Three years after Google’s third-party cookie deprecation was supposed to be the defining moment of digital advertising, the cookieless future has arrived in a much messier form than predicted. Third-party cookies are technically still around in Chrome, though deprecated and unreliable. Safari and Firefox have been cookieless for years. Mobile attribution has its own privacy regimes. The attribution stacks running in 2026 reflect this messy reality.

Here’s an honest look at where cookieless attribution actually sits this quarter, and what’s working in production.

The current state of the cookie

Worth being precise about what’s actually true in 2026. Third-party cookies still work in Chrome but with significant restrictions and user controls that have eroded coverage to maybe 40-50% of traffic depending on segment. Safari has been blocking by default for years - Intelligent Tracking Prevention has continued to tighten through 2025. Firefox is similar.

The result: third-party cookie data exists but covers a minority of traffic and skews toward users who haven’t actively engaged with privacy controls. This is a bad demographic to optimise for - typically older, less tech-engaged users.

The practical implication is that any attribution model relying primarily on third-party cookies is making decisions based on partial and biased data. Most serious attribution programs have already shifted away from this dependency.

What’s actually working in production

Several patterns have emerged as workable cookieless attribution approaches in 2026.

First-party data infrastructure with identity stitching. The teams doing this best have invested heavily in their own first-party data - CDP implementations, server-side event capture, identity resolution across logged-in sessions. This isn’t cheap to build but it produces durable attribution capability that doesn’t depend on browser cookies.

Server-side event sharing with platforms. Sending conversion events server-to-server (via Conversions API for Meta, Enhanced Conversions for Google, similar APIs across platforms) has become standard. The match rates are imperfect but meaningfully better than relying on browser-side tracking alone.

Marketing mix modelling. MMM has been the biggest beneficiary of the cookieless shift. Aggregate-level modelling doesn’t depend on user-level tracking, so MMM works regardless of cookie state. The renaissance of MMM that started around 2023 has continued and accelerated.

Incrementality testing. Geo-based holdouts, audience holdouts, and structured experimentation have become the gold standard for measuring channel effectiveness in a cookieless world. These methods don’t care about cookies at all - they measure causal lift directly.

Modelled conversions. Both Google and Meta now provide modelled conversion data that fills gaps left by tracking limitations. The accuracy of these models has improved meaningfully through 2024-2026, though there’s still significant variance by category.

What hasn’t worked

Several approaches that got hyped post-cookie haven’t really delivered.

Probabilistic identity resolution at scale. Vendors promised they could stitch identity across devices and sessions using probabilistic methods. The reality has been that match rates are mediocre, false positive rates are high, and the regulatory environment has gotten more hostile to these approaches. Most teams we talk to have stopped paying for these services.

Universal IDs in advertising. UID 2.0 and similar industry initiatives have had limited adoption beyond a handful of major publishers and platforms. The promise of a cookie-replacement identity layer hasn’t materialised at the scale needed to be useful.

Server-side tracking as a solution to cookie problems. Many teams adopted server-side tracking thinking it would solve their measurement gaps. It helps with some specific issues (browser blocking of pixels) but doesn’t fundamentally address the underlying privacy regime - users who don’t want to be tracked still aren’t tracked.

The shift in measurement philosophy

The deeper change isn’t tactical - it’s how marketing teams think about measurement.

The old model: track everything individually, attribute precisely, optimise to user-level outcomes. This world is largely gone for any team operating at scale across multiple channels.

The new model: measure aggregate impact through MMM and experimentation, supplement with available platform-level signal, accept inherent uncertainty in attribution, and optimise based on directional confidence rather than precise allocation.

This is a healthier model in many ways. It puts more emphasis on causal understanding (what actually drives results) and less on credit assignment (which channel deserves the win). The teams that have made this shift psychologically tend to be more effective marketers, in our experience.

The platforms’ role

Worth flagging how the major platforms have responded. Meta and Google have both invested heavily in their first-party measurement capabilities - Meta’s Conversions API and Advantage+ products, Google’s Enhanced Conversions and Performance Max. These tools work better when fed strong first-party signal from the advertiser, which incentivises advertisers to invest in their own data infrastructure.

The platforms have also become more aggressive about claiming attribution credit through black-box modelling. Reported conversions in modern PMax or Advantage+ campaigns include significant modelled attribution that can’t be independently verified. This makes the platforms’ reported numbers unreliable as a sole attribution source.

The pragmatic response from advertisers in 2026 has been to treat platform-reported numbers as one input alongside MMM and incrementality testing - not as the ground truth.

What we’d recommend in 2026

Practical recommendations for teams thinking about their cookieless attribution stack:

Invest in first-party data. This is the foundation everything else builds on. CDP, server-side event capture, identity resolution for logged-in users, durable storage of customer data. If you don’t have this, fix it first.

Run an MMM program, even if simple. You don’t need to spend $300k on Quantium or one of the boutiques (though they do good work). Open-source tools like Meridian, Robyn or LightweightMMM can produce useful MMM with internal effort. Some MMM is much better than no MMM.

Build incrementality testing into your operating cadence. The discipline of regular holdout testing pays back enormously. Make it a quarterly or monthly process for your largest spends.

Use platform tools but don’t trust their numbers blindly. Send strong first-party signals through Conversions API and equivalents. Use the platforms’ modelling. But validate against MMM and incrementality - don’t just believe what they report.

Accept the uncertainty and communicate it clearly. Stop pretending you can attribute conversions precisely to channels at user level. You can’t. Be honest about what you know and don’t know, and frame decisions accordingly.

The longer arc

We expect the next 12-24 months to see continued maturation of these patterns rather than any dramatic shift. The cookie-related hand-wringing has largely resolved into a stable equilibrium where serious advertisers have built durable cookieless attribution capability and the rest are operating with degraded measurement they don’t fully realise.

The gap between attribution-mature and attribution-weak organisations has widened, not narrowed. Teams that invested in the infrastructure and methodology shifts are now meaningfully better at measurement than peers who waited. That advantage compounds.

If you’re in the second group, the work to catch up isn’t impossibly hard. But it does require intentional investment in data infrastructure, modelling capability, and experimentation discipline. The teams getting started in 2026 are 18-24 months behind the leaders, but starting now beats starting in 2027.

The cookieless world isn’t fully here yet, but the cookie-dependent world is gone. Build for the actual reality, not the one you wish persisted.