What a brand-ad pause actually measures

One reason to question a brand-ad budget is that people may reach the same site through its organic result. One reason to defend it is that a rival may take the top paid position. Both mechanisms have evidence behind them. They do not give us a single percentage of brand spend worth keeping.

Start with two historical results. In Edmunds’ randomized 2015 brand-ad pause, organic gains offset about 43% of the measured paid-session decline. Total measured search-origin sessions still fell. In eBay’s 2012 Yahoo/MSN pause, an adjusted model implied about 99.5% retention of total referred clicks. That second percentage has a different denominator. Reconstructing it explains more than putting the two numbers side by side.

Paid down, organic up, total still down

Edmunds assigned 105 of 210 US geographic markets to suspend 6,587 keywords containing its name on Bing and Yahoo; the other 105 kept advertising. Competitors remained eligible to advertise. The pause ran from October 14 to November 5, 2015.

Coviello, Gneezy and Goette’s Table 2 puts each market’s estimated changes on the same baseline: its average daily total measured search-origin sessions before the experiment. Expressed per 100 of those baseline sessions, the unadjusted estimates are 9.8 fewer branded paid sessions, 4.2 more organic sessions, and 5.6 fewer total sessions, relative to markets assigned to retain the ads.

The organic rise was substitution in the aggregate accounting, but it was smaller than the paid decline. The remaining total loss was 5.6 per 100 baseline measured sessions; its approximate individual 95% normal interval is a loss of 2.3 to 8.9. The 43% offset and 57% unreplaced fractions are point estimates, without ratio confidence intervals.

Here “100” is a unit of scale, not a tracked group of 100 people. The study did not observe where the missing sessions went. They cannot be assigned to rivals, direct visits or lost purchases.

Nor was this complete elimination of paid traffic. Other Edmunds campaigns continued to produce some branded paid sessions. The experiment estimates the effect of assignment to suspend the specified terms, with that residual traffic present. Its organic-session measure also does not establish that every organic query contained the brand.

Accounting for baseline paid share changes the estimates to −10.2 paid, +4.0 organic and −6.2 total per 100 baseline sessions: about 39% offset. Partial replacement remains the finding under that specification. Larger estimated total losses in markets with higher baseline paid share describe heterogeneity; paid share itself was not randomly assigned. Design, uncertainty and calculation.

Why eBay’s 99.5% answers a different question

In March 2012, eBay stopped brand-containing paid terms on Yahoo/MSN, using Google brand traffic as a seasonal comparison. This was a platform pause, rather than the Edmunds geographic randomization.

The adjusted MSN coefficient for total paid-plus-organic referred clicks was −0.00529 in logs. Exponentiating it gives 0.9947, or about 99.5% total-click retention. Its approximate 95% normal interval for the percentage change runs from −3.9% to +3.0%. The point estimate is close to full retention; the uncertainty permits a loss as well as a gain.

A footnote supplies the other denominator: the roughly 0.5% total-click loss was roughly 1.5% of paid clicks. On those rounded quantities, about 98.5% of the paid-click decline was offset in aggregate. A small loss relative to all search clicks can be a larger loss relative to the paid component.

So, for an organic-substitution argument, take the paid denominator to the meeting: about 43% offset in the unadjusted Edmunds specification versus about 98.5% implied by eBay’s rounded footnote. Keep the design, site, period and measured unit attached. The differences establish variation; they do not identify which difference between the businesses caused it. eBay’s Table A1 and denominator note.

A pause also changes the rival’s opportunity

Removing an owner’s ad while rivals can remain above the organic result creates a different search page from removing all top ads. That distinction changes what a click estimate can tell us about brand defense.

In Simonov, Nosko and Rao’s 2014 Bing experiment, users were assigned different caps on the number of top paid ads. Cap zero removed all top ads. The 824 brands that usually advertised on their own names also had the owner in the first organic position. Moving from cap zero to cap one added 2.27 owner clicks per 100 searches, while the owner ad received 36.4 paid clicks per 100 searches. The extra owner clicks were about 6.2% of that paid volume.

This is strong evidence of organic crowd-out in that contrast: much of the ad’s traffic would otherwise have reached the owner. But its ads-off condition also removed competing top ads. It does not directly measure an owner switching off while leaving rivals above it.

A different set of 181 brands, which did not bid on themselves, shows the rival opportunity: with a competitor in the top paid slot, competitors received about 17–18 clicks per 100 searches across the study’s organic-click-rate groups. The 824 self-bidders and 181 non-self-bidders are different populations. Combining their results into one matched “with our ad / without our ad” comparison would erase that selection. Bing caps and populations.

Simonov and Hill’s 2021 final publisher abstract gets closer to the question of removing the owner’s top paid link. Competitors collectively received 6–15% of brand-query traffic when that link was removed, versus 1–2% when present. Among rival clicks, 20–47% returned to Bing in under 30 seconds, versus 7% after owner-link clicks. These ranges describe conditions, not confidence intervals.

A quick-back is an imperfect success proxy; staying away for 30 seconds is not a recorded purchase. These results come from the final publisher abstract; full-text methods were unavailable. The earlier preprint remains a separate edition. Final abstract and edition boundary.

There is also a useful counterexample to the claim that a better rival position guarantees a gain. Golden and Horton studied two competing service marketplaces in March 2014. One randomized all Google search ads off in half of 210 US markets. The other’s ad on the first firm’s brand query moved up about 0.931 positions, approximately second to first. Yet the study detected no positive gain in that rival’s brand-ad clicks or overall registrations.

The rival-click estimate was negative, with an approximate 95% normal interval spanning −34.1% to +5.0%. Its registration estimate was about +0.9%, with an interval of −5.2% to +7.4%. Those results constrain the observed gain; they do not prove exact zero or exclude later, unmeasured transfer. Twenty-one days were usable after exclusions. Marketplace intervention and endpoints.

The separate question: did business value change?

A lost owner session and a rival click are traffic outcomes. Even a measured registration stops short of profit or lifetime customer value.

There is, however, a brand-only business endpoint in the eBay paper. In January 2013, eBay suspended Google ads for the keyword “eBay” in eight randomly assigned German states and retained them in the other eight. The outcome was the total dollar value of goods purchased by eBay users—gross transaction value, not booked fee revenue or profit. This is a separate experiment from the 2012 MSN click pause.

Table A3 reports the effect of ads being on under three time-control specifications. Each point estimate is near zero, and each is too uncertain to distinguish from zero. The negative signs mean slightly lower estimated transaction value with ads on; reading them as sales losses caused by pausing would reverse the contrast.

Percentage effects and intervals are Cascader calculations from the log coefficients and state-clustered standard errors in Table A3. The intervals use a normal approximation; a short window and only 16 states limit precision. Coefficients and calculation.

This is evidence against a clearly detected transaction-value benefit in that historical eBay test. It is also compatible with modest effects in either direction. It supplies neither a universal zero-sales result nor a precise return on the German ad spend, which is not disclosed.

Golden and Horton measured another business endpoint: the owner’s overall new registrations fell by about one fifth when its search ads were off. But that treatment stopped brand and nonbrand ads together. The owner-registration loss cannot be assigned to brand ads alone. The German eBay experiment is closer to a brand-only value claim; the marketplace experiment is closer to an all-search-acquisition claim.

Where Google’s large pause studies fit

These click studies return to the replacement question: was the owner’s organic route available, and where?

Google’s 2011 study reported a paid-click-weighted 89% incremental-ad-click ratio across 446 selected advertiser pauses. The ratio was modeled loss of total paid-plus-organic clicks divided by modeled paid-click loss. These were observed spend pauses, screened for a drop over 95% and model quality; about 55% of candidate pauses passed. They mixed query types and measured clicks.

A different 2012 pause set helps explain why organic availability matters. Its fitted click-incrementality was about 50% with a same-domain organic result in rank one, versus about 99% without one on the first page. Organic rank was not randomized, and ad position varied with it.

Their selected, modeled, mixed-query results are useful context for search advertising. They cannot supply a randomized brand-only customer-value effect. Google studies, samples and model.

Take the comparison that answers the claim

For “the organic result will replace our paid traffic,” the Edmunds arithmetic and eBay footnote let you compare replacement relative to the paid traffic removed. The point estimates differ substantially, and the original conditions travel with them.

For “rivals will take that traffic,” use evidence that preserves rivals’ eligibility when the owner’s top paid link disappears. Simonov and Hill’s final abstract is relevant to that opportunity; the Bing all-top-ads-off comparison answers a different counterfactual. Golden and Horton show why a rival’s improved position still needs its own click and registration measurement.

For “the brand budget creates customer value,” eBay’s German experiment gets as far as brand-only gross transaction value, with noisy estimates near zero. The marketplace registration loss belongs to an all-search-ad pause. Neither a traffic ratio nor quick-back behavior completes the value claim.

What would materially advance the budget question is a matched brand-only owner-off comparison with rivals still eligible, measuring the owner’s business outcome across firms and retaining spend and value information. That is an evidence requirement, not a pause recommendation for an account. The historical originals here do not observe that whole relationship.

Return to the originals

These notes preserve the editions and calculations behind the comparisons above. PDF page numbers count from the first page of the linked file. The source-edition cutoff is September 25, 2026, 03:36:57 UTC; the results describe historical interventions, not current platform performance.