- eBay’s 99.5% is a share of all clicks, and the estimate isn’t statistically different from zero. On paid clicks, the paper’s own footnote says about 1.5% was lost. [1]
- That agrees with the Bing experiment across 824 brands, which implies about 6.2% net new when no rival bids. [2]
- The real disagreement is Edmunds, which lost about 60% of paid clicks, against Golden and Horton, whose rival gained nothing from the top slot. [5][6]
- Your breakeven is CPC divided by the value of a click. The same net-new share can pay in one account and burn money in another.
- Google’s 89% covers all search ads rather than brand, and it comes from advertisers who chose to pause. [7]
- In our simulation, a four-week geo holdout can settle this for many mid-size advertisers, if you measure the right traffic.
The meeting goes the same way every time. Someone in finance has read that eBay turned off its brand ads and kept 99.5% of the traffic. Someone in paid search says competitors will steal the clicks, and quotes Google’s 89%. Both numbers are real, and neither one answers the question in the room.
Start with 99.5%. It’s in the Blake, Nosko, and Tadelis paper, and it’s the share of all eBay clicks from MSN, paid and organic together, retained after brand ads stopped in March 2012. The coefficient behind it has a t-statistic of 0.30, so by our calculation the loss of total clicks could plausibly sit anywhere from about -3% to +4%. On the denominator people assume they’re hearing, the paper has a footnote: about 1.5% of paid clicks. [1] The study’s US test on Google had no control group at all, and its German one read revenue across just 16 states.
Google’s 89% has its own problem. It comes from 446 studies of advertisers who chose to cut search spend by at least 95%, it covers all search ads rather than brand, and it counts clicks rather than sales. [7] Google’s later figure of 50% at organic rank 1 covers any query where the advertiser ranks first, and its fit excludes 58 of the 390 studies, those that came in below the line. [8]
One denominator
Put everything on a single metric: s, the share of your paid clicks that are net new, meaning clicks you’d have lost entirely without the ad. Then the map looks different from the debate.
| Situation | Net-new share of paid clicks | Basis |
|---|---|---|
| Alone on the page, very strong brand | 1.5%, up to about 11% by our interval | eBay on MSN, 2012, with Google as control [1] |
| Alone on the page, average advertiser | About 3.6-6.2%, and the paper calls its figure a bound | Bing, 824 brands, randomized, 2014 [2] |
| Low competition, measured on sales | 9% | Haus customers, vendor data, 2025 [9] |
| Rivals below you, if you lose the top slot | 22-38% | Bing, working paper, 2017 data [4] |
| High competition, measured on sales | 37% | Haus customers, vendor data, 2025 [9] |
| Peer rivals, brand and task queries | 57-61% | Edmunds, Bing and Yahoo, 2015 [5] |
Read the first two rows together. The Bing authors say their result is consistent with eBay’s. [2] The clash people cite doesn’t exist on a common denominator.
The real clash is further down. Edmunds switched off brand ads across half of 210 US markets and lost about 60% of its paid clicks, with no Google in the test. [5] Golden and Horton had a company shut off all Google search ads while a close rival moved up to the first slot on its name, and they could see the rival’s data. The rival’s clicks didn’t rise, and they ruled out a gain above 6% in signups. [6] That firm had only 4.3% brand awareness and the test lasted 21 usable days, so don’t read it as a law either.
Our best guess at what separates them is the query. Edmunds’ 6,587 keywords included task searches like “Edmunds used Honda Civic,” where a paid landing page may beat whatever ranks organically, while Golden and Horton describe navigational searches. Neither paper tests this, so treat it as inference. It does suggest splitting your pure name from your name plus a modifier before you decide anything.
Also weigh the newest evidence with care. A 2026 paper reports no significant sales effect from stopping brand search across eight small direct-to-consumer brands. [10] We could only read the abstract, and with sales as the outcome, a small brand’s null may simply be too noisy to mean much.
The number that decides it
Here’s why the map alone can’t settle your case. What matters is whether your share clears your breakeven, and that’s arithmetic: a paid click has to cost less than net-new share times what a click is worth.
The net-new share your brand clicks must reach to pay
Use your brand campaign’s CPC and its conversion rate, and the margin you actually keep on an order. The defaults are illustrative.
Model: CPC / (conversion rate × value × margin), the same algebra Google’s 2011 paper uses. [7] The shortcut is 1 / (reported ROAS × margin). A result above 100% means no share can pay. It assumes a click arriving through organic is worth as much as a paid one, and if your paid page converts better, your real bar is lower.
The same share gives opposite answers. All four accounts below are invented, to show the mechanics.
| Account | CPC | Value of a click | Breakeven | Result |
|---|---|---|---|---|
| Uncontested online store | $0.45 | $2.29 | 19.6% | At 6.2% net new, loses $20,514 a month on $30,000 |
| Store in a bidding war | $1.50 | $1.91 | 78.4% | At 61% net new, still loses $6,667 a month |
| B2B software | $4.00 | $96.00 | 4.2% | At 6.2% net new, makes $9,760 a month on $20,000 |
| Low-margin marketplace | $0.30 | $0.25 | 120% | Can’t pay at any share |
That second row is the one defenders of brand bidding should sit with. A rival raises your share of net-new clicks, and it raises your price too, because Google charges you what it takes to beat the ad below yours. [13] The Bing paper’s finding that defense pays rested on brand CPCs about ten times lower than other queries. [2] If yours aren’t, defense may not pay at all.
And it doesn’t have to be on or off. A bid cap of your estimated share times the value of a click keeps you present where you’re alone and cheap, and keeps you visible in Auction Insights, which shows nothing below 10% impression share. [14]
“But they might come in if we leave.” You can put a number on that too. Keeping a campaign as insurance makes sense only if the chance of entry exceeds (breakeven - share if alone) / (share if contested - share if alone). With a 19.6% breakeven, 6.2% alone, and 36% contested, that’s 45%. Since rival entry is visible and you can switch back within minutes, pausing and watching usually beats paying the premium.
Which way the signals point
| Signal | Points to | How strong |
|---|---|---|
| Breakeven under about 5% | Keep, even if you’re alone | Strong, it’s arithmetic |
| Breakeven over 40% | Cut or cap hard, even if contested | Strong, it’s arithmetic |
| No rival in Auction Insights, and you hold the top slot | Test a cut | Strong on clicks [1][2] |
| Rival takes the top slot during a small pause | Keep, at a cap | Medium, and Golden and Horton disagree [2][6] |
| You also sell on Amazon or through resellers | Measure marketplace sales too | Medium, vendor data only [9] |
| Brand plus a task or comparison | Test separately | Our inference from Edmunds [5] |
| Ambiguous name, not first organically | Keep | Our inference, since the Bing sample required first place [2] |
| Low awareness alone | No call | Contested [2][6][10] |
The test
Pick the rule before you see a result: cut if the top of the interval sits below your breakeven, keep if the bottom sits above it, and otherwise extend. If you can’t say that now, you’ll argue about it later.
Two things matter more than you’d expect, based on a simulation we ran with assumptions of our own, calibrated against the standard error Edmunds reported. First, measure the right traffic. If paid brand clicks are 40% of what you’re measuring, 210 US markets over four weeks give an interval of about ±7 points, which settles the question at a 19.6% breakeven. If they’re 15%, it’s about ±20, which doesn’t. So narrow organic to homepage and brand-page landings. Second, don’t lean on sales unless you have volume. Under about 5,000 conversions a week, the interval gets too wide to help, and at 1,000 it’s still about ±29 points over four weeks. Run a placebo check on your own data first.
Watch for what the traffic number hides. One Haus customer gained 1.36% on its own site from brand ads and lost 2.7% on Amazon, netting 0.26%. [23] Clicks on rivals’ ads can also be low quality: on Bing, 20-47% of them returned to search within 30 seconds, against 7% for the brand’s own link. [3]
If you’d rather not build this, Google’s Conversion Lift covers Search, with a stated minimum of 1,000 observed conversions and $5,000, though the page says it isn’t available for all accounts. [16] Just remember who holds the control group.
Going dark also raises a legal question: can a competitor bid on your name? Generally, yes. Google doesn’t restrict trademarks as keywords, even after a complaint, but it does restrict their use in ad text by a direct competitor. [17] In the US, Lerner and Rowe found 236 possible confusion calls across 109,322 impressions, or 0.216%, and the Ninth Circuit ruled for the bidder. [18] In the EU, the Court of Justice held that Google isn’t liable for selling the keyword and that an advertiser can be if users can’t easily tell who the ad is from. [19] This isn’t legal advice, and truce agreements between competitors carry their own antitrust risk.
Who’s telling you what
Look at who wrote each number. eBay’s staff wrote the paper that justified eBay’s cut. The Bing work was mostly completed at Microsoft, which sells these ads. [2] Google’s staff wrote the 89% and the 50%, and an academic estimate attributes 28-36% of Google’s search revenue to brand and navigational queries. [21] Haus sells the test it recommends. None of that makes them wrong. It does mean none of them should decide your budget, and Google’s own help page now pitches brand exclusions as a way to “avoid paying for clicks from users who are already navigating to your site.” [11]
Calculate your breakeven this week. It takes ten minutes, and you may find you don’t need the test at all, because at a 5% bar or a 60% bar most of the argument is already over.
Sources
- Blake, Nosko, and Tadelis · Consumer heterogeneity and paid search effectivenessEconometrica 2015, eBay field experiments; the authors worked at eBay
- Simonov, Nosko, and Rao · Competition and crowd-out for brand keywords in sponsored searchMarketing Science 2018, randomized on Bing, January 2014
- Simonov and Hill · Competitive advertising on brand search: traffic stealing and click quality, abstractMarketing Science 2021, click quality and stolen traffic
- Simonov and Hill · Working paper2018 draft, cost per incremental click with one to three rivals
- Coviello, Gneezy, and Goette · A large-scale field experiment to evaluate the effectiveness of paid search advertisingCESifo working paper 2017, Edmunds on Yahoo and Bing
- Golden and Horton · The effects of search advertising on competitorsManagement Science 2021, the only one observing the rival’s data
- Chan, Yuan, Koehler, and Kumar · Incremental clicks impact of search advertisingGoogle, 446 pause studies, all search ads rather than brand
- Chan, Kumar, Ma, and Koehler · Impact of ranking of organic search results on the incrementality of search ads (Google, March 2012)
- Haus · When is branded search worth the investment?vendor analysis of its own customers, September 2025
- Beknazarova and colleagues · Do brands need search advertising? Abstract via OpenAlexnine field experiments with eight small brands; we read only the abstract
- Google Ads Help · About brand exclusionsGoogle’s stated purpose for the control
- Google Ads Help · Apply brand exclusions to Performance Max or Search campaignsexclusions for Search sit inside AI Max, and how Performance Max applies them
- Google Ads Help · Actual cost-per-clickyou pay what beats the Ad Rank of the competitor below you
- Google Ads Help · Use auction insights to compare performanceno insights below 10% impression share
- Google Ads Help · Implement campaigns for geo experimentspre-test length, Presence targeting, and budget shifting
- Google Ads Help · Set up Conversion Lift based on usersminimum thresholds and availability
- Google Advertising Policies · Trademarkskeywords are not restricted, and competitor ad text can be
- US Court of Appeals, Ninth Circuit · Lerner and Rowe v. Brown Engstrand and ShelyOctober 2024 opinion on keyword bidding and confusion
- Court of Justice of the EU · Press release 32/10, Google FranceMarch 2010, on who is liable for keyword advertising
- Google Ads Help · About Performance Maxexact match keywords are prioritized over Performance Max
- Gleason and colleagues · Search engine revenue from navigational and brand advertising, abstract via OpenAlexUS panel estimate of revenue from navigational and brand searches; we read only the abstract
- Google Ads Help · Answering your top questions about Performance Maxword-in-brand traffic is blocked only if it clearly refers to the brand
- Haus · How Orthofeet used brand search testing to uncover hidden Amazon competitionnamed customer in a vendor write-up, 70/30 US split over two weeks
Frequently asked questions
Should I bid on my own brand name if no competitors are bidding on it?
Probably test a cut. Alone on the page, the best evidence puts net-new clicks in single digits of paid clicks: about 1.5% at eBay and about 4-6% across brands on Bing. That can still pay if your clicks are valuable enough, so compare it with your own breakeven before deciding.
What percentage of branded search clicks are actually incremental?
It depends heavily on the setting. Published estimates run from about 1.5% for a strong brand with no rivals, to 22-38% where rivals sit below you, to 57-61% at Edmunds with peer rivals on Bing and Yahoo. Most of this data comes from 2012 to 2017, and we found no public, controlled, brand-only estimate on Google that also observed the rival.
Did eBay really lose nothing when it turned off brand search ads?
The 99.5% figure is real, but it is the share of all clicks from MSN that eBay kept, and the estimate is not statistically different from zero. On the paid-click denominator, the paper’s own footnote says about 1.5% was lost. The US test on Google in July 2012 had no control group, and traffic fell 3.2%.
How do I calculate whether brand search ads are worth the money?
Divide your CPC by the value of a click, which is conversion rate times order value times margin. That gives the share of paid clicks that must be net new for the spend to break even. If your account reports ROAS, the shortcut is one divided by reported ROAS times margin. Then run a holdout to estimate your actual share.
Will competitors take my traffic if I stop bidding on my brand?
The evidence conflicts. Edmunds lost about 60% of paid clicks in a Bing and Yahoo test, and the authors think rivals may have taken about half. In Golden and Horton’s test on Google, a rival moved up to the first slot and gained no measurable clicks or signups. Rivals with low click-through rates may not capture much, and entry can be monitored and reversed.
How long should a brand search holdout test run?
Four weeks is a sensible default, with at least three times that as pre-test data, which is Google’s own guidance for geo tests. In our simulation, paid share of the measured traffic matters more than volume, so narrow your organic data to homepage and brand-page landings. Under about 5,000 conversions a week, sales alone is a weak primary outcome.
Can competitors legally bid on my brand name?
Generally, yes, if the ad itself is clean. Google does not restrict trademarks as keywords, even on complaint, though it does restrict them in ad text from a direct competitor. In the US, the Ninth Circuit found almost no confusion in Lerner and Rowe. In the EU, the Court of Justice held that Google is not liable and that an advertiser can be, if users cannot easily tell where the ad comes from. This is not legal advice.
Mikołaj Salecki
Writes about media, tech, and AI business for people who actually run digital. Former agency lead. Skeptic of frameworks that read better than they perform.
