Ask a marketer whether retargeting is dead and you will get a flinch before an answer. The flinch is the honest part. Everyone has watched the addressable pool shrink, watched Safari users vanish from their remarketing lists, watched the open-web display ad that used to chase a cart abandoner across the internet quietly stop chasing.
So here is the answer up front, before the explanation. The specific machine that powered retargeting for fifteen years, the third-party cookie, is decayed past the point of repair on most of the web. The thing that machine did, showing an ad to someone who already engaged with you, is not dead at all. It moved. It now runs on data you collect yourself, and on that footing it frequently performs better than the old version ever did.
Origin: a pixel, a cookie, and a trail across the web
Classic retargeting was a simple trick built on a quirk of how browsers stored state. You put a small piece of JavaScript, a pixel, on your website. When someone visited, the pixel dropped a third-party cookie in their browser and added that anonymous browser to a list. The cookie belonged to the ad tech vendor, not to your site, which is the part that mattered. Because the same vendor's cookies were present on millions of other sites, the vendor could recognise that browser again somewhere else entirely, and serve it an ad for the product it had looked at on yours.
That is the whole mechanism. A visitor browses your store, leaves without buying, then sees your ad on a news site, in a social feed, in search results, nudged back toward the checkout they abandoned. The cookie was the thread that tied your site to every other site.
It worked, and it scaled. Criteo, founded in Paris in 2005 and shipping its first product in 2008, built a multibillion-dollar business almost entirely on dynamic retargeting: not just any ad, but an ad showing the exact product the person viewed, priced and pictured. For a decade, retargeting was the line item nobody questioned. The dashboards looked spectacular. Hold that thought, because the dashboards are part of the problem.
Present: the machine broke, slowly, then all at once
The third-party cookie did not collapse on one date. It was strangled over eight years, mostly by browser makers who decided cross-site tracking was a privacy problem.
Apple started in 2017 with Intelligent Tracking Prevention in Safari. The early versions capped how long a third-party cookie could live and partitioned it; later ITP releases tightened that to outright blocking, and in March 2020 Apple shipped full third-party cookie blocking by default, with the WebKit team stating plainly that cookies for cross-site resources were now blocked across the board. Safari joined Firefox, which had already turned on similar protection. Brave never allowed them. Apple's app tracking rules then cut off most mobile advertising IDs, and consent banners across regulated markets gave a real slice of users a working opt-out.
The arithmetic is the point. Safari runs on roughly a sixth of browsers worldwide and a larger share in wealthy markets like the United States, per StatCounter's running tally. Firefox and Brave add a smaller piece. Add app environments and consent opt-outs and a large minority of the open-web audience simply cannot be cookied. Google kept third-party cookies alive in Chrome after years of promising to remove them, and then shut down its Privacy Sandbox replacement APIs in late 2025, so Chrome retargeting still functions. But a retargeting pool is only as large as the browsers you can actually tag. When a third of your would-be audience is invisible, the cart abandoner you most want to reach is, with uncomfortable odds, on a device you cannot see. Supermetrics, summarising the shift, notes that the cross-web tracking model that powered retargeting is breaking down into smaller, noisier pools.
That is the decay. Open-web, third-party-cookie retargeting reaches fewer people, recognises them less reliably, and measures them worse. On that narrow definition, yes, retargeting died.
The verdict, stated plainly
But that narrow definition was always a confusion of a tactic with its plumbing. Retargeting is not "third-party cookies." Retargeting is the decision to spend ad budget on people who have already shown interest rather than on strangers. That decision is sound. What changed is the data you use to act on it.
Three forms of retargeting work right now, and none of them depend on a third-party cookie.
The first is first-party retargeting, and it is where the action is. Instead of renting a vendor's cross-web cookie, you build the audience from data you own and the customer consented to give you: email subscribers, account holders, past purchasers, app users. You hash that list and upload it to a platform's own matching tool. Google's Customer Match and Meta's Custom Audiences both work this way, taking hashed identifiers and matching them against logged-in users. The audience is smaller than a sprawling cookie pool, but it is real. These are people who actually transacted with you, not anonymous browsers a data broker guessed were in-market. Google has lowered the Customer Match threshold so a Search campaign needs only 100 matched users to activate a list, which tells you how central first-party audiences have become to its system. Cleaner input tends to produce a better result, which is why first-party retargeting often beats the third-party version it replaced rather than merely surviving it. Building that audience well is its own discipline, covered in first-party data for advertising.
The second is retargeting inside the walled gardens, which never needed the third-party cookie in the first place. Google, Meta, Amazon and TikTok run on logged-in identity. When someone is signed into Facebook or YouTube, the platform knows exactly who they are without any cross-site cookie, and it maintains that identity and its targeting regardless of what browsers do. Retarget a website visitor through a Meta pixel or a Google tag and the match happens against an account, not a third-party crumb. This is a large reason budget keeps consolidating into the big platforms: their retargeting kept working while the open web's degraded.
The third is the set of channels that never touched programmatic display at all: on-site, email, and CRM retargeting. Showing a returning visitor a tailored on-site message, emailing a cart abandoner, syncing a CRM segment of lapsed customers into a paid campaign. These run on your own database and your own consent. They were always cookie-independent. They simply got more important as the alternative shrank.
For prospecting, the part retargeting was never meant to do, contextual and lookalike targeting carry the load. Contextual places ads by what a page is about rather than who is reading it, so it needs no identity at all and regained real standing as cookieless browsing became normal. Lookalike modelling takes your best first-party customers as a seed and finds new people who resemble them. Neither is retargeting. Both are the top-of-funnel complement that feeds the audiences you will later retarget.
The caveat that predates the cookie crisis
There is an older problem with retargeting, and it has nothing to do with privacy. Retargeting has always flattered itself in the reporting.
By design, retargeting shows ads to people already close to buying. Many of them would have bought with no ad at all. Under last-click attribution, the retargeting ad that appeared shortly before a purchase claims the credit, even when it changed nothing. The dashboard shows a glorious return. A good chunk of that return is conversions you already had.
This is not a hunch, it is a measurement finding. The cleanest demonstration comes from outside retargeting but proves the same trap. Economists Blake, Nosko and Tadelis ran a large field experiment at eBay, turning off paid search across 68 US markets and watching what happened. For brand keyword ads the effect on sales was, in the researchers' words, indistinguishable from zero: traffic the paid ads used to win simply arrived through free organic results instead. The work was published in Econometrica. The lesson generalises directly to retargeting. When you advertise to people whose intent already exists, clicks and conversions correlate with your ad without being caused by it.
The fix is incrementality testing. You hold out a random control group that sees no retargeting, run the campaign to everyone else, and measure the difference. That difference, not the attributed total, is what retargeting actually produced. eMarketer reports 52 percent of US brand and agency marketers now use incrementality testing, and that incremental return runs lower than reported return because it sets a higher bar. The same coverage notes Google cut the minimum budget for an incrementality experiment from roughly 100,000 dollars to about 5,000, so the test is no longer reserved for the largest advertisers. Run it. Retargeting that survives a holdout test is worth the money. Retargeting that does not survive one was buying conversions you already owned.
Future and impact: how to run retargeting now
Retargeting in a post-cookie world is narrower, more honest, and more demanding than the old version. A few principles make it work.
Build the audience from first-party data, not a rented pool. Your customer list, your subscribers, your logged-in visitors, captured with clean consent and kept fresh. That is the durable input, and it gets better as you invest in it.
Judge it on incremental lift, never on last-click. Hold out a control group, measure the gap, and accept that the honest number will be smaller than the dashboard's. A retargeting program you have not tested against a holdout is a program you do not actually understand.
Cap frequency and suppress converters. The fastest way to waste a retargeting budget is to keep paying to show ads to someone who already bought, or to show the same person the same creative until they resent it. Practitioner benchmarks land around three to five impressions per week for many consumer cases, lower for considered B2B purchases, and the moment someone converts they should drop out of the prospecting list and into a suppression list.
Treat retargeting as one stage, not the whole funnel. It re-engages demand that contextual and lookalike prospecting created. It cannot manufacture demand on its own, and a budget that is mostly retargeting is a budget mostly spending on outcomes it did not cause.
For decision-makers, the practical shift is from a rented capability to an owned one. Old retargeting was something you bought from a vendor with a cross-web cookie. New retargeting is something you build from your own data and your own consent, then activate across platforms that respect it. That is more work. It is also more defensible, because it does not sit on infrastructure a browser update can switch off.
So, is retargeting dead? No. The cookie that carried it is finished on most of the open web, and pretending otherwise wastes money. But retargeting was never the cookie. It was the idea that warm audiences deserve different spend than cold ones, and that idea is intact. It just runs on data you own now, gets measured by what it truly added, and rewards the marketers who stopped admiring the dashboard and started reading the holdout.
Council summary
This post argues that retargeting is not dead, but a confusion between a tactic and its plumbing has made it look that way. The third-party cookie that carried open-web retargeting for fifteen years is genuinely finished on Safari, Firefox and Brave, and the post is honest that pretending otherwise burns budget. Its real teaching is the distinction it draws: retargeting is the decision to spend on warm audiences rather than cold ones, and that decision now runs on first-party data, walled-garden identity, and owned channels like email and CRM, often more effectively than the rented version it replaced. The post also surfaces an older flaw the cookie crisis distracted from, that retargeting flatters itself under last-click attribution, and points the reader to incrementality holdouts as the only honest test. The takeaway for a decision-maker is concrete: rebuild retargeting on data you own, judge it on incremental lift rather than the dashboard, cap frequency, suppress converters, and treat it as one funnel stage rather than the whole plan.
Comments