For five years the retargeting conversation was really a conversation about a deadline. Third-party cookies were going away, Chrome would replace them, and every agency deck carried a slide about preparing for it. The deadline was called off. On 22 April 2025 Google confirmed it would not roll out a new prompt to phase out third-party cookies, and on 17 October 2025 it retired ten Privacy Sandbox interfaces outright. The replacement died before the thing it was meant to replace.
Key takeaways
- Chrome keeps third-party cookies; Privacy Sandbox APIs were retired on 17 October 2025.
- Safari and Firefox still block them by default, so reach was never universal anyway.
- Lambrecht and Tucker found dynamic retargeted ads underperform generic equivalents on average.
- Google Ads dropped its audience minimum to 100 active users across all networks in December 2025.
The deprecation that was called off, and what it does not fix
Third-party cookies remain available in Chrome with no announced end date. Google cited low adoption and ecosystem feedback when it retired Attribution Reporting, Topics, Protected Audience and seven other Privacy Sandbox interfaces across Chrome and Android in October 2025. Three narrower pieces survived: CHIPS for partitioned cookies, FedCM for federated login and Private State Tokens for anti-fraud signals.
None of that restores the coverage a 2015 retargeting plan assumed. Safari and Firefox block third-party cookies by default and have done for years, so a meaningful share of your audience was already invisible to a cookie-based list before any of this. Consent is the second constraint and the firmer one: under the GDPR and the ePrivacy rules, dropping a tracking cookie in the European Union requires consent that is freely given and revocable, and no browser decision changes that. Anyone who spent 2024 planning a cookieless future did not waste the effort. They just did the work for a different reason than the one on the slide.
The statistic that will not die
Nearly every piece written about retargeting repeats that retargeted ads are 76 percent more likely to be clicked than ordinary display ads. We went looking for the study. There is not one. The figure circulates from a marketing vendor’s blog post with no sample size, no time period, no advertiser set and no method, and it has been copied for a decade because it is convenient rather than because anyone checked it. This site is not going to keep it in circulation.
What can be said with a source is more interesting and less flattering. Anja Lambrecht and Catherine Tucker, publishing in the Journal of Marketing Research in 2013, ran a field test of dynamic retargeting, meaning ads that show the specific product a visitor browsed. Their finding: dynamic retargeted ads were, on average, less effective than their generic equivalents. The exception was narrow but instructive. When a consumer’s browsing suggested their preferences had actually formed, for example after visiting review sites, the dynamic version stopped underperforming.
Read plainly, that says showing someone the exact sofa they glanced at is not automatically the strongest creative. It works on people who have decided what they want and not yet decided where to buy it. Everyone else is being reminded of a preference they had not formed.
Why the measurement problem is bigger than the targeting problem
Retargeting has a structural flaw that has nothing to do with cookies: it targets people who were already going to convert at a higher rate than average, then takes credit for it. Every platform report will show it as your best-performing line item, and most of that gap is selection, not causation.
The scale of the error is documented. Brett Gordon, Florian Zettelmeyer, Neha Bhargava and Dan Chapsky compared observational measurement against randomised experiments using 15 large advertising studies at Facebook, covering roughly 500 million user-experiment observations and 1.6 billion ad impressions, and published the comparison in Marketing Science in 2019. Their conclusion was that the observational approaches commonly available to advertisers often fail to recover the causal effect that the experiments measured.
The practical answer is a holdout. Withhold retargeting from a randomly selected share of the eligible audience, then compare conversion between the two groups. It costs you a slice of a campaign that may or may not be working, and it is the only method that answers the question you are actually asking. If a platform’s attribution and your holdout disagree, believe the holdout.
What genuinely changed in the platforms
| Change | When | What it means for a campaign |
|---|---|---|
| Chrome keeps third-party cookies | 22 April 2025 | No forced migration, but no reprieve on consent or on other browsers either |
| Privacy Sandbox APIs retired | 17 October 2025 | Anything built on Topics or Protected Audience needs rebuilding on conventional targeting |
| Google Ads audience minimum cut to 100 active users in 30 days | December 2025 | Small sites can finally run remarketing on Search, Display and YouTube alike |
| Older customer lists keep the 1,000 matched user floor | Lists uploaded before 1 February 2024 | Re-upload a legacy list rather than waiting for it to qualify |
The audience minimum is the underrated one. A list of 100 active users is small enough that a modest site can build genuinely narrow segments, cart abandoners in one market, pricing-page visitors in another, instead of pooling everyone into a single list because the platform demanded volume.
How we would build one now
- Segment by intent, not by visit. Someone who reached a pricing page and someone who bounced off the homepage are different problems and should never share a creative.
- Exclude converters on day one. Nothing burns goodwill faster than advertising a product to the person who just bought it, and the exclusion takes minutes to set up.
- Cap frequency and rotate creative on a schedule, not when someone notices fatigue in the numbers. Wear-out shows up in brand perception well before it shows up in click-through.
- Test generic against dynamic rather than assuming the product-specific version wins, given what the field data says.
- Keep a permanent holdout, even a small one, so the incrementality question stays answerable at any point in the year.
Creative quality does more work here than most media plans allow for, which is a subject in itself and one we have covered separately in what actually stops a scroll. A well-segmented list carrying a forgettable banner is still a forgettable banner.
Where retargeting stops being the right tool
Retargeting is a paid solution to a problem that is sometimes free to solve. If a visitor left because the shipping cost appeared at the last step, no amount of remarketing on Google Ads, Facebook or Instagram fixes it, and the spend simply buys repeated exposure to the same objection.
It is also worth asking whether the person is reachable through a channel you own before you rent one. Someone who gave you an email address can be re-engaged for the cost of sending, which is why we keep pointing teams back to owned channels before they raise a retargeting budget. The honest version of this discipline is that it recovers a slice of demand you already created. It does not create demand, and treating it as a growth channel is how a line item quietly consumes a quarter of a media budget while reporting the best numbers on the sheet.
Before you rent the audience back, check the one you own
Re-engaging a subscriber costs the price of a send, and most teams under-invest in it while paying to reach the same person through an ad exchange.
Sources: Google, 22 April 2025, for the decision not to introduce a new standalone prompt to phase out third-party cookies in Chrome, and October 2025 announcements for the retirement of ten Privacy Sandbox interfaces including Attribution Reporting, Topics and Protected Audience, effective 17 October 2025, with CHIPS, FedCM and Private State Tokens retained. Anja Lambrecht and Catherine Tucker, “When Does Retargeting Work? Information Specificity in Online Advertising”, Journal of Marketing Research, volume 50, issue 5, 2013, pages 561 to 576, for the finding that dynamic retargeted ads were on average less effective than generic equivalents, and for the review-site exception. Brett R. Gordon, Florian Zettelmeyer, Neha Bhargava and Dan Chapsky, “A Comparison of Approaches to Advertising Measurement: Evidence from Big Field Experiments at Facebook”, Marketing Science, volume 38, issue 2, 2019, drawing on 15 advertising experiments, approximately 500 million user-experiment observations and 1.6 billion impressions, for the finding that commonly used observational methods often fail to recover experimentally measured causal effects. Google Ads Help and reporting on the December 2025 rollout, for the minimum of 100 active visitors or users in the last 30 days across Search, Display and YouTube, and for the 1,000 matched user requirement retained on customer lists uploaded before 1 February 2024. The frequently quoted claim that retargeted ads are 76 percent more likely to be clicked traces to a marketing vendor blog post without stated sample, period or methodology, and we have not reproduced it as a finding. Platform rules in this area change often; verify current thresholds before launching. Updated August 2026.
