Ask five agencies what AI has done to Google and Meta Ads and you will hear five versions of one answer. The machine chooses the audience now. Creative is the lever. Feed it good data, keep the structure simple, and let it work.
That answer is right. Corient lays it out well for Meta, RedPandas makes the point that the marketer’s role has been redefined rather than removed, and Birch and In Marketing We Trust cover how ads now sit around Google’s AI Overviews. We agree with nearly all of it.
Our founder worked inside both Google and Meta before starting Dadek Digital, so we have seen these systems from both sides of the table. Two things get far less airtime than they deserve. The machine is extremely good at hitting the target you give it, and it has no opinion on whether it is the right target. And it now writes the report on its own performance as well. Those two facts decide whether AI makes you money or just spends it faster.
Why this matters
The machine took over the doing
Start with targeting. Meta’s Andromeda system narrows tens of millions of candidate ads to a few thousand for each person, and Meta says it allowed a 10,000-fold increase in model capacity (Engineering at Meta). Detailed targeting exclusions are gone, and Meta says an advertiser test showed a 22.6% lower median cost per conversion without them (Meta for Developers). Interest and demographic settings are increasingly treated as suggestions rather than limits (RedPandas). Advantage+ is the default for new sales, leads and app campaigns, and Meta’s Advantage+ products passed a US$75 billion annual revenue run rate in Q2 2026 (Corient).
Bidding went the same way. Google’s Smart Bidding sets a bid in every single auction, optimising for conversions or conversion value (Google). Creative is heading there too: more than a million advertisers used Meta’s generative AI tools to create over 15 million ads in a single month, and that was back in 2024 (Engineering at Meta).
The jobs that used to fill a media buyer’s week, stacking interests, adjusting bids by device and hour, splitting keyword lists by match type, are now mostly done by the platform. Mostly done well, too.
Search changed shape
On Google, ads now appear above, below and inside AI Overviews, and the inside placement is live in Australia (Google). Those ads are matched to both the query and the content of the AI answer, which is why Google requires broad match or keywordless targeting, such as AI Max, Performance Max, Shopping or Dynamic Search Ads, to reach them (Google).
You cannot target the AI Overview placement, you cannot opt out of it, and Google offers no separate reporting for it (Google). Impressions can hold steady while click-through falls, because an ad sitting under a long AI answer has to wait for the scroll (Birch).
The machine now writes the report as well
This is the change we would watch most closely, and the one that gets the least attention.
The Google Ads homepage now opens with AI-generated insight cards, new Dashboards turn a text prompt into a report with an automatic summary of the “why”, and Google Analytics benchmarks you against similar businesses (Google). Meta’s incremental attribution setting reports the conversions Meta predicts its ads caused (Corient). Meta’s reporting API stopped returning the 7-day and 28-day view windows from 12 January 2026 (Meta for Developers).
None of this is sinister, and we use these tools ourselves. But for the first time, one system picks the audience, sets the bid, can assemble the ad, and then explains how it all went. In any other part of your business, you would not let a supplier audit its own invoice.
The bill is rising while this happens. Meta’s average price per ad was up 14% year on year in Q3 2026 (Meta).
What has not changed
AI changed who does the work. It did not change who decides what the work is for. Four jobs are still yours, or your agency’s.
Deciding what a customer is worth
Google’s own Smart Bidding guide lists “increase profit” as the goal for Target ROAS and Maximise conversion value (Google). Both strategies maximise whatever value you send them. In most accounts, that value is revenue.
So a $200 order at 10% margin and a $200 order at 60% margin look identical to the machine, and it will cheerfully find you more of the first kind. Lead generation has the same trap: optimise for form fills and Meta finds people who fill in forms (Corient).
The fix is a business decision, not a setting. On a Dutch software retailer’s Google Ads account, we split Shopping and Performance Max by margin tier and made new customer acquisition cost the target instead of blended ROAS. Google’s systems did the bidding. We decided what they were bidding towards. Across a ten-week Q4 the account delivered €1.07M in revenue at 13.2x ROAS, with new customer acquisition cost down 26% (Dadek Digital).
Feeding it correct data
Smart Bidding needs conversion tracking to work at all, and Google suggests judging it over periods with at least 30 conversions, or 50 for Target ROAS (Google). A Meta ad set usually needs about 50 optimisation events in a week to leave the learning phase (Corient).
The machine learns from exactly what you send it. Duplicate purchases, a pixel firing twice, junk leads counted as wins: it learns those too, and faster than any person could. Meta’s own figure for connecting your CRM, so the system can optimise on lead quality rather than lead volume, is a 19% lower cost per quality lead on average (Corient). It is Meta’s number, so treat it as a reason to test, not a promise.
Owning the offer
AI made ads cheap to produce. It did nothing for the reason to buy. Corient puts it bluntly: with a weak offer, “better AI just finds the wrong people faster.” The platform can write ten headlines in a minute. It cannot tell you why someone should pick you over the business in the next ad.
Keeping score from outside the platform
The reporting changes make this job more important, not less. Keep one number the platforms do not calculate: revenue from your own systems against total ad spend, also known as MER. If Google, Meta and your MER all improve together, believe them. If the platforms improve and MER does not, believe MER.
What people overestimate
That AI Overviews killed paid search
In late 2025, a widely shared study from Seer Interactive found paid click-through on informational queries fell 68% where an AI Overview appeared (Seer Interactive).
Seer’s broader 2026 update, covering 53 brands, found paid click-through on AI Overview queries edged up through 2025, from 14.6% to 16.2%, while paid click-through on queries without an AI Overview fell from 26% to 21.8% (Seer Interactive). Google also says ads only show with AI Overviews when certain conditions are met, including detected commercial intent (Google).
The damage is real, but it sits mostly on informational searches, which were rarely the money searches for most advertisers. Check your own search terms before you move budget on a headline.
That you no longer need anyone at the wheel
Which leaves the question underneath all of this: do you still need a media buyer?
You need less of someone pushing buttons, and more of someone deciding what the buttons are for. Google frames its own new tools as amplifying the marketer’s expertise, not replacing it (Google). The hours have moved from settings to inputs: the value, the data, the creative brief and the check at the end. There are fewer of those hours, but each one is worth more.
On a budget of $5,000 to $30,000 a month, nobody needs to touch bids every morning. You do need someone who knows which number to point the machine at.
That simpler means any structure will do
Consolidating campaigns is right, but the reason is arithmetic, not fashion. Take $10,000 a month at a $100 cost per acquisition. That buys about 100 conversions a month. Split it across six campaigns and each gets around four conversions a week, well below what either platform asks for. Two well-funded campaigns learn. Six thin ones guess.
On a mid-sized budget, every split you add still has to pay for its own data.
Where to start this week
Three checks, none of which need a new tool.
Write down what each campaign is told to value. If the answer is revenue, ask whether it should be margin or new customers.
Compare the conversions each platform reports with your own records for the last 30 days. A big gap is a data problem before it is a performance problem.
Pick one number you calculate yourself, revenue against total ad spend, and judge both platforms on it every month.
AI will keep taking over more of the doing. That is good news, as long as someone on your side still owns the deciding.
Dadek Digital works in exactly that gap: we set the value the platforms optimise towards, keep the data they learn from clean, and judge the results on your revenue rather than their report. If you would rather run it yourself, the whole method is written down in The Dadek Digital Pyramid, in an Ecommerce and a Lead Generation edition.