store of your market, so the second question is a lookup.
| Questions asked about one market | AI agent with data access | MessCube |
|---|---|---|
| 1st | $1.42 · ~16 min | Platform fee · ~20 min |
| 2nd | $1.42 again · ~16 min again | Included · seconds |
| 10th (running total) | $14–30 · ~2.5 hours | Included · seconds |
| 50th (running total) | $71 · ~13 hours | Included · seconds |
We ran that to its conclusion: 36 consecutive questions in one session. The AI's answers stayed accurate — we predicted they'd degrade, they didn't, and we published that too. What degraded was time. By question 18 a single answer took 9½ minutes. The same questions against our store returned in a tenth of a second.
Multiply by a loaded hourly rate and the "cheaper" option stops being cheaper somewhere around question three.
Finding three: AI without your market's data invents it
The third contender — the frontier model with no data, the configuration behind most AI marketing tooling — was fluent, confident and specific. It named competitors. It ranked angles. It sounded exactly like the others.
43% of the competitors it named weren't running ads where it said they were. Real companies, wrong market, stated with total confidence. Its top-ranked angle for one market didn't exist there at all: it described the well-known challenger brands, while the actual money was going to a category of advertiser it never mentioned.
It got the strategy right and the market wrong. That is the most expensive kind of wrong, because it reads as insight.
What this means for you
You are already paying for this. The question is what you get back.
Performance teams burn 30–50% of creative budget testing concepts that were never going to work. Not waste through incompetence — waste through not knowing what the market has already proven.
The difference between the two options is not the software line. It's that one is a fixed platform fee against an asset that compounds, and the other re-buys the same market data every time someone asks a question. Fifty questions on the agent is $71 in data and roughly 13 hours of someone waiting. At any loaded rate you care to use, the waiting is the expensive part.
Decisions your team can show their work on.
Every theme we report carries named advertisers and the specific ads behind it. When someone asks "how do we know?", there's an answer that isn't "the AI said so." We checked 381 of 381 citations in our own benchmark against the underlying ads.
And the read is stable: ask the same question five times, get the same answer five times. A chat-based tool gives you a different answer each time, which is fine until you've built a quarter's plan on one of them.
The market you're actually in, not the one you assume.
In our own test the AI's picture of a market was the brands you'd name from memory. The real picture — from the ads actually running — was a completely different tier of advertiser doing the volume, with different angles, different price points, different hooks.
We also read the ads a text-only tool can't: image and video creative is 100% analysed, versus roughly half a market being readable from copy alone.
The AI did four things better than us. It cited how long an ad had been running as proof of spend. It noticed a brand ran zero ads on a platform they'd assumed they were on. It spotted an offer nobody in the market was making — whitespace. And it found competitors our system hadn't discovered on one platform.
Three of those were things our data could already answer that we simply weren't showing. One was a real gap. All four are fixed or scheduled, and they're in the public technical paper with the same prominence as everything we won.
We publish the losses because a benchmark where the author wins everything isn't a benchmark. It's a brochure. You can check ours.
See your market's actual read
Give us your brand. We'll show you the angles running against you right now, who's running them, and the ads behind every claim.
Every number here comes from the full technical paper, published with its methodology, its limitations and the raw transcripts at messcube.com/blog/measured-advantage-benchmark. Check it.