Cookieless retargeting means re-engaging site visitors using consented first-party data, server-side measurement and page context instead of third-party cookies. The best mix for most teams is consented first-party audiences, server-side tagging, contextual overlays and probabilistic modelling. Start by auditing your consent flow and mapping Advanced Consent Mode to your CMP, because every other piece depends on it.
Key Takeaways
Effective cookieless retargeting relies on first-party data, consent-aware tagging, contextual targeting, and AI-driven modelling to maintain reach and compliance.
| Point | Details |
|---|---|
| Prioritize first-party data | Use consented site data, logged-in IDs, and CRM lists as the foundation for retargeting efforts. |
| Implement consent-aware tagging | Configure server-side tags and Advanced Consent Mode to respect user choices and support measurement. |
| Utilize contextual targeting | Target audiences based on page content and context instead of browsing history for privacy compliance. |
| Leverage AI and modelling | Fill data gaps with AI-driven probabilistic models, but always verify with holdout tests. |
| Thrive Agency’s approach | We integrate strategy, creative, and measurement, handling setup if internal resources are limited. |
Table of Contents
- What is cookieless retargeting and why does it matter?
- How does cookieless retargeting work in practice?
- How do you set up cookieless retargeting, step by step?
- How do you measure cookieless retargeting performance?
- What creative changes lift cookieless retargeting results?
- Which tool categories should you evaluate?
- What can you expect from a managed implementation?
- Agency perspective: what to stop doing
- How we can help with cookieless retargeting
- FAQ
- Sources
Third-party cookies are the old plumbing of retargeting: a tracker on one site recognises a browser on another. Browsers and regulators have spent years making that plumbing unreliable, and Google’s Privacy Sandbox update (October 2025) now points the project at interoperable attribution standards and privacy-preserving measurement APIs. The direction is clear even where the timeline keeps shifting. The Chromium team set it out back in 2020.
Retargeting itself is worth keeping. A visitor who looked at a product has told you more than any lookalike model can guess. What changes is how you recognise that person and what you are allowed to do with the signal.
Why move now rather than later:
- Signal decay is gradual, not a cliff. Each browser or consent change quietly shrinks your audiences, so waiting means you rebuild under pressure.
- First-party data compounds. Every consented email, login and site event you collect this quarter works for you next quarter.
- Privacy-aware setups can help teams respond to changes in rules and browser defaults. Consent-aware tagging still needs to be reviewed when those changes happen.
- Testing takes time. Incrementality tests and modelling need history, so teams that start earlier can gather evidence sooner.
Several methods can help. None is enough alone.
First-party audiences. These come from signals you own: site events, logged-in IDs and CRM lists. You upload hashed lists to ad platforms or build audiences from consented site events. Where it falls short: reach is capped by how many visitors identify themselves or consent.
Server-side tagging with Advanced Consent Mode. A web container collects consent, and a server container processes the data and fires consent-aware tags, as Google’s server-side consent mode guide describes. When consent is denied, Advanced Consent Mode sends anonymised pings, which lets Google Ads and GA4 use advertiser-specific modelling rather than a general industry model. Where it falls short: modelled conversions are estimates, not observed ones.
Contextual targeting. Ads match the page the person is reading, not their cross-site history. Shopify’s primer (October 2025) describes it as content-first advertising without tracking identifiers. Where it falls short: it reaches an audience in the right mood, not the specific person who left your basket.
Probabilistic modelling and synthetic personas. AI fills gaps where signals are thin. The IAB State of Data 2025 (March 2025) reports a shift toward AI-assisted campaign workflows and synthetic data. Treat it as augmentation. Where it falls short: a model trained on thin data produces confident noise, so it needs a holdout to check it.
Pro Tip: Rank your methods by signal quality, not novelty. Consented first-party data first, modelling last.
Do these in order. Skipping the first two is how teams end up with modelling built on junk.
- Audit your consent banner and CMP. Consent Mode is not a substitute for a CMP. Without a CMP recording granular choices, tags fire blind and modelling can be useless or non-compliant.
- Map CMP signals to tag triggers. Persist the consent state across pages and sessions so a denied choice stays denied.
- Stand up a server container. Configure consent-aware tags so measurement pings reach analytics even when cookies are refused.
- Switch on Advanced Consent Mode, and assess whether URL passthrough suits your setup. Check the platform documentation to understand how it affects click attribution when cookies are unavailable.
- Build and refresh first-party audiences. Use site events and CRM imports, then export to ad platforms and private match lists.
- Layer contextual targeting and modelled personas. These reach people who decline cookies.
- Run small incrementality tests. Hold out a group, then compare attribution models against the lift you actually measured. Our own write-up on incrementality testing for Meta ads covers how we design holdouts.
Pro Tip: Fix steps 1 and 2 before touching anything else. Modelling on top of a broken consent map teaches the model the wrong thing, and you will not see it in the dashboard.
Measurement is where most teams lose their nerve. Dashboards show fewer observed conversions, so people assume performance fell. Often the signal fell, not the sales.
Here is how to read the data without fooling yourself:
- Know which model the platform uses. Advertiser-specific modelling uses your own consented and anonymised signals. A general industry model leans on averages from other advertisers. Check which one your setup triggers.
- Treat incrementality as the verification layer. Platform attribution reports conversions it can see. A holdout can help estimate whether ads caused additional conversions. It is one way to check platform attribution independently.
- Plan around current limits. Google’s relevance and measurement FAQs say frequency capping and cross-site frequency signals remain a hard problem, and the alternatives can add latency or noise. Cap frequency inside your own channels and expect some overlap.
- Turn on the platform features. Enhanced conversions, first-party match and aggregated reporting all help, and identifiers should persist only where lawful.
Advertiser-specific modelling is a Google Ads and GA4 capability that depends on Advanced Consent Mode being implemented, not just a consent banner being present (Google Developers).
We cover the reporting side in our piece on Google Ads attribution columns.
Less behavioural data means the creative has to do more of the targeting. A generic “still thinking about it?” reminder was always weak. Without a precise audience behind it, it is dead weight.
- Ask for a specific action. Prompts that invite a choice, such as “customise your pack”, give the platform a clearer optimisation signal than another brand reminder.
- Build modular creative. Swap the headline, product and offer by page context or by a known first-party action, so a visitor from a recipe page sees something different from one on a pricing page.
- Test fast, and tie every test to a holdout. Without a held-out group, a creative “win” is often just the audience mix shifting. That is how teams end up reporting uplift that never existed.
Why this works: when targeting gets blunter, the message becomes the filter. People who care self-select by clicking, and that behaviour is the signal you can still collect.
Pro Tip: Change one variable per test. If you swap audience, offer and creative at once, you will never know which one moved the result.
Which tool categories should you evaluate?
Ask capability questions, not brand questions. Here is what to check in each category.
- Customer data platforms (CDPs): identity stitching quality, real-time audience exports, and how retention and deletion policies handle consent changes.
- Server-side tag managers: latency, how consent maps from your CMP, who hosts and maintains the server container, and what that costs in engineering time.
- Contextual DSPs and partners: taxonomy depth, scale in your markets, and whether the creative can adapt to the context it lands in.
- Owned channels (email, SMS) and first-party match: deliverability, opt-in quality and how sequencing avoids pestering one person on three channels.
- AI orchestration and modelling tools: governance, explainability and data provenance. The IAB companion guide says agencies should integrate probabilistic modelling and synthetic personas, and the full report notes that many agencies use accessible AI tools while far fewer run proprietary or organisation-wide integrations. If a vendor cannot explain what trained its model, walk away.
Our artificial intelligence archive shows how we think about these tools.
What can you expect from a managed implementation?
An integrated team shortens test cycles because measurement, creative and server work sit in one room. There is no hand-off where the tagging brief gets lost between the media buyer and the developer.
Here is the proof statement we use, and what a real one needs:
Before: [metric] at [number]. After: [metric] at [number]. Timeframe: [weeks]. A proof point without all three is a testimonial, not evidence.
You can read our published client results in the Sorted Drinks case study and the Bonaire case study.
An operational note: if your internal team lacks engineering bandwidth, CMP mapping and server container setup is the work that stalls longest. An agency can usually run it in parallel with creative testing, so you are not waiting on one before starting the other.
Agency perspective: what to stop doing
Stop firing tags blind. Stop leaning only on third-party cookies and calling the shortfall “attribution drift”. Some performance losses may come from consent-mapping problems that teams have not checked.
Our short checklist: CMP mapping, server container, first-party audiences, contextual layer, creative tests, incrementality holdouts. In that order.
Happy to help you start.
— Harry
We handle the strategy, creative and measurement under one team: consent mapping, server-side tagging, creative testing and incrementality measurement, built around how your business actually sells. If your team lacks the engineering time, we can run the setup while you keep the campaigns live. Book a short consultation or an implementation audit on our consultation page, or see the full range on our services page.
FAQ
Is Google still phasing out third-party cookies?
Google has reshaped its plans several times, so check the current position before you commit a roadmap. Its Privacy Sandbox update now focuses on interoperable attribution standards and privacy-preserving measurement APIs. Build for a world where cookies are unreliable either way.
Are tracking cookies illegal?
Whether tracking cookies are allowed depends on the applicable privacy rules and the circumstances. Ask your legal team about your specific case. A consent platform that records granular choices is the practical starting point, as the Google consent guidance makes clear.
What is Advanced Consent Mode?
It is a setup that sends anonymised pings when a visitor denies consent, so platforms can estimate conversions. It allows advertiser-specific modelling instead of a general industry model.
Does contextual targeting replace retargeting?
No. Contextual targeting reaches people by what they are reading, not by what they did on your site. Use it to widen reach alongside consented first-party audiences, not instead of them.
Sources
- IAB State of Data 2025: AI on the Brink of Transforming Advertising
- Contextual Targeting: What It Is and How It Works (2025) – Shopify
- Developers
- Update on Plans for Privacy Sandbox Technologies
Recommended
- Incrementality Testing for Meta Ads – Two Perspectives on Social Performance Advertising Measurement
- Google Ads Attribution: Check the Columns Before You Change Bids
- The Seven-Figure Blueprint: Advanced Facebook Ads Scaling Strategy for 2026
- Increase Ecommerce Conversions 40% with YouTube Ads, Thrive Checklist
