AI assistant on a live ad account: read-only first or full write access from day one?

Take a dental clinic with two Google Ads campaigns that connects an AI assistant to the account. Plenty of people will just grant write access and go, figuring preview-before-apply covers most of the risk. I'd keep the first two weeks read-only. Not because the AI will do something wild, but because you don't yet know how it reads your account.

Ask it for a search term report, a conversion summary, and a list of what it would change, then compare that with what you'd have concluded yourself. If it mixes up a brand campaign with a generic one, or misreads which conversion action is primary, you want to catch that before it proposes a bid change. Where do others draw the line, especially people who've been running this for a while?

Agreed on the read-only window. One check I'd add: compare its conversion totals against the Primary/Secondary setting on each action. Google only bids on primary actions, so an AI that adds up everything, page views and call clicks included, will sound confident and still be wrong.

It depends on whether anyone has checked the primary conversions against real outcomes. Matching the Primary/Secondary setting only proves the AI read the account correctly, not that the setting is right. Dental accounts often have a "form submit" marked primary that fires on the thank-you page, or a call action that counts every tap on a mobile number.

If someone has already reconciled those numbers against booked appointments or CRM leads, a short read-only window mostly just tests how well the AI reads the account. If nobody has, the AI will optimize faithfully toward a broken number, so that reconciliation has to happen before any bid change gets proposed. In that case I'd ask the client how many patients they booked last month before looking at anything Google reports.

Agreed on reconciling against booked patients first. I'd also look at the Count setting on each conversion action, since it's easy to miss. For a form or booking confirmation it should be One per click. If it's set to Every, a patient who submits twice, or a confirmation page that refires on reload, gets counted twice, and the numbers look strong until the appointment book doesn't match.

This matters more with Performance Max than with Search. PMax has no keyword list holding it back, so it will chase whatever cheap action the primary setting rewards, and a doubled lead count can make a weak campaign look like it's scaling well.

So when you do the patient-count comparison, check the Count setting and the attribution window on those actions too. Both take a minute to fix, but finding out a month later is what costs money.

Depends on how long the account has been on Every. Switching Count doesn't rewrite past conversions, so the older months will still look inflated. If you're reconciling against booked patients, only use the period after the fix, or the comparison will mislead you.

The conversion lag report matters here too, along with the attribution window. A dental patient might book a consult weeks after the first form submit, so if you only compare the post-fix period, make sure it's long enough for those leads to close. Otherwise the newest month will look like a collapse when it's really just lag.

Waiting out the lag only helps your reading of the numbers, and it does nothing for the campaign. Google keeps bidding on whatever the primary action reports, so a doubled count keeps training the account for the whole lag period.

A faster route is to match CRM leads to the day they came in and compare them with Google clicks on the same click dates. That gives you a usable read in days, and you can fix the primary action before the optimizer learns much more from the bad signal.