AI agents build a lot of my clients' marketing now. Ads, blog posts, social posts, website pages. They're fast, and they're good. They're also wrong often enough that nothing they make goes live until I've checked it.
This is the AI marketing quality control checklist I run on every piece. Use it whether you run the agents yourself or pay someone to. If whoever does your marketing can't describe something like it, that's worth a conversation.
Mistakes I caught this week
First, why this matters. These are real, from my own work in the last few days.
| What happened | Why it mattered | The check that caught it |
|---|---|---|
| Building a Google Ads search ad, the agent pinned a headline to position 2 instead of position 1. | Pinning Headline 1 was the whole A/B test. The wrong slot would have quietly ruined it. | Reading every setting against the plan, not just the ad text |
| The agent clicked a checkbox twice, turning a setting on and then right back off. | The screen looked finished. The setting wasn't applied. | Checking the saved result, not the agent's summary of it |
| While checking conversion tracking, the agent found one site had no form listener at all. | Conversions on that site had never recorded, so every report undercounted leads. | Testing tracking with a real submission before trusting it |
Two of those were the agent's own mistakes. The third was the agent catching a gap that had been sitting there unnoticed. Both kinds are why review exists. I walk through the campaign build in build a local Google Ads campaign with AI, and the tracking fix in how to set up Google Ads conversion tracking with AI.
What does AI marketing quality control actually mean?
It means a person who knows the business checks the work against a standard before customers see it. Not a skim. A real check against a list.
The reason is simple. AI writes and clicks with the same confidence whether it's right or not. It'll state a price you've never charged in the same calm tone it uses for your phone number. The confidence tells you nothing, so the checklist has to.
My rule: agents can draft, build and fill in anything. Nothing publishes, sends, spends money or gets deleted until I approve it. Anthropic's safety guide for Claude in Chrome lands in the same place: review proposed actions before you approve them, because you remain responsible for what the agent does.
Facts, claims and the offer
1. Facts and claims
I have the AI list every factual claim in the piece, then I check each one myself. Prices, years in business, licenses, warranties, response times, anything about how a product or platform works.
- Numbers. Every number gets traced to a source or to the owner. If neither, it comes out.
- Superlatives. "Best," "fastest," "number one." Can you prove it? If not, cut it.
- Reviews and testimonials. Real ones only, from real customers, with permission. The FTC's rule on consumer reviews and testimonials prohibits buying or selling fake ones. An AI will invent a glowing quote without blinking if nothing stops it, so I confirm every quote is real.
2. The offer
This one causes the angriest phone calls. AI loves to round an offer up. "Free estimates" turns into "free service calls." A seasonal discount gets an end date that doesn't exist, or loses the one it had.
- The exact offer. Does it match what your front desk will honor when someone calls?
- Dates and fine print. Expiration dates, exclusions, service area limits.
- Services. Does it promise anything you don't do? Agents fill gaps with what's typical for your industry, not what's true for you.
Voice and local details
3. Brand voice
Read it out loud. If you wouldn't say it to a customer standing in your shop, change it. I look for the generic AI tells: throat-clearing openers, buzzwords, everything in threes. I go through those in why all AI content sounds the same.
4. Local details
This is where agents slip most quietly, because their mistakes look plausible.
- Towns and neighborhoods. Do you actually serve every place it named? Did it misspell one?
- Phone number and address. They should match your Google Business Profile exactly.
- Hours. Including holiday hours, and whether you really answer on Saturdays.
- Local references. If it mentions a landmark, a season or a local quirk, is it true?
Links and tracking
5. Every link and every tag
A great ad that sends people to a broken page is worse than no ad. Broken tracking is worse than that, because you'll make decisions on bad numbers for months without knowing it.
- Click every link. On a phone, not just a laptop. Make sure it lands on the right page, not the homepage.
- Campaign tags. Links in ads, emails and social posts should carry UTM tags, which are short labels on the end of a link that tell Google Analytics where a visit came from. Google's URL builder guide explains them.
- Conversion tracking. Before any ad goes live, I submit a real test lead and confirm it shows up. Google Tag Manager's preview and debug mode shows which tags fired on the page before anything is published.
- Phone numbers. Tap the number in the ad or on the page. Does it ring your business?
Policy compliance: Google, Meta and reviews
6. Ad platform rules
Agents know the platforms in general. They don't always know the rules in detail, and a disapproved ad on a Friday afternoon costs you the weekend.
- Google Ads character limits. Responsive search ad headlines max out at 30 characters and descriptions at 90, according to Google's help page on responsive search ads. Agents sometimes miscount or cut a word in half to fit.
- Google's editorial policy. Google's editorial requirements prohibit things like gimmicky capitalization or punctuation, unnecessary repetition, and putting a phone number in the ad text.
- Meta special ad categories. Ads about employment, housing, financial products and services, or social issues and politics have to be declared, per Meta's documentation. That catches small businesses off guard. A "we're hiring technicians" ad is an employment ad. A "financing available" ad may fall under financial products and services.
7. Review and profile rules
Anything that touches reviews gets extra care. Google's prohibited content policy removes reviews posted because of an incentive, and says businesses may not selectively solicit positive reviews from customers. So no "happy? leave a review, unhappy? call us" messages, no discounts in exchange for reviews, and no review replies that argue with a customer.
Accessibility and the final approval
8. Can everyone use it?
- Alt text on images. The W3C's images tutorial says images must have text alternatives that describe their information or function. Agents often skip it or write "image of a truck."
- Readable text. Text on graphics should be big enough to read on a phone, with enough contrast against the background.
- Clear link text. "Schedule a furnace tune-up" instead of "click here."
Final human approval
Last step, every time: a person who knows the business approves it. Then it goes live. Then I look at the live version once more, because what got saved and what's actually showing aren't always the same thing.
Two prompts for a second-pass review
I use AI to help with review too, as a second set of eyes, never the final one. These work best in a Claude or ChatGPT project that already has your business documents loaded.
Prompt: claims and offer audit
Review the draft below before it goes live. Your job is to find problems, not to rewrite it. Use the business documents in this project as the source of truth.
List every factual claim, number, price, offer, date, service, town, phone number, review and testimonial in the draft. Mark each one Confirmed (say which document confirms it), Conflicts (quote what the document says instead), or Unverified. Flag every superlative like "best" or "fastest." Don't change the draft and don't publish anything. Give me the list and stop.
[paste the draft]
Prompt: ad policy pre-check
Check these ads for [Google Ads or Meta] before I submit them. Context: my business is a [type of business] in [area], and the ads promote [offer].
For Google responsive search ads, count the characters in every headline (30 max) and description (90 max) and list any that run over. Flag gimmicky capitalization or punctuation, repeated words, phone numbers in the ad text, and any claim I'd need to prove. For Meta ads, tell me whether the ad might fall under a special ad category such as employment, housing, or financial products and services, and why. Where you're unsure about a policy, say so and tell me which official policy page to check. Don't edit or submit anything. Report back and wait for me.
[paste the ads]
The AI catches things I'd miss on a tired Friday. I catch things it misses every day of the week. That's the point of having both.
Alex Lamy16 years in marketing, from radio to agencies to running his own shop. Now he runs a team of AI agents that do the marketing for St. Louis small businesses. You talk to him, and he checks every piece.
Want this handled for you?
$499 a month, flat. My AI agents produce it, I check every piece, and we meet weekly. Month-to-month, and you own everything. Want a reference first? I'll connect you with someone I did similar work for.
Book a 30-minute call
