Sam Ford Collins
A weekly grading · Video series

AI slop, annotated.

Real AI-generated marketing, pulled from the wild and graded against Anthropic's 4 D's of AI fluency. What went out, why it reads as slop, and the line that would have caught it before it carried the brand's name.

New episodes land on LinkedIn and YouTube every week. The archive lives here. Companion reading: Where AI belongs, and what stays human.

Latest · This week on LinkedInThe caption that cost millions: you spend seven figures on a sponsorship, then let a bot announce it with a "that's not X, that's Y." (DiGiorno lives in all of us.)

Watch on LinkedIn →
Start here · The grading system

The 4 D's, in seven minutes.

Every episode grades one real ad against Anthropic's 4 D's of AI fluency: Delegation (what goes to AI at all), Description (how it's briefed), Discernment (a human with taste judging the output), and Diligence (owning the result). Watch this first and the grades explain themselves.

Episode 03 · July 2026

The bicycle Meta improved.

REI ran campaigns on Meta. The Advantage+ enhancements, on by default, decided their bicycle needed a second set of handlebars — under the seat, facing backwards. Nobody chose this. That's the point. Auto-enrollment means the platform is editing your product photography unless you tell it not to.

The line that would have caught itAI gets a territory, not a blank page. If a platform can alter the product image, the territory was never drawn. Go check what Advantage+ is "enhancing" in your campaigns this week.
Watch on YouTube →
Episode 02 · June 2026

Skechers takes the test.

One real Skechers ad, graded against the 4 D's. It earns some and drops others, and the difference is visible from across the room. The pattern to watch: the D's it fails are the human ones. Delegation was fine; discernment never showed up for work.

The line that would have caught itThe dividend is judgment, not time. If AI saved the team an afternoon and nobody spent ten minutes of it looking hard at the output, the trade went backwards.
Watch on YouTube →
Episode 01 · June 2026

The first specimen.

Narwhal Labs, graded against all four D's. Let's just say it didn't earn them. Technically fine, instantly generic — the work could belong to any brand, which is precisely the problem when the work carries yours.

The line that would have caught itSome work is the brand. If the output could sit under anyone's logo, it never should have gone out under this one.
Watch on YouTube →