What we switch on in a local Ads account

Google I/O is on the 20th and Marketing Live is the day after. Those two days reliably produce a list of new automation in Google Ads and, about a week later, a round of calls asking whether we should turn it on. Meanwhile AI Mode, which opened in Labs in March, keeps reaching more people than it did the month before.

We would rather agree the rules before the pitch than after it. So this is what is already on in the local accounts we run, what stays off, and how we will decide about whatever shows up in two weeks.

Two months of asking AI Mode our clients’ questions has not changed what we said in March, with one addition: the questions people type into it keep getting longer, and the constraint now arrives in the first sentence, evenings or a plan name or a same-week appointment. What has not changed at all is that there is nothing on that surface to buy, no placement and no setting, which is the sentence to repeat to whoever is calling your office this month to sell AI ranking.

What we switch on

The paid side is where the switches actually are, and most of the useful ones have been available for a while. In a typical local search account we run with these on.

1. A conversion action that counts booked work. Not a form load, not a page view, not a click to call that nobody answered. If the bidding is optimizing toward the wrong event, everything downstream is decoration.
2. Enhanced conversions for leads, plus offline import. The booking happens in the CRM days after the click. Sending that outcome back is what makes automated bidding useful instead of confident and wrong.
3. Smart Bidding, once there is volume. Target CPA on the booked action. Below roughly fifteen conversions a month we would rather run manual caps than hand a thin signal to a bidding model.
4. Campaign-level negative keywords in Performance Max. These arrived in January and they are the main reason PMax became usable again for local accounts. Brand exclusions go in at the same time.
5. Location targeting set to presence. Presence or interest is the default and it quietly spends your budget on people three states away reading about your city.
6. Generated asset suggestions, reviewed by a person before they run. The drafting is genuinely useful. The publishing is not automatic in any account we manage.
7. Call reporting, with the opening hours correct in the ad schedule. Paying for a click at nine at night and sending it to a voicemail box is the most expensive thing in most local accounts.

What we leave alone

1. Auto-apply recommendations. Off in every account, permanently. It changes match types and budgets while you sleep and then reports a higher optimization score for doing it.
2. Automatically created assets on a dental or medical account. Machine-written claims about treatment are a compliance problem, not a testing opportunity.
3. Search partners and display expansion inside a small search campaign. The traffic is cheap and it is not your patient.
4. Broad match with no conversion volume and no negative list. Broad match plus a real signal is fine. Broad match plus hope is a donation.
5. Anything sold as guaranteed placement in an AI answer. See above.
6. Any new campaign type in its announcement week. The first cohort tests it with their own money. We would rather read their results in July.

How we will judge whatever lands on the 20th

Four questions, in this order, for every new thing announced:

• Does it change targeting, matching or bidding? Those three are the ones that move money.
• Can we measure it against a baseline we already wrote down?
• Can we turn it back off, and how quickly?
• What does a fair test cost, and which account can afford to run it?

Then the rule that saves us every year: one change at a time, two weeks apart, and nothing new the same week a core update starts.

What to do this month

• Write down your baseline before the announcements: booked jobs per week by channel, cost per booked job, brand against non-brand. After the 20th you will not be able to reconstruct it.
• Check what your primary conversion action actually counts. Fix it if it counts form loads.
• Turn off auto-apply recommendations today, in every account, including the ones your agency runs.
• Add campaign-level negatives and brand exclusions to any Performance Max campaign that still has none.
• Ask AI Mode about the services you most want to sell, and fix any wrong fact it repeats at the source: your site first, then your profile.
• Put an hour in the diary for the 21st to read the announcements yourself rather than five newsletters summarizing each other.

The pattern of the last two years has been that the platforms automate the buying and leave the judgment to you. What you feed the automation is the part you still control, and it comes down to accurate conversion data in Google Ads and an accurate business sitting behind it. Those are the same facts that decide whether a written answer names you, which is why AI search visibility and local search end up being fed from one set of inputs rather than two.