9 Ways AI Gets Your Marketing Numbers Wrongwith a 30-second check for eachwith a 30-second check for each

9 Ways AI Gets It Wrong Explainer on the nine predictable mistakes AI makes with marketing numbers, with a 30-second check for each.

Fast isn't the same as right

What you get (+1 bonus)

Nine mistakes, nine worked examples.
From mixed definitions to averages of averages, each mistake shown on a small made-up shop or lead-gen dataset, with every right answer calculated in code.
The nine at a glance.
A one-table summary of all nine with the tell for each, such as platform numbers that add up to more than your real total, or a figure you can't find in any source.
Why each one happens.
The mechanism in plain English, from maths done in the chat to GA4's (other) row, plus what code execution fixes and what it still leaves to you.
Spot it, then prevent it.
For each mistake, the question that catches it in about 30 seconds and the sentence to add to your next question that prevents it next time.
The scored check, on one page.
Nine checks to run on any AI answer, scored out of 9, plus a prompt to paste first. The printable one-pager adds three habits that prevent most of the mistakes.
+1 bonus
An AI skill for Claude, ChatGPT, Gemini or Copilot
that checks any AI-produced number against the nine ways, gives the verdict first and, when your data is in the chat, has your assistant recalculate it in code.

Free · about 15 seconds

Who it's for, and who it's not for

A small-business owner at a desk types a question into an AI assistant on a laptop, with a notebook of handwritten figures beside it.

Who it's for

Owners and marketers of online shops and lead-gen businesses who ask ChatGPT, Claude, Gemini or Copilot about ROAS, orders, leads or cost per lead.
Agencies and freelancers who report numbers to clients and want to catch an AI mistake before it reaches a client report.
Anyone who uses AI for marketing numbers and wants to keep using it, with a 30-second way to tell which answers to act on.

Who it's not for

Data teams who already run every calculation in SQL or Python, with tests. You'll recognise most of what's in it.
Anyone looking for a ranking of AI models. The chapters describe mechanisms and don't claim that any particular model made any particular mistake.

What's inside

PDF · About 25 min read

Honest note

The check tells you when an answer needs a second look. It can't repair the data underneath. If your purchase or lead tracking is broken, the fix is in your tags, and that's a job for whoever implements your tracking. Every example uses a small made-up dataset, built to show one mechanism. AI tools change monthly, so the guide dates its platform facts and re-checks them every quarter.

Free · about 15 seconds