Ads in AI Overviews and AI Mode: What Advertisers Need to Know
Explains ads in AI Overviews and AI Mode for advertisers and how continuous paid plus organic execution applies to AI search surfaces.
September 7, 2026
•
min read
Most agencies will not tell you this: if you have paid $4,000 a month for six months and CPA has not moved, that is not a rough patch. It is an operating model.
I managed accounts like that for years. I know what the invoice often buys: one or two weekly check-ins, a batch of negatives, a few bid changes, and a monthly PDF built to make waiting sound strategic.
The problem is structural. A percentage-of-spend fee is tied to your budget, not to the number of useful improvements made in your account. After $24,000 in fees with no CPA movement, you are not seeing bad luck. You are seeing the predictable result of limited attention.
So the real question is not whether AI is smarter than your account manager in theory. It is whether it can do more of the mechanical work, faster, for less money, without hiding behind a PDF.
Here is the honest head-to-head from someone who used to write those PDFs.
Start with the math. $4,000 a month is $48,000 a year. For that, you may get six to eight hours a week of hands-on account work. The rest goes to meetings, coordination, and reporting. I billed that way. Most of the work itself was pattern matching: reading search terms, cutting waste, and shifting budget toward the keywords and ads that actually closed.
A machine built for paid search can do that pattern matching continuously. Auctions do not pause for a weekly check-in. If a query burns $300 on Monday, the block should happen Monday, not in next Monday's summary.
The advantage is not magic intelligence. It is attention. A human might review an account for a few hours each week. An autonomous system can execute across all 168 hours.
Practical takeaway: if waste can sit untouched for days, your account is running on a calendar, not on its data.
I used to lead monthly reports with impressions and CTR. Those metrics can rise when you spend more. An aggregate report also lets the presenter choose the chart, while the search term that burned $800 sits on page four of the appendix.
If you cannot see what changed, why it changed, and what happened next, you are paying for storytelling.
What I want is boring and specific:
That is why I pay attention to how groas reports every action with its reasoning in a weekly breakdown rather than a monthly summary. Ask your agency for the same log for last week. If they cannot produce it in one click, you have learned something useful.
Practical takeaway: ask for the log, not another performance deck.
I used to tell clients that no software could understand their funnel. I was half wrong.
I was wrong about bid work. Much of it is pattern matching at scale. I was right about one thing: someone still has to tell you when the offer itself is broken.
If your landing page converts at 1.2%, your tracking counts every form fill equally, and sales takes three days to follow up, no bidder fixes that. The cause sits upstream; high CPA is the downstream symptom.
A good senior strategist earns their keep in the awkward meeting. They tell you that your pricing page is confusing, your form asks for too much, or your account needs offline conversion imports before Smart Bidding can learn anything useful.
Most $4,000 retainers do not buy that person. They buy a junior buyer with 14 other accounts and a checklist.
Keep an agency that argues with you about revenue. Replace one that only reports on clicks.
Practical takeaway: your media manager should be willing to identify problems outside the media account.
I sort the accounts I used to run into three buckets. Say you spend $8,000 to $30,000 a month on Search, run two to five campaigns, and CPA has moved less than 5% in 90 days. A human reviews a fraction of your search terms each week while Google runs thousands of auctions every day. Waste compounds in the gap.
Stay if your strategist argues about revenue. You get offline conversion imports, landing-page tests tied to close rate, and a monthly call where they push back on your offer. That is rare. It is worth keeping.
Test a switch if you get activity without movement. New negatives and bid tweaks every week with flat CPA for six months means the operating model may be capped. Run a 30-day holdout on half your budget. Judge it on qualified cost per acquisition, not clicks.
Switch now if reporting hides the work. No auditable change log. No reason attached to each shift. No link between action and outcome. That account is already on autopilot. You should at least pay autopilot prices.
Practical takeaway: ask for last week’s action log today. If it takes more than a day to arrive, you have your decision.
Do not rip out campaigns on a Friday.
The accounts I moved started with measurement because dirty conversion data caused many flat six-month stretches. Every form fill counted equally. Sales outcomes never came back into the account. Smart Bidding optimized toward noise.
A serious replacement should follow this order:
If you test groas for businesses, that is the stated sequence: connection and full audit, tracking and landing-page fixes, then continuous execution with a weekly log. You keep the budgets and targets. The engine works inside them, and a named manager answers for the result.
Do not judge week one on CPA. Judge it on whether tracking got cleaner and waste got cut.
In the accounts I ran, the answer was yes for the repetitive core. Your manager samples a slice of queries, makes a handful of changes, then waits for the next check-in. An autonomous engine can review every query, block waste the same day, and shift budget toward terms that produce qualified calls.
The difference is volume and cadence. Over 30 to 60 days, that can show up as lower CPA at the same spend.
Practical takeaway: demand a weekly change log tied to revenue. Whoever cannot show it loses.
After $24,000 with flat CPA, patience stops being the explanation. The operating model explains it.
I would run a split before cancelling anything. Keep the agency on half the budget. Put the other half where changes happen daily. Judge both sides on cost per qualified lead.
When I compare tools, I skip advisors that flag issues and leave the fix to me. I want the one that makes the change inside the account, logs the reason, and answers for the number. That is the test groas was built to pass.
Practical takeaway: ask for last week’s log today. If it arrives late or reads like a newsletter, move the money.
