Should You Bid on Your Brand Name? A Geo Holdout Test

There are two positions on bidding for your own brand terms, and they are usually held with more conviction than evidence. One says you must defend the space, because if you go dark on your own name a competitor will bid on it and intercept customers who were looking for you. The other says you are paying for clicks you would have won for free, since you already rank first organically for your own name, so the spend is a tax on traffic you already own. Both arguments are reasonable. Both are almost always made from anecdote, a story about the one time a competitor appeared, or the one account where pausing brand made no difference.

The reframe that ends the argument is simple: you do not have to guess. Whether brand bidding adds incremental revenue, or merely cannibalises your own organic clicks, is a question you can measure directly with a geo holdout test. This article is the method, the design, the maths, the readout, so that the next time finance asks you to justify the brand line, you can answer with an experiment instead of an opinion.

What the argument is actually about

Underneath the whole debate sits one number: the cannibalisation rate. It is worth defining precisely, because most discussions talk past it. When someone searches your brand name, you typically already hold the top organic listing. If you also run a paid ad, some of the people who click the ad would have clicked the organic result anyway, arriving at your site at no cost. The cannibalisation rate is the fraction of paid brand clicks that would have converted through organic regardless. The higher it is, the more the paid spend is simply buying traffic you already had.

The reason the argument never resolves is that this rate is not a constant. It differs by brand strength, because a well-known brand with fierce loyalty has a high cannibalisation rate, its people will find it either way, while a weaker or newer brand may genuinely convert more when it reinforces the organic listing with a paid one. It differs by competitor presence, because if rivals bid on your name, the paid slot is defending against interception rather than duplicating organic. And it differs by the layout of the results page itself, since the more the paid ad pushes your organic listing down the screen, the more the two cannibalise each other. A number that swings on all three of those cannot be settled by a general rule, which is exactly why it has to be measured for your brand, in your market, against your competitors.

Test design

The instrument for measuring incremental effect is a geo holdout. You split the country into two sets of regions, keep brand ads running in one set (the control) and switch them off in the other (the test), and compare what happens to total brand revenue in each. If revenue holds up in the regions where you stopped paying, the spend was largely cannibalising. If it drops, the spend was incremental. The design is simple in principle and easy to get wrong in the details, so here is how to do it properly.

Start with region selection, and match on behaviour, not population. The test and control groups have to behave alike before you intervene, which means selecting regions that had similar brand traffic and brand revenue trends in the pre-period, not simply regions of similar size. Two cities of equal population can have very different brand demand, so matching on population alone gives you groups that were never comparable and a result you cannot trust. Match on the pre-period brand revenue and traffic curves, and aim for several regions in each group rather than one against one, so a single local anomaly cannot swing the outcome.

Then check parallel trends, the step most tests skip and the one that makes the result defensible. Before you switch anything off, plot the brand revenue of your intended test and control groups across the pre-period. The two lines should move together, rising and falling in step. If they already diverge before the test begins, the groups are not matched and any difference you see afterward is contaminated by that pre-existing gap. Only when the pre-period lines track each other closely do you have a clean baseline to measure against.

Next, the sample-size question, which is where most brand bidding tests quietly fail. A test can only detect an effect larger than its resolution, and that resolution is the minimum detectable effect. The MDE depends on how much brand traffic and revenue you have, how variable it is, and how long you run, and it sets a floor on what the test can prove. The relationship with spend is the part to internalise: the more brand revenue flows through the test, the smaller the effect you can detect, so a large brand can resolve a subtle few-percent cannibalisation while a small brand can only ever detect a large swing. Consider three cases at a fixed duration. A high-spend brand pushing substantial revenue through the test might resolve an effect of a few percent. A mid-spend brand might only reliably detect a difference in the region of ten percent. A low-spend brand might need the effect to be twenty percent or more before the test can see it against the noise. None of these is wrong, but you have to know your resolution before you start, because a test that cannot detect the effect you care about will return a false “no difference” and you will draw the wrong conclusion.

Duration is tied to the purchase cycle, not to a fixed number of days. The test has to run long enough to capture at least one full cycle from first brand search to purchase, plus a stable pre-period for the matching, because if your customers typically take three weeks to decide, a two-week test measures half a decision. For most consumer brands that means something in the range of four to six weeks of live test, on top of a clean pre-period. Longer cycles need longer tests.

Finally, hold everything else constant. The whole logic depends on the only difference between test and control being the brand ads, so anything else that moves brand demand has to be kept uniform across both groups for the duration. That means not launching a regional promotion in one set of markets, not concentrating PR or out-of-home in one group, and not letting other channels shift their geographic weighting mid-test. If you cannot hold something constant, at least record when it changes, because an uncontrolled variable you documented can be reasoned about, while one you did not notice invalidates the result silently.

What to measure

The single most common mistake in brand bidding analysis is measuring the wrong thing, and it always flatters the ads. The metric that matters is total brand revenue in the test regions versus the control regions, not paid brand revenue, and how your analytics credits that revenue across channels shapes what you can see, so it is worth understanding your attribution settings before you begin. The entire question is whether turning the ads off reduces the total you earn from brand demand, so the total is the only number that answers it.

Alongside the total, track organic brand clicks as the compensating variable, because this is where the cannibalisation shows up directly. When you switch off brand ads in the test regions, watch what happens to organic brand clicks there. If they rise to absorb most of the lost paid clicks, and total revenue holds, you have watched cannibalisation happen in real time: the demand simply moved from the paid listing to the free one. If organic does not compensate and total revenue falls, the paid clicks were incremental. The relationship between the two listings is the mechanism, so measuring both is what makes the result interpretable.

This is also why paid ROAS is the wrong readout, and a dangerous one. The return on ad spend of a brand campaign will almost always look excellent, because brand searchers are the warmest, highest-intent traffic you have, people already looking for you by name. They convert at high rates whether or not the ad is what brought them, so the paid campaign takes credit for revenue that the organic listing would have captured for free. A brand campaign with a stunning ROAS can be almost entirely cannibalising, and the ROAS figure will never tell you, because it only counts the clicks the ad was paid for, never the ones it quietly displaced. The Ehrenberg-Bass tradition of marketing science, and the effectiveness work of Binet and Field at the IPA, both point the same way: the number that matters is incremental effect on the total, not the flattering return on the slice you happened to pay for.

Reading the result

When the test finishes, the outcome takes one of three shapes, and each points to a different decision.

The first is full cannibalisation. Total brand revenue in the test regions holds steady even though you stopped paying, because organic brand clicks rose to absorb the lost paid ones. This means the brand spend in those conditions was buying traffic you already owned, and the honest reading is that you could redirect most of that budget with little revenue loss. It is the outcome finance tends to suspect, and the one worth confirming rather than assuming.

The second is partial cannibalisation, and it is the most common real-world result. Total revenue drops in the test regions, but by less than the full value of the paused clicks, because organic absorbed some but not all of the demand. Some of your brand spend was incremental and some was cannibalising, and the size of the drop tells you the split. This rarely justifies either extreme position, which is why both camps in the original argument are usually wrong.

The third is negative, meaning total revenue drops sharply when you stop bidding, by as much as or more than the paused spend. That points to real incrementality, and the usual cause is competitor conquesting: when you vacate your own brand term, rivals bidding on your name intercept the demand, so going dark does not just lose you the paid click, it hands the customer to a competitor. In that case the brand spend is genuinely defending revenue.

That competitor dimension is a variable to handle deliberately, not a footnote. Before and during the test, monitor who else is bidding on your brand terms in both groups. The awkward case is a competitor who begins bidding on you partway through the test, because it changes what you are measuring midway, turning a cannibalisation test into a defence test in the affected regions. Record it as an event with its date, and check whether it correlates with the point where test and control diverge. A divergence that starts the day a competitor appeared is telling you about competitive defence, not about baseline cannibalisation, and conflating the two leads to the wrong budget decision.

The scenarios where the test is not needed

A test costs time and a controlled reduction in spend, so it is worth knowing when you can skip it because the answer is already clear.

The first case is a regulated category with mandated landing pages. In some regulated sectors the paid ad has to route to a specific compliant page, or specific disclosures have to appear that the organic listing does not carry, so the paid slot is doing a compliance job the organic result cannot. There the ad earns its place regardless of cannibalisation, and the question is moot.

The second is a brand with a weak organic presence for its own name. The whole cannibalisation argument assumes you already hold the top organic slot for your brand, so the paid click might be redundant. If you do not own that slot, because your site is new, your brand name is contested, or your organic visibility is genuinely poor, then the paid ad is not duplicating a strong organic listing, it is the only reliable way to appear, and you should keep it while you fix the organic gap.

The third is a brand with an ambiguous name that shares its query with something else. If your brand name is also a common word or the name of another well-known thing, then a search for it is not a clean brand query; it is a mixed bag of people looking for you and people looking for something else. That muddies both the organic performance and the test, and it usually justifies keeping a paid presence to claim the searches that are genuinely for you, since you cannot rely on organic to disambiguate.

What to do with the answer

Assume the common result: partial cannibalisation. You now know that some of your brand spend is incremental and some is not, and the job is to keep the incremental part while stopping the waste. The move that follows is a defensive floor, not full coverage all year round.

A full-coverage strategy pays to appear on every brand search regardless of whether anyone is threatening the slot, which the test has just shown you is partly wasteful. A defensive-floor strategy instead protects against the situations where the spend is genuinely incremental, competitor interception, while not paying full price for the rest. In practice that means using impression-share targets set to hold a strong presence rather than a total one, so your ad shows reliably enough to defend against a competitor spike without buying every single brand impression at full cost across the whole year. You are buying insurance against the negative-outcome scenario, sized by what the test told you the incremental portion was worth.

The discipline here is to let the measured cannibalisation rate set the budget, rather than defaulting to nothing or everything. The two original camps argue for the extremes precisely because they never measured the middle, and the middle is where almost every real brand lands. Set the floor to the level the test justifies, and revisit it if your competitive situation changes, which is the cue to run the test again.

The cost of the test against the line it governs

The only real cost of running this test is the controlled revenue dip in the test regions during the holdout, plus the analyst time to design and read it. Weigh that against what it governs: the entire annual brand-search line, paid every month, often for years, on the strength of an assumption nobody has checked. A few weeks of partial holdout in a subset of regions is a small, one-off price for putting a recurring budget line on evidence instead of belief.

That is the real argument for the test. Not that brand bidding is good or bad, it is usually partly both, but that a recurring line item large enough for finance to question deserves to be governed by measurement rather than by whichever anecdote was told most confidently. Run the holdout once, set the floor it justifies, and you have replaced an argument that never resolves with a number you can defend. That is a measurement problem before it is a bidding one, which is the way we tend to approach it in our paid search work, and where it fits the wider plan is something we work through as part of digital strategy.

FAQs

How long should a brand bidding holdout test run?

Long enough to cover at least one full purchase cycle plus a stable pre-period for matching, which for most consumer brands means roughly four to six weeks of live test on top of a clean baseline. The rule is tied to your buying cycle rather than a fixed number of days: if customers typically take three weeks from first brand search to purchase, a two-week test only captures half a decision. Longer consideration cycles need proportionally longer tests to avoid measuring an incomplete journey.

What is a normal cannibalisation rate on brand search?

There is no single normal rate, which is exactly why it has to be measured rather than assumed. It varies widely with your brand strength, how strongly you hold the organic listing for your own name, whether competitors are bidding on you, and how the results page is laid out. A dominant, well-loved brand with no competitors on its terms may see very high cannibalisation, while a weaker or contested brand may find much of its brand spend is genuinely incremental. The honest answer is to run the holdout and find your number.

Will pausing brand ads hurt my organic rankings?

No. Paid activity is not a ranking factor, so pausing brand ads does not lower your organic position. What does happen is that total brand clicks can shift between the paid and the organic listing: when the ad disappears, some of the clicks it was receiving move to your organic result instead. That redistribution is the very thing the holdout measures, but it is a movement of clicks between your own listings, not a change to how Google ranks you.

What if a competitor bids on my brand during the holdout?

Record it as an event with its date, and check whether it correlates with the point where your test and control regions diverge. A competitor appearing partway through changes what the test is measuring in the affected regions, turning a cannibalisation test into a defence test, so you need to know it happened to read the result correctly. If the divergence starts the day the competitor appeared, the test is telling you about competitive defence rather than baseline cannibalisation, and you should treat those regions accordingly.

Can I run this test on a small budget?

Yes, with one honest limitation: a smaller spend raises the minimum detectable effect, so the test can only find larger effects. With less brand revenue flowing through the holdout, subtle differences get lost in the noise, and the test may only reliably detect a swing of twenty percent or more. That is still useful, it can catch a large cannibalisation or a strong incremental effect, but you should not read a null result on a small-budget test as proof that brand bidding has no effect, only that it has no large one.

Last reviewed: September 2026

This article provides general information about paid search measurement and testing. It is not specific marketing, financial or statistical advice for your business, and the right test design depends on your data, your market and your resources. Treat benchmark ranges as illustrative, and validate any test design against your own numbers before acting on the result.