Ask why a company pays for clicks from people who typed its own name, and the answer is almost always a number from the platform: brand campaigns convert at a fraction of the cost of everything else. That number is true and answers a different question. It counts the conversions that followed the click. It does not say how many of those people would have clicked the result directly beneath the ad, for free. That second number is the value of the line, and in most companies it has never been measured, because the brand budget was never run as an experiment.
The usual defense of brand spend is a counterfactual, and it is stated by the party that sells the clicks. I called that a Seller's Counterfactual in July, and brand search is the purest case I know of. It is also where credited and caused drift furthest apart, a point I made in passing in Break-Even ROAS Is a Floor. What has changed this month is that the question is finally being asked by the people holding the budget, and the answer is going to be decided in 2027 plans that are being drafted now. A test has to be designed before the result is known, which is what a tolerance band is for.
What moved in the first week of November
Three things, all reported second-hand and all worth stating carefully.
On November 2 Barry Schwartz at Search Engine Roundtable reported that Google had moved AI Overviews for many brand searches into the right-hand column, where a knowledge panel used to sit. His examples were US desktop results, and he cited no statement from Google. PPC Land, reporting a study from Ahrefs, said AI Overviews had appeared on 63 of 100 large brand names by the end of September, against 5 of 98 at the start of July. I have not read the Ahrefs study itself.
Around the same date, PPC News Feed spotted a sentence on Google's help page about how Performance Max interacts with other campaigns, as reported by PPC Land and by Relevant Audience: the campaigns "may also filter out navigational or brand-specific search queries" so that dedicated brand Search campaigns serve them, with advice to give those campaigns enough budget. When I read that help page on November 14, the sentence was not there. The page carries no date and no change note. Either the text was removed, or it lives on a version I was not shown. I am not going to tell you what Google's documentation says about your brand queries, because this week I could not pin it down myself.
And on November 2 Digiday reported that buyers are questioning Google search budgets openly. One agency founder put the amount in question at around 10 to 20 percent of the Google budget. One agency executive described 10 to 30 percent of search budgets being reallocated. Digiday also cited a 5 percent rise in branded cost per click in the third quarter, from Tinuiti. Those are estimates from three people, not a market measurement, but they are the people who sign insertion orders.
Put together, the page around your brand is changing, the instructions about who should serve your brand queries may have changed and changed back, and buyers are deciding how much of the line to keep. None of those three facts tells you whether your brand budget works. That is the point of this essay.
The concession
Brand search can be a very good buy, and in plenty of accounts it is. If competitors bid on your name, an absent brand ad hands them the top of the page at the moment of highest intent. Brand ads let you control the message, send people to the right page during a sale, and capture sitelinks and real estate. For some businesses, the incremental share of brand clicks is high, and paying for them is cheap insurance. For a company with no competitors on its name and a strong organic listing, a lot of that spend buys clicks it already had.
Neither case is the general one. That is exactly why the line needs a test rather than a rule. "Always run brand" and "never pay for your own name" are both claims about a counterfactual nobody in the room has measured.
Why it was never tested
Brand campaigns look like the safest line in the plan. The reported cost per conversion is the lowest in the account, the return is the highest, and nobody wants to be the person who switched off the campaign that "works." The evidence that it works is the platform's attribution. The evidence that would show whether it works, a holdout, has to be designed and paid for, and on the largest platform it is not freely available. Google's Conversion Lift page, which I read on November 14, says the product is not available to all Google Ads accounts and directs you to your Google representative. A geography-based version covers Search, and the page notes its budget requirement tends to run higher. I made the broader case about granted evidence in Who Is Allowed to Measure?.
So the line renews every year on the seller's word. Nobody chose that. It is a default owner at work: the decision about how much brand demand to buy back from the platform belongs, in practice, to whoever sets the campaign budget, and that person is graded on the very number that makes the line look good.
A brand budget that has never been tested is a subscription nobody signed.
Turning the argument into a test
A counterfactual audit of brand search does not need a vendor study. It needs a holdout you design, a band you set before the data comes in, and a person who agrees in advance to act on the result.
The holdout. Switch brand campaigns off, or cut them sharply, in a set of matched regions for four to six weeks while leaving them on elsewhere. Measure total brand-driven orders (paid plus organic plus direct) in both groups, from your own order system rather than the ad platform. If you cannot split by geography, alternate on and off by week in a single market, accepting a noisier read. Run it outside your peak season, and check first whether competitors bid on your name, because the answer differs sharply when they do.
The band. Before the test starts, write down three outcomes and what each will mean for the 2027 budget. Express each one as the share of the conversions the brand campaigns had been credited with that fail to show up in total orders once the ads stop. If more than, say, 60 percent go missing, the line is mostly incremental and stays. If fewer than 20 percent go missing, the line is mostly buying clicks you would have had, and it gets cut back to defensive coverage on the queries where competitors appear. Anything in between gets a smaller budget and a re-test in two quarters. The numbers are yours to choose. What matters is that they are chosen first.
The signer. Name the person who will move the money based on the result, and have finance co-sign the band. If nobody will commit before the data arrives, the test will be argued away whichever way it comes out.
Here is the arithmetic for an invented case, so you can check it on paper. A company spends $50,000 a month on brand search, and the platform credits 1,000 conversions, a reported cost of $50 each. The holdout shows that 70 percent of those customers arrived anyway through the organic listing or by typing the address. The incremental conversions are 300, and the real cost per incremental conversion is about $167. That might still be worth paying. It is a very different number to put next to a cost per cart or a cost per funded account than $50 is.
Google also offers a paid and organic report that shows how often your ads and your organic listing appear on the same searches. It is useful, and it describes overlap, not cause. It cannot tell you how many paid clicks would have gone to the organic result had the ad not been there.
If a holdout is impossible this quarter, there are weaker readings that are still better than none. Report brand and non-brand returns separately, every time, so the brand number stops flattering the blended one. Check how often competitors appear on your brand terms in the auction data, since that is the single biggest driver of whether brand clicks are incremental. And look back for accidents: a day the brand campaign ran out of budget, a week it was paused by mistake. Those are unplanned holdouts, noisy and short, and they are often the only causal evidence a company already owns.
The counterfactual just moved
There is a second reason to test now rather than lean on an old result, and it is the part of this I find most interesting.
The value of a brand ad is the difference between the page with the ad and the page without it. That page is not fixed. When an AI summary sits above your organic listing, the free click beneath the ad is harder to reach, and paying for the top slot may buy more than it did. When the summary moves to the side, as it reportedly did for many brand searches this month on desktop, the organic listing may get some of its prominence back, and the same ad may buy less. I do not know which way it nets out for any particular brand, and I doubt anyone outside Google does yet.
What I am fairly sure of is that it is not a constant. Brand incrementality is a reading taken on a particular page, on a particular date, with particular competitors in the auction. A holdout run in 2024 measured a results page that no longer exists. So the band you write down should carry its own expiry: re-test when the page around your brand changes materially, or every two quarters, whichever comes first. That turns a single argument about whether brand works into a standing decision with a review date, and in my experience that is the kind of decision that survives a budget cycle.
It also changes who should own the line. A setting that has to be re-measured whenever the seller rearranges the page is not a campaign setting. It belongs to whoever owns the forecast, because that person absorbs the error when the reading goes stale.
The claim
On February 28, 2027, Google's Ads documentation will still not offer advertisers a standard report that estimates the share of paid brand clicks that would have gone to their organic listing without the ad. Measuring that will still require a lift study arranged through a representative, or a holdout the advertiser designs.
If a self-serve report of that kind exists on that date, I was wrong, and I will say so here.
The brand line is a renewal decision, and right now it renews on the seller's evidence. Before the 2027 plan locks, give it an owner, a holdout, a band set in advance and a date to look again. If the test says brand search is earning its keep, you will defend it with your own numbers next time someone asks. If it says otherwise, you will have found money in the one place nobody wanted to look. The same test applies to every line the seller grades for you, which is the argument of The AI You Audit Is the AI You Own.