What if you stopped all paid ads for 90 days? What the data says

eBay switched off brand search ads and kept 99.5% of the traffic. Brands that went dark for a year lost 16% of sales. Both are true, and the difference is your business.

Large digital billboards at night in Times Square, New York
Photo: Kevani for Wikimedia Foundation, CC BY-SA 4.0, via Wikimedia Commons

If you stopped all paid ads for 90 days, the spend that only harvested people already looking for you would barely be missed, and the spend that found new customers would be. eBay paused its brand-keyword search ads and kept 99.5% of the traffic through free listings, while Google's own pause studies found 89% of paid search clicks were not replaced. Over 90 days most brands lose a slice of revenue that grows week by week; over a year, Ehrenberg-Bass found brands that went dark lost 16% of sales on average, and small brands lost more.

Key takeaways
  • eBay kept 99.5% of brand-search traffic after it stopped bidding on its own name.
  • Google's 400+ pause studies found 89% of paid search clicks were not replaced by organic ones.
  • Uber cut about two-thirds of a roughly $150 million ad budget with no real drop in sign-ups.
  • Brands that went dark lost 16% of sales after one year and 36% after three (Ehrenberg-Bass).
  • With these example inputs (weekly carryover 0.95), the ad-driven sales lost average only 29% over the 13 weeks, with about half the effect still left at day 90, so short tests understate the loss. Calculated in this post from the adstock formula (Robyn docs) and a weekly carryover of 0.95, an example input and not a measured figure.

Every founder who has watched a ROAS dashboard has had the thought: what if half of this is credit for sales that would have happened anyway? It is a fair suspicion, and the field experiments say it is often right. One company tested it at a scale few others could.

The story

In 2017, the watchdog group Sleeping Giants criticised Uber for ads appearing on Breitbart. Uber cut about $15 million of ads in response and saw no meaningful fall in new users.

Kevin Frisch, then running acquisition, kept digging. Some ad networks reported more daily clicks than they had active users; some apps watched what people searched in the app store and claimed credit for installs that were really organic. Uber went on to switch off about two-thirds of a roughly $150 million budget with no real drop in sign-ups, and launched what The Hustle calls the first major ad-fraud lawsuit. Frisch later told the story on the Marketing Today podcast.

The Hustle: Uber was swindled out of $100m in ad spend and no one is talking about it

Uber is the extreme case. Pausing is also how brands talk themselves into a slow decline. The useful answer is not "stop" or "don't", but which ads, for how long, and how you would know.

99.5%of eBay's brand-search traffic retained after it stopped bidding on its own name
89%of paid search clicks not replaced by organic, across 400+ Google pause studies (2011)
−16%average sales after a year without mass-media ads (Ehrenberg-Bass)
₹71,621 crIndia's digital ad spend in 2025, 59% of all ad spend (dentsu)

What happens to brand search when you stop paying?

Brand search ads, the ones that appear when someone types your name, are the safest spend to cut: most of those clicks move to your free listing. The cleanest evidence comes from eBay.

Consumer Heterogeneity and Paid Search Effectiveness: A Large-Scale Field ExperimentThomas Blake, Chris Nosko and Steven Tadelis · Econometrica 83(1), 155–174 · 2015 (NBER Working Paper 20171)

eBay stopped bidding on brand keywords such as "eBay" on Yahoo and MSN (now Bing). About 99.5% of the lost paid clicks came back through natural search links. eBay then stopped all non-brand bidding across roughly 30% of US metro areas for about two months. Naive regression said paid search returned over 4,100%; the experiment found the whole paid search programme added only 0.66% to sales (95% interval −0.42% to 1.74%), for an estimated ROI of −63%. Ads did influence new and infrequent buyers, but frequent eBay users, who would have bought anyway, absorbed most of the spend. The limit: eBay is a household name; a brand nobody knows gets far more from ads.

Read the paper (free)

The same pattern shows up in newer tests. Haus, which runs geo-holdout experiments for brands, reported in September 2025 that branded search produced significant lift for 82% of brands facing heavy competitor bidding but for only 35% of brands with little competition.

If rivals are not bidding on your name, your brand ads are mostly a toll you pay to yourself.

What happens to non-brand search and prospecting?

Non-brand search and prospecting ads, the ones that reach people who have not heard of you, are where stopping hurts: most of those clicks are not replaced. Google's statisticians studied over 400 advertisers who paused search campaigns and estimated that, on average, 89% of paid clicks were incremental, meaning organic clicks did not make up for them.

A 2012 follow-up split that by organic rank:

  • 50% of ad clicks were incremental when the advertiser already ranked first organically;
  • 96% were incremental when it ranked fifth or lower.

Read those two studies together and the rule is simple. The better your organic position and the stronger your brand, the less your search ads add. The newer and less visible you are, the more of your traffic the ads are actually creating. A D2C brand two years old is not eBay.

Type of paid spendEvidenceWhat a 90-day pause likely does
Brand search, no competitors biddingeBay: 99.5% of traffic retained; Haus: lift in 35% of low-competition brandsLittle change; save most of it
Brand search, competitors biddingHaus: lift in 82% of high-competition brandsRivals take some of your demand
Non-brand searchGoogle: 89% of clicks incremental on average (96% if you rank 5th or lower)Real traffic and sales loss
App-install and affiliate networksUber cut most of its install spend with little change, after finding attribution fraudOften nothing, if the "conversions" were fake
Brand building (TV, video, social reach)Ehrenberg-Bass: sales index 84 after one year darkSlow decline that shows up after the 90 days
From harvesting demand to creating itHarvestsCreatesBrand search,no rivalseBay: 99.5%came back freeInstall andaffiliateUber: two-thirdscut, no changeBrand search,rivals biddingHaus: lift for82% of brandsNon-brandsearchGoogle: 89%incrementalBrand reach,TV, videoEhrenberg-Bass:−16% in a yearPause: mostly saves moneyPause: loses customersOrder is a judgement from the studies cited, not a measured scale
The table above as a spectrum. The further right a rupee sits, the more a 90-day pause costs, and the later the bill arrives.
Digital advertising screens on buildings in Times Square
Reach media rarely shows up in last-click dashboards, which is why it is the first thing cut and the last thing missed.Photo: Kevani for Wikimedia Foundation, CC BY-SA 4.0, via Wikimedia Commons

Why do sales fall slowly instead of all at once?

Sales fall slowly because advertising has carryover: an ad seen today still nudges purchases for weeks, and that memory decays rather than switching off. Marketing mix models capture this with an adstock, a running total of past advertising that loses a fixed fraction each week.

\[ A_t = x_t + \lambda A_{t-1}, \qquad \text{after the stop: } A_t = A_0\,\lambda^{t}, \qquad t_{1/2} = \frac{\ln 0.5}{\ln \lambda} \]

Here \( x_t \) is spend in week \( t \) and \( \lambda \) is the weekly carryover. Two cases show the range:

  • \( \lambda = 0.7 \): the half-life is about two weeks and, averaged over 13 weeks, about 82% of the ad-driven sales are gone.
  • \( \lambda = 0.95 \), an example of a slow brand effect: the half-life is about 13 weeks and, averaged over 13 weeks, only 29% of the ad-driven sales are lost (about half the effect is still left on day 90). Calculated in this post from the adstock formula (Robyn docs) and a weekly carryover of 0.95, an example input and not a measured figure.
100%50%0024681012Weeks after the ads stop (13 weeks = 90 days)λ = 0.7: half-life about 2 weeksλ = 0.95: half-life about 13 weeksShare of ad memory left after the stop
Geometric adstock drawn from the equation above. A fast-decaying response campaign is nearly gone inside the test; a slow brand effect has barely started to fade by day 90.

That is the trap: a 90-day test of brand advertising mostly measures the period before the damage arrives. I explain how these models are fitted in marketing mix modelling with Meridian and Robyn.

Worked example: the ₹1 crore brand

Use the calculator to check what you just read with your own revenue, spend and margin.

Should you pause ads for 90 days?

Work the default example, with these example inputs. A brand with ₹1 crore a month in revenue spends ₹15 lakh a month on ads, 30% of revenue is truly ad-driven and the margin is 40%.

  1. Decay factor. \( \phi = \tfrac{1}{13}\sum_{k=1}^{13}\lambda^{k} = 0.178 \) for \( \lambda = 0.7 \), so 82.2% of ad-driven sales are lost on average. Calculated in this post from the adstock formula and the example inputs above.
  2. Revenue lost. \( 3 \times ₹100\text{ L} \times 0.30 \times 0.822 = ₹74\text{ L} \) over 90 days. Calculated in this post from the adstock formula and the example inputs above.
  3. Margin lost. \( ₹74\text{ L} \times 0.40 = ₹29.6\text{ L} \). Calculated in this post from the adstock formula and the example inputs above.
  4. Spend saved. \( 3 \times ₹15\text{ L} = ₹45\text{ L} \). Net: about ₹15 lakh better off, for now. Calculated in this post from the adstock formula and the example inputs above.

The general rule falls out of the algebra. Over the pause, the margin you lose per rupee of revenue is \( s\,m\,(1-\phi) \); the money you save per rupee of revenue is your spend ratio. Pausing pays, in the short run, when the spend ratio is larger.

\[ \frac{\text{spend}}{\text{revenue}} > s\,m\,(1-\phi) \quad\Rightarrow\quad 0.15 > 0.30 \times 0.40 \times 0.822 = 0.099 \ \checkmark \]

Spending 15% of revenue to protect 9.9% of revenue in margin is a bad trade, so in this example the pause wins by about ₹15 lakh. Raise the truly ad-driven share to about 46% and the two sides balance; above that, the pause loses money even inside 90 days.

Calculated in this post from the adstock formula and the example inputs above (0.15 against 0.30 x 0.40 x 0.822 = 0.099; break-even at 0.15 / (0.40 x 0.822) = 46%).

What did brands that went dark for a year lose?

Brands that stopped mass-media advertising for at least a year lost on average 16% of sales after one year, 25% after two and 36% after three, according to the Ehrenberg-Bass Institute. Big brands that were already growing held up; small and declining brands fell fastest.

When brands go dark: what happens to sales when brands stop advertisingAdam Gelzinis, Rachel Kennedy, Virginia Beal, Nicole Hartnett and Byron Sharp, Ehrenberg-Bass Institute · Journal of Advertising Research (early access) · 2021

The team tracked 70 competing brands in an Australian consumer-goods category for over 20 years and found 57 cases where a brand cut all mass-media advertising for at least a year. The average sales index fell to 84 after one year, 75 after two and 64 after three. Small brands (under 250,000 units a year, n=23) declined fastest. Brands already declining before the stop roughly halved over two years, while big and medium brands that were growing kept growing for one to two years. The limit: one country, one category, and the average is unweighted, so a few small brands falling sharply pull it down.

Read the summary (free)
Sales index after brands stopped mass-media advertising
See the data as a table
Average brand (57 cases) (index)
Year 0100
Year 184
Year 275
Year 364

Source: Ehrenberg-Bass Institute, Gelzinis et al. · Australian consumer goods category, 20+ years of data. Small and declining brands fell faster than the average.

This is what a 90-day experiment cannot see. Your dashboard after three months may look fine, because the ads you cut were partly harvesting, and the decay of memory structures has barely begun. The bill comes in quarters four to eight. I make the longer argument in brand vs performance marketing.

Is any of your ad spend simply fake?

Sometimes a large share is, and a pause is the cheapest way to find out. The best-known case is Uber, in the story above. Its former head of acquisition, Kevin Frisch, found that much of the app-install advertising he was paying for was claiming credit for installs that would have happened anyway; reports put the budget Uber could cut at two-thirds to three-quarters of online spend, with next to no change in rider sign-ups.

The mechanism was attribution fraud: apps on Android phones fired fake clicks so that ad networks got paid for organic installs.

The opposite case: Airbnb

Airbnb is the opposite kind of example. During the pandemic it cut marketing hard, and its CMO later said traffic had recovered to about 95% of 2019 levels without the old performance spend. The company did not stop advertising; it moved money from search and performance into brand campaigns.

A household name with 95% of traffic arriving directly is a very different business from yours, unless you are a household name.

A large building-sized digital billboard in New York
Brands that can afford to go quiet are usually the ones whose names are already on every screen. The evidence on going dark is kindest to big, growing brands.Photo: Kevani for Wikimedia Foundation, CC BY-SA 4.0, via Wikimedia Commons

How should you test a pause without betting the company?

Run a holdout, not a full stop: pause in some regions or for some audiences, keep the rest live, and compare. eBay, Bose and the Haus experiments all used geographic splits for exactly this reason.

  1. Start with brand search. Pause it in half your cities for four weeks and compare total brand traffic and orders, not just paid clicks.
  2. Audit install and affiliate partners. Cut the bottom third by quality and watch organic installs and sales; this is where fraud hides.
  3. Hold out prospecting by geography. Stop non-brand search and cold social in matched regions for six to eight weeks.
  4. Keep brand reach running. Do not cut video, TV or reach campaigns in the same test, or you will not know what caused what, and the damage arrives after your test ends.
  5. Read the result in incremental revenue. Platform ROAS will mislead you; compare test and control regions directly.

For Indian D2C, the arithmetic of acquisition also changes the answer: if contribution margin after returns and COD failures is thin, the share of revenue your ads must truly create to pay for themselves rises sharply. See CAC payback and LTV for Indian D2C.

So, should you stop all paid ads for 90 days?

Stopping everything at once is the worst version of the experiment: you learn least and risk most.

  • Often free money: brand search with no rival bidders, and unverified install networks.
  • Costs it back with interest: prospecting and brand reach, which save cash for a quarter and lose it over the next two years.

Across India, digital already takes 59% of ₹1.21 lakh crore in ad spend in 2025 (dentsu); much of it is harvesting, and some of it is the only reason new customers ever hear your name.

What if next?

Questions people ask

Should I stop bidding on my own brand name?

Test it rather than guess. If competitors are not bidding on your name and you rank first organically, the eBay and Haus evidence suggests most brand clicks will come back free. If rivals bid on your brand, pausing often hands them customers who were looking for you.

How long do the effects of advertising last after you stop?

Direct-response effects fade within weeks, while brand effects decay over months to years. The Ehrenberg-Bass data show average sales still falling three years after a brand went dark, which is why short pause tests understate long-term loss.

Why did my sales not drop when I paused Meta ads?

Several reasons can coexist: the ads were reaching people who would have bought anyway, carryover is still supporting sales, other channels picked up the slack, or attribution was over-crediting the ads. Watch the next two to three months, and use a regional holdout next time.

What is a good test to measure if ads are working?

A geographic holdout: switch ads off in a random set of matched regions and compare sales with regions where ads stay on. It measures incremental sales directly, unlike platform-reported conversions.

Sources

  1. Blake, Nosko and Tadelis, NBER/Econometrica: Consumer heterogeneity and paid search effectiveness (2014–2015; source for 99.5% retention, 0.66% sales effect and −63% ROI)
  2. Google Research: Studies show search ads drive 89% incremental traffic (2011; source for 400+ pause studies and 89%)
  3. Google Research: Impact of organic ranking on ad click incrementality (2012; source for 50% and 96% by organic rank)
  4. Ehrenberg-Bass Institute: What happens when brands stop advertising (2018 report, JAR 2021; source for the 84, 75 and 64 sales indices)
  5. Haus: When is branded search worth the investment? (Sep 2025; source for 82% vs 35% lift)
  6. Adweek: Bose has paused paid search in half its US markets (2025)
  7. The Hustle: Uber was swindled out of $100m in ad spend and no one is talking about it (source for the 2017 Breitbart trigger, the first $15 million cut, the two-thirds of ~$150 million and the Marketing Today podcast)
  8. Inverse: Uber burned through $100 million thanks to digital ad fraud (Jan 2021; source for the Uber attribution-fraud case)
  9. The Drum: Airbnb CMO on ditching performance marketing (Aug 2023; source for traffic at 95% of 2019)
  10. Social Samosa: Digital becomes India's largest ad channel, dentsu report (Feb 2026; source for India ad spend figures)
paid adsincrementalitybrand searchadstockEhrenberg-BasseBay experimentgoing dark
Sapta

Sapta (Saptarshi Sarkar) is a mechanical engineer turned marketer in Bengaluru and a solo touring rider. He writes the seven Saptaverse blogs. Work with him or read more of Marketblog.

Keep reading in Marketblog

From the other blogs