An AI assistant says your tracking needs fixing. Should you act on it?
More and more clients bring me a list of urgent fixes from ChatGPT, Claude or Gemini. Some are right. Many are confident answers about a website the tool never actually saw.
Published
Treat an AI recommendation as a question to check, not a task to do. An AI assistant can't see your live site, your GTM container, your consent settings or your sales records, so it fills the gaps with what's common elsewhere and presents the result as urgent and certain. Before you change anything, ask what evidence would show the problem is real, look for it in GTM Preview or GA4 DebugView, and compare it with your own business records. If it holds up, fix it. If you can't tell, have someone look at the actual setup.
Seeing this in your own setup?
Tell me what you're working on
Why AI tracking advice goes wrong
- It can't see what's running. An assistant works from your description or a screenshot, not from your site, your tags or the data layer. Two sites with the same report can have completely different causes.
- It fills gaps with the typical setup. If your form, checkout or consent banner works differently from the common case, the advice was written for someone else's site.
- Its knowledge lags the products. GA4 and Google Tag Manager change most months. In October 2026 alone, GTM changed how config commands appear in the data layer.
- It sounds certain either way. A wrong answer reads exactly like a right one, and urgent framing makes waiting feel risky.
Recommendations that deserve a second look
These come up again and again. Each can be right on some sites and harmful on others.
- "Add the GA4 tag to every page." If Google Tag Manager already sends GA4 data, a second tag doubles page views and conversions.
- "Turn on form tracking in GA4." Next to an existing form event, this can count one inquiry twice. Which form event to count.
- "GA4 and Google Ads don't match, so your tracking is broken." They count differently, so some gap is expected. The question is whether it can be explained. Why reports disagree.
- "Your Direct traffic is too high. Fix your UTMs." Sometimes. Consent banners, redirects and how a visit started can all push traffic into Direct, and none of those are UTM problems.
- "Switch to server-side tracking" or "use advanced Consent Mode to recover data." Both can be the right move, but they carry real costs and consent decisions. They're not quick wins. What each option involves.
- Steps for menus that don't exist. Instructions for settings Google has moved or renamed, or for Universal Analytics features GA4 never had.
How to get more out of AI advice
- Ask for questions, not fixes. "What would I check to confirm this?" gets you further than "How do I fix this?"
- Give it real context: your website platform, how GA4 is installed and whether you use a consent banner. Never paste customer details, passwords or API keys.
- Check the evidence yourself in GTM Preview, GA4 DebugView or Realtime, and against your CRM or order records.
- Ask what the fix could break: double counting, consent behavior, or reports other people rely on.
- Change one thing at a time, note the date, and compare the numbers before and after.
When to bring in a person
If a recommendation touches conversions, consent or numbers you report to leadership, have someone look at the actual setup before you change it. That's a common reason clients contact me now: send me the AI's recommendation and your website, and I'll tell you whether it applies and what I'd do instead. Often that takes one short conversation, and it can save weeks of chasing the wrong problem.