Before you turn on AI Max, fix the conversion data feeding it

Most accounts turning on AI Max are making a mistake. Not because AI Max is bad. Because they are feeding it garbage conversion data and expecting the algorithm to compensate.

AI Max is the right answer to Conversational Discovery Ads. Google rewrites your copy to fit the format, expands your URLs, handles your matching. For most advertisers that is genuinely better than manual. Embrace it.

But AI Max learns from your conversion signals. If you are counting form views as leads, if your GCLID passback is broken, if you are pulling in soft micro-conversions to pad the volume, the algorithm optimises toward that behaviour at scale. Faster, more aggressively, and with more of your budget than any manual campaign would.

The accounts that will win with AI Max are not the ones who resist it or the ones who flip the switch and walk away. They are the ones who spent a week fixing their measurement layer first.

What "garbage conversion data" actually looks like

The specific patterns we see most often in pre-AI Max audits:

• Form view conversions counted as leads. The user landed on the contact page, the page fires a "view" event, the event is set as a conversion. Now the bidder thinks any user who arrives on /contact is a lead.
• Conversions where verified status is "Not verified" or "No recent conversions". The conversion exists in the account, but Google's tracking does not have a clean ping in 30+ days. The algorithm uses historical data with no recent ground truth.
• GCLID passback broken between your site and your CRM. Conversions happen, the platform sees them, but the offline conversion import has not flowed for weeks. Bidder learns from incomplete data, then misreads quiet patches as performance drops.
• Primary conversions set to weight 1 across the board when actual values differ by 10x. Bidder treats a £30 newsletter signup the same as a £300 demo booking.
• Soft engagement events (scroll depth, time on site, video plays) sitting in the primary column to "pad the volume." Bidder learns toward engagement, not toward revenue.
• Multiple conversion actions firing for the same business event (form fill plus thank-you-page view plus email confirmation), each counted independently. The bidder sees three conversions where there was one event.

Each of these individually does not bring an account down. AI Max with all of them combined is a different problem. The algorithm does not pause and ask whether the signal makes sense. It optimises toward the signal you gave it.

Why this matters more with AI Max than with manual

A manually managed campaign has a human in the loop. Junk conversions get noticed in weekly reviews. Patterns that do not match business reality get flagged. The damage is contained by the cadence of human oversight.

AI Max removes most of that cadence. The system makes more bid decisions per day, expands into more queries, rewrites more copy, all based on the conversion signal you provided. The system is faster and more aggressive than any manual setup, which is the entire point. Faster and more aggressive at optimising toward a bad signal is faster and more aggressive at burning budget.

The accounts that struggle with AI Max in the first 60 days almost always have a conversion-data issue underneath, not an AI Max issue.

The pre-launch checklist

Before you activate AI Max, run this audit:

• Open Conversions in Google Ads. Check the "Status" column. Every primary conversion should be verified or have a recent conversion. Anything else either gets fixed or moved to secondary.
• Verify GCLID passback if you use offline conversion import. Run a test conversion from a known click through to your CRM and confirm it appears in Google Ads within 24 hours.
• Check your primary conversion values. If everything is weight 1, you are telling the bidder all conversions are equal. They are not. Either set realistic values per action or implement Conversion Value Rules.
• Audit which conversions are primary vs secondary. Anything that does not correspond to actual revenue or a qualified lead should be secondary.
• Compare reported revenue in Google Ads against actual revenue in your CRM or accounting system for the last 30 days. The gap (and which direction it points) tells you whether your measurement layer is reading too high, too low, or close to reality.

If any of these checks come back uncomfortable, that is the work. Not turning on AI Max sooner.

What to actually do

• Fix the measurement layer this week, before any campaign-level changes.
• Implement at least one independent revenue check (Google Ads vs CRM monthly, at the campaign level).
• Once measurement is clean, activate AI Max in one campaign first, not the whole account. Watch for two to four weeks before rolling out further.
• Document the conversion configuration before AI Max launches so you can compare bidder behaviour against the baseline.
• Most agencies skip this step because fixing conversion tracking is not a retainer line item. Launching AI Max is. Pick the agency that does the unglamorous work first.

Before you activate AI Max, open your conversion actions and check what percentage have verified status. Then check whether your primary conversion action maps to actual revenue, not traffic. If either answer is uncomfortable, that is the work.

The black box performs exactly as well as the signal you put in.

If you want a free pre-AI Max conversion-data audit, book a free audit and we will catalogue every signal the algorithm will optimise toward before you flip the switch.