The industry spent roughly five years building for a world without third-party cookies. That world did not arrive. What arrived instead was stranger and less discussed: the cookies stayed, and the replacements were cancelled.

That inversion is the actual measurement story of 2026, and most of the vocabulary still in use predates it. “Post-cookie” describes a transition that was called off. What practitioners are actually operating in is post–Privacy Sandbox, which is a different problem with a different shape.

What Was Cancelled, Precisely

On October 17, 2025, Google announced the retirement of the bulk of the Privacy Sandbox initiative. The list is worth reading in full, because its breadth is the point:

  • Topics API (Chrome and Android)
  • Protected Audience API / PAAPI (Chrome and Android)
  • Attribution Reporting API (Chrome and Android)
  • Private Aggregation, including Shared Storage
  • Protected App Signals
  • On-Device Personalization
  • IP Protection
  • Related Website Sets
  • SDK Runtime

Those were not peripheral experiments. Topics was the interest-based targeting replacement. Protected Audience was the remarketing replacement. Attribution Reporting was the conversion measurement replacement. The three functions that third-party cookies performed each had a designated successor, and all three successors were withdrawn on the same day.

A smaller set survived: CHIPS for partitioned cookies, FedCM for federated sign-in, Private State Tokens for fraud signals, and Google’s stated support for the W3C’s interoperable Attribution proposal. What survived are plumbing and fraud primitives. What was cut was the advertising functionality.

On the same day, the UK Competition and Markets Authority released Google from its Privacy Sandbox commitments, concluding that competition concerns no longer arose once cookie deprecation was abandoned. The CMA received 15 consultation responses. All 15 opposed the release. It proceeded anyway.

That detail matters beyond the procedural curiosity. The commitments regime was the mechanism through which the industry had formal input into browser-level advertising infrastructure. Its removal means the next change of this magnitude arrives without a consultation attached.

So What Is Measurement Actually Running On?

With the designated replacements gone and cookies intact but increasingly user-governed — Chrome having moved third-party cookie handling toward an explicit user choice rather than a silent default — measurement in Q3 2026 runs on a blend rather than a standard. In practice that blend is:

Observed events, where consent and browser state permit. Still the highest-fidelity signal, now covering a smaller and less representative share of traffic than it did in 2021.

Modeled conversions, filling the gap where observation fails. This is the largest change in practice and the least examined. Modeled conversions are estimates produced by the same platform that sells the media, and the modeling methodology is generally not disclosed at a level that permits independent validation.

Server-side collection, which moves the measurement boundary from the browser to the advertiser’s own infrastructure. This is genuinely more durable, and it is why the data-layer in-housing trend and the measurement question are the same conversation.

Panel and incrementality methods, including geo experiments and holdout testing, which never depended on identifiers and are consequently having a second life. Media mix modeling — written off as obsolete for a decade — is back in budget conversations for the same reason.

The honest summary is that measurement has moved from a shared standard to a set of vendor-specific estimates that do not reconcile to each other. Two platforms reporting on the same campaign will produce different numbers, and there is no longer an agreed method for adjudicating between them.

Read the Agenda, Not the Announcements

A useful way to see what the buying side is actually anxious about is to look at what practitioner-led events choose to program, since those agendas are set by people spending money rather than by people selling tools.

MediaPost’s Data & Programmatic Insider Summit — an invite-only format whose agenda is assembled by former marketers around in-house brand and agency practitioners — convenes at South Lake Tahoe on August 16–19, 2026. Its published agenda covers agentic AI, contextual strategy, inventory quality, retail media, CTV, attention metrics, and, in the organizers’ framing, the measurement frameworks that actually earn leadership trust.

That last phrase is the most revealing item on the list. It is not a technical topic. It concerns whether a CFO believes the number. When practitioner agendas start programming the credibility of measurement rather than its mechanics, the field has moved from a methods problem to a trust problem — and trust problems are not solved by another API.

Note also what is not on that agenda: identity resolution as a headline topic, and the successor-to-cookies framing generally. The industry has stopped waiting for a replacement.

What This Means for Planning

Three practical consequences follow, none of which require predicting anything.

Stop budgeting for a deprecation event. There is no date to plan against. Roadmaps built around a cookie sunset should be rewritten around consent rates and modeled-conversion share, which are the variables that actually move.

Instrument what you can own. Server-side collection, first-party identity resolution, and log-level access to your own campaign records are the parts of the measurement stack that cannot be retired by an announcement from someone else. The case for owning the data layer rests on exactly this.

Ask what share of your reported conversions are modeled. Not whether modeling is used — it is — but what proportion, and how that proportion has moved over the last four quarters. Most platforms will answer this if asked directly, and the trend line is usually more informative than the absolute number. A reported figure drifting steadily toward estimation is a different asset than the one your benchmarks were built on.

The transition everyone prepared for was cancelled. The one that actually happened was never announced as a transition at all, which is why so much planning language is still describing the wrong thing. Our earlier look at the cookie delays traced how the timeline kept slipping; this is what the end of the slipping turned out to look like.