Free Checklist

The AI Gold Test

10 checks we run before trusting any AI-generated campaign

Every agency you talk to this year is going to tell you they use AI. That's not information anymore — it's table stakes, like saying they have a website. The question that actually tells you something is what the AI is doing, who's checking its work, and whether that work survives contact with your real business.

This is the checklist we run internally before any AI-touched output — an ad, a landing page, a targeting recommendation, a budget shift — goes live on a client account. We're publishing it because the same 10 questions work on any agency or tool you're evaluating, including the one you're paying right now.

Print it, run it on your current setup, and see what comes back.

The 10 Checks

The Mechanism Check

Ask them: “Walk me through exactly how the AI reached this recommendation.”

Gold

They can trace it — the inputs it used, the pattern it found, the specific reason this ad, audience, or bid change came out the other end. It sounds like an explanation, because it is one.

Makeup on a Rock

"The AI analyzed the data and optimized for performance." That sentence is true of every tool on the market and explains nothing.

Why it matters: If nobody can explain why the AI did something, nobody can tell you when it's about to do something dumb — until it already has.

The Real-Data Check

Ask them: “Was this built and tested against my actual account history, or is this the same output every client in my category gets?”

Gold

The recommendation references your numbers — your cost per lead, your seasonality, your actual customer behavior — not category averages.

Makeup on a Rock

The strategy deck could have your competitor's name swapped in and nothing else would change.

Why it matters: AI trained on generic patterns gives you generic output. Your account isn't generic — it has a history, and a good system uses it.

The Human Fingerprint Check

Ask them: “Who looked at this before it went live, and what did they actually change?”

Gold

A named person, a specific edit — "our media buyer cut the AI's suggested audience from 2M to 400K because it was too broad for your service area." A human made a judgment call and can describe it.

Makeup on a Rock

"Our team reviews everything" with no detail on what review means or what, if anything, it ever changes.

Why it matters: Review that never changes anything isn't review — it's a rubber stamp with a human's name on it.

The Real-Conversion Check

Ask them: “What is the AI actually optimizing for — and is that the thing that pays your bills?”

Gold

It's optimizing for a tracked action tied to revenue — a booked appointment, a completed purchase, a qualified lead — and they can show you the tracking that proves it.

Makeup on a Rock

Impressions, clicks, "engagement," or reach. Numbers that go up while your phone stays quiet.

Why it matters: An AI optimizing for clicks will happily get you more clicks from people who were never going to buy anything.

Dental

Say your agency reports "we generated 40,000 impressions this month." Ask instead: how many of those turned into a booked new-patient appointment your front desk can confirm? If they can't answer, they weren't tracking the thing that fills your chair.

eCommerce

Say your dashboard shows a healthy ROAS but your bank balance disagrees. Ask what the AI counted as a "conversion" — an add-to-cart? An initiated checkout? Only a completed, paid order is the number that matters to your cash flow.

The Economics Check

Ask them: “What do you think a lead — or a customer — is actually worth to me, and where did that number come from?”

Gold

They can state a number and defend it, and it reflects your margins, not a generic industry benchmark pulled off a blog.

Makeup on a Rock

Blank stare, or a number that's clearly copy-pasted across every client in the vertical.

Why it matters: An AI system optimizing without knowing your economics is optimizing blind. It can hit its own targets and still lose you money.

Dental

Say a new-patient lead is worth $800+ in first-year production for an implant case, but only $150 for a routine cleaning. A system that treats every lead as equal will happily spend your budget filling the calendar with the wrong kind of appointment.

eCommerce

Say your margin on your hero SKU is 60% and on a clearance item it's 12%. A system optimizing for raw revenue instead of margin can hit a record sales month and still shrink your profit.

The Audit Trail Check

Ask them: “Can you show me a log of what changed on my account, when, and why — going back?”

Gold

A real record — dates, changes, the stated reason for each one — that you can review any time, not just at the monthly check-in call.

Makeup on a Rock

"We can put together a report if you want one." Meaning: nothing is being tracked as it happens, it gets reconstructed after the fact when asked.

Why it matters: If you can't see what changed and why, you can't tell the difference between a system that's learning and one that's guessing.

The Stress-Test Check

Ask them: “Tell me about a time the AI got it wrong. What happened, and who caught it?”

Gold

A specific story — the AI suggested something off (an audience too narrow, a bid too aggressive, copy that missed the mark), a human caught it before it spent real budget, here's the guardrail that exists now.

Makeup on a Rock

"It doesn't really get things wrong" — which either means it's never been checked closely enough to fail, or nobody's willing to admit it does.

Why it matters: Every AI system is wrong sometimes. The question isn't whether — it's whether a human catches it before or after it costs you money.

The Template Check

Ask them: “If I saw this exact strategy running for another client in my category, would I recognize it?”

Gold

No — because it was built around your specific positioning, your specific market, your specific gaps.

Makeup on a Rock

Yes, obviously, because it's the same playbook with your logo dropped in.

Why it matters: A template can be produced instantly and cheaply, which is exactly why AI makes it easy to hide one behind a personalized-looking wrapper.

The Consistency Check

Ask them: “Show me this result held over consecutive months, not just the best one.”

Gold

A trend line — several months in a row, with the range shown, not cherry-picked.

Makeup on a Rock

One great month, presented as if it's the norm.

Why it matters: AI systems can get lucky in a given week the same way anyone can. A result that holds for a year tells you the system is actually working, not that you caught a good stretch.

Dental

"We got a dental practice's cost per lead down from $888 to $56" means a lot more when it's the number after 11 straight months, not a single strong week during a slow season for the practice's competitors.

eCommerce

A single 8x ROAS week is easy. What's harder — and worth asking for — is 3.4x to 6.5x held for 12 consecutive months, because that's a system, not a spike.

The Show-Me Check

Ask them: “Can I see the actual dashboard, workflow, or system — right now, live — instead of a slide describing it?”

Gold

They pull it up. You're looking at the real thing.

Makeup on a Rock

A case study slide, a screenshot from a pitch deck, or "we can set up a demo of that" that never quite happens.

Why it matters: Anyone can describe a machine. Showing it is the only version of proof that can't be faked in a deck.

Scoring Guide

Score one point for every check that came back gold. Half a point for a partial or vague-but-honest answer.

8–10

Gold.

This is a system you can trust with real budget. Keep asking these questions periodically — gold can tarnish if nobody keeps checking.

5–7

Partial.

There's a real process under here, but pieces are missing — usually the audit trail or the economics. Ask them directly to fix the weak checks before you increase spend.

Under 5

Makeup on a rock.

The AI claim is doing more work than the AI itself. Painted to look solid, nothing underneath if you scratch it. Time to ask hard questions or start looking elsewhere.

What This Checklist Is Actually For

We didn't write this to describe what we do — we wrote it because most owners have never been handed a way to check. Run it on your current agency before you run it on us. If a vendor doesn't want to answer these ten questions, that's your answer.

If you want a second opinion on where you'd actually score, we'll run a free growth assessment on your account — no sales call attached, just a straight read on where the leaks are and what we'd check first.

Send me the free growth assessment