The AI Bid Strategy Everyone Is Copying, and the One Signal It Cannot See

"Google will not automatically adjust your bidding targets or budgets." That sentence sits in a Google Ads help page describing a change that takes effect on August 17, and it is the most honest description of automated bidding anybody published this year. The system will spend toward the number you typed. Whether that number still reflects what a customer is worth to you is not a question it was built to ask.

I manage enterprise Google Ads accounts and I am not going to argue that Smart Bidding is bad at its job. It prices auctions better than any human ever did, and the accounts I have watched try to go back have not enjoyed the experience. The limitation is narrower than the LinkedIn version of this argument suggests, and it matters more.

Smart Bidding is Google's family of auction time bid strategies, including Target CPA and Target ROAS, that sets an individual bid for every auction using signals Google observes about the user and the context. Almost every serious account runs it now. That is the setup for the problem, not evidence of one.

Everyone Copied the Same Bid Strategy, and It Mostly Worked

The move to automated bidding was correct and the industry was right to make it fast. A human being cannot price an auction against device, physical location, time of day, browser, remarketing list membership and the query itself, in the fraction of a second available, several million times a month. The machine can. Accounts that made the switch generally got cheaper conversions and stopped losing money to the sort of manual bid mistakes nobody catches until the monthly review.

What nobody planned for is what happens once everyone does it. When every advertiser in an auction hands the same system a similar target and a similar conversion feed, the bidding layer stops being a place where anyone gets an advantage, and the differentiator moves upstream to what you feed it. That is a less comfortable place to compete, because it requires knowing things about your own business that the platform cannot tell you. I have written before about the broader cost of leaning on automation by default, and this is the sharpest version of it.

So What Is the One Signal It Cannot See?

Smart Bidding optimizes toward the conversion actions you configured, at the values you assigned them, across whatever queries your match types let into the auction. Every one of those is something you decided. The algorithm reads outcome and it reads context, and it has no way to form a judgment about whether a particular search was worth showing up for.

Two queries can carry an identical conversion rate and represent completely different buyers. One produces a lead that closes in three weeks at a healthy margin. The other ties up a salesperson for a month and buys the cheapest thing you sell, or never buys at all. If both fire the same conversion action at the same assigned value, the system sees two identical events and correctly decides to go find more of both.

Google has quietly made this harder to audit. Its own documentation now says that for searches arriving through Lens, AI Mode, AI Overviews and autocomplete, "the search term shown in your reporting represents the best approximation of the user's intent." The report is an inference layer rather than a transcript. Every negative keyword you write, every intent cluster you build, and every match type decision you make on top of that report is being computed against Google's paraphrase of your customer.

The August 17 Change Makes That Blind Spot More Expensive

Here is the mechanic, since some readers arrived only for this. Starting August 17, campaigns that are limited by budget and running a target based bid strategy will perform more consistently toward the target you set. Google's own worked example is a campaign with a ten dollar Target CPA that has been achieving a five dollar actual CPA. After the change it will deliver more closely to ten dollars. Google has said plainly that it will not adjust anybody's targets for them, and that campaigns overperforming against their targets may be affected if the advertiser takes no action. Target Impression Share and manual CPC are outside the scope, and Google's own FAQ is explicit that the auction itself is not changing.

Every explainer published so far stops at that paragraph. The part worth arguing about is what the additional spend actually buys.

A budget limited campaign that has been beating its target has been banking efficiency it never asked for. After the change it stops banking that efficiency and starts converting it into volume. Volume in a budget limited campaign comes from further down the same query curve, which is to say from the searches the campaign was previously too constrained to reach. Those queries were at the margin for a reason. Whether they are worth their new price is exactly the judgment the bidding system does not make, and the reporting will show you the aggregate rather than the decision. As of publication this has not taken effect yet, so treat the framing as a forecast rather than an observation.

Two Weeks Later, the Matching Default Moves Too

In the first week of August, Google began emailing advertisers a second date. From September 1, campaigns running automatically created assets or the campaign level broad match setting are upgraded to AI Max for Search automatically, with search term matching switched on, plus text customization for the automatically created assets campaigns. Nobody has to opt in.

Search term matching is the mechanism that decides a query is close enough to your keyword without being the keyword, and it is where the approximation in your reporting comes from. Mike Ryan at Smarter Ecommerce measured how far it has already spread, across 383 million EMEA ecommerce impressions: expansion on exact match keywords has climbed from roughly nothing in January 2025 to nearly 29 percent. More than a quarter of what your exact match reporting attributes to exact match arrived through matching Ryan describes as basically broad match.

So the recalibration lands on August 17, and two weeks later the matching default moves. One decides how hard a budget limited campaign pushes toward its target. The other widens what was eligible to absorb that spend to begin with, which is the larger change and is getting a fraction of the attention. Both arrive switched on, and the report you would audit them with is the one Google now calls an approximation.

Does Going Manual Fix It? No, and That Is the Wrong Fight

No. Going back to manual bidding gives up the auction time signals that were the actual win, in order to solve a problem that does not sit at the bid layer at all.

There is good evidence that second guessing the bidder makes things worse. Optmyzr has been measuring Black Friday and Cyber Monday against seasonality adjustments for three years running. In the 2024 cohort, 5,886 accounts, the ones applying adjustments saw ROAS fall 15.7 percent while the ones that left the system alone saw it rise 5.7 percent, and 2022 and 2023 lean the same direction. Frederick Vallaeys wrote the recommendation without hedging it: for mainstream retail events, do not use seasonality adjustments, and let Smart Bidding handle the macro trends.

The complication worth stating is that revenue growth was actually higher for the accounts that adjusted, in all three years, alongside sharply higher CPC inflation. What the adjustment bought was volume at a worse rate. Which is close to the trade the August 17 recalibration is about to make for budget limited campaigns whether they ask for it or not.

That study is about overriding the bidder's forecast of its own conversion rate. It says nothing about which queries you allow into the auction or what values you send back, which are different interventions belonging to different people. The bid price should go to the machine, which is better at it. The judgment about what a search is worth to your business stays with you, because nothing in the platform has access to the information required to make it.

What Query Level Truth Actually Looks Like in an Account

Start with the finding that should have changed more minds than it did. Andy Goodwin, who runs the paid search practice at Brainlabs, published twenty three tests across sixteen already mature advertisers. At campaign level the AI driven expansion looked like a 7 percent lift in conversion value, generated by queries the campaign had never targeted before. Then he reconciled those queries against the whole account. Only 46 percent of them were actually new to it. The other 54 percent had already been captured somewhere else in the same account, and what survived the reconciliation was roughly 3 percent incremental conversion value, which he was careful to scope to accounts already running heavy broad match.

More than half of what the automation "found" was already yours. You cannot see that from inside campaign reporting, because campaign reporting does not reconcile across campaigns. It takes an account wide view at the query level, which is work somebody has to choose to do.

Smarter Ecommerce reached a compatible conclusion from a different angle, looking at more than 250 retail search campaigns with AI expansion active. Median revenue went up 13 percent, median cost per acquisition on those incremental conversions went up 16 percent, and the ROAS spread across the sample ran from 42 percent above baseline to 35 percent below it. Mike Ryan's conclusion was that turning it on is essentially a coin toss where you may see a lift but efficiency likely will not follow. A spread that wide makes the average close to meaningless for any individual account. You are either on the good side of it or the bad side, and platform reporting will not tell you which.

The method is not exotic. Export the search terms over a long enough window to matter. Group them by what the buyer appeared to want rather than by the words they share, since string similarity and intent similarity are different things and n gram grouping quietly merges them. Then price the groups against whatever downstream outcome you actually have, even an imperfect one. What falls out is usually a cluster of terms with respectable platform conversions and almost no downstream value, sitting next to a small cluster carrying the account. Feed that judgment back through conversion values, structural splits and negatives used for steering rather than cleanup. This is the work we productized as query level paid search optimization, and it is the layer the bid strategy cannot reach.

The report you are exporting is also incomplete. A July 2025 analysis by Collin Slattery, twenty million dollars of spend across 933 campaigns, put hidden terms at 26.7 percent of search spend, and nobody has replicated it since, so read it as a reason to keep your own query history rather than as today's number.

What to Check Before the Recalibration Hits

Find the campaigns that are both limited by budget and currently running under their target. Those are the ones that will move, and in most accounts the list is shorter than people expect.

Pull the query mix on exactly those campaigns, not on the account. You are looking for what sits just outside what the budget currently reaches, because that is what the extra spend is about to buy.

Check whether your reported conversion values reflect anything downstream. If every lead is worth the same number in the account, the bidder has been told that every lead is worth the same number, and it has been optimizing accordingly for as long as that has been true.

Decide your targets deliberately before August 17 rather than discovering them afterward. A target set eighteen months ago against a different cost structure is about to become binding in a way it has not been, and Google has said it will not touch it for you.

And if anything in the account runs automatically created assets or campaign level broad match, that is a separate decision with a September 1 deadline, because the upgrade does not wait for an answer.

The Advertisers Who Win the Next Year Will Not Have a Better Bid Strategy

Bid strategy has commoditized. Everyone has access to the same auction time model, and the accounts running it well are separated by very little at that layer. Query judgment has not commoditized, because it requires knowing what a customer is worth to you after the click, and no platform has that information.

Two dates in the same month, both switched on by default, both moving the boundary of what your budget buys without moving the target you set against it. The system will spend to the number you gave it, and it will find the volume somewhere. Deciding where is still your job.

FAQ

Can Smart Bidding see the search term when it sets a bid? Yes. The query is one of the signals available at auction time, which is a different thing from the query being managed. The algorithm can factor a search term into a bid and still bid on terms you would never have chosen, because it is optimizing toward the conversion action you configured rather than toward your definition of a good customer. Negatives and a real search terms review remain the only way to take a term out of consideration.

Does the August 17 change affect campaigns that are not limited by budget? Google's stated scope is campaigns that are limited by budget and running a target based bid strategy. Campaigns that are not budget constrained are described as continuing to behave as they do today. App campaigns, video reach and video view campaigns are excluded, and so are Target Impression Share and manual CPC.

What happens on September 1 and do I have to do anything? Campaigns using automatically created assets or the campaign level broad match setting are upgraded to AI Max for Search automatically, with search term matching switched on, plus text customization on the automatically created assets campaigns. If you want to keep control of which queries are eligible, that is a decision to make before the date rather than after, since the upgrade proceeds without an opt in.

Should I lower my targets before August 17 to protect efficiency? Google's own guidance is to update targets to match recent performance if you want to maintain current results, since it will not do that for you. Whether you should is a business question rather than a platform one. A campaign that has been quietly beating its target has been telling you something about what that traffic is worth, and the right response might be a lower target or might be a larger budget.

Does adding negative keywords still help under automated bidding? It helps, and the reporting will not credit it. The bidder has no visibility into whether a query that converts occasionally goes on to close, so it cannot flag that spend as wasted. Pulling a long window of search terms and grouping them by recurring intent surfaces patterns that no in platform report will hand you.

Is manual bidding ever the right answer in 2026? Rarely, and almost never for the reason people give. The cases that hold up involve very low conversion volume where the strategy never meaningfully exits learning. Wanting more control over which queries you buy is a real instinct pointed at the wrong control.