Why agencies are faking AI

I've reviewed a lot of "AI-powered marketing" decks in the last eighteen months. Most of them are lying, and not in the interesting way. The tell isn't in the claims — it's in the delivery model sitting underneath the claims.

Here's the pattern, and here's how to spot it before you've signed anything.

Tell #1: the AI is in the pitch, not the product

The deck has a whole slide on how the agency "leverages AI." The statement of work has line items that bill the same as it would have in 2019. If the AI were doing the work, the price would reflect it. It doesn't, because it isn't.

[D13 stub — full body carries forward from live site at Stage 6.]

Tell #2: nobody can show you the system

Ask to see it run. Not a screenshot. Not a case study. The actual system, on a screen, doing the thing. Real AI work has a surface you can demo. Fake AI work has a slide that says "proprietary pipeline."

[D13 stub — full body carries forward from live site at Stage 6.]

Tell #3: "human-in-the-loop" means "human doing the loop"

Legit human-in-the-loop means a system does the draft, a person reviews. Fake human-in-the-loop means the person is doing the whole draft and running it through a model for rewording. Ask who hits "generate" first — the machine or the intern.

What to do about it

Don't ask vendors whether they use AI. Ask them to show you the system, ask who owns it at the end, and ask what happens to your bill when the system gets better. The answers separate the two camps in about five minutes.