Over the past few years, audience extension has become a strategic priority for programmatic ad sales teams. The idea is simple: retarget a publisher's audiences outside its own sites.
It lets:
buyers increase reach against their target audience and lock in targeting quality, while tapping into impressions that may perform better or into new formats;
publishers' ad sales teams capture more budget and add a complementary offer built for tighter KPIs.
That makes it good for the whole programmatic ecosystem. It strengthens the Open Web's overall value proposition: audience segment vendors sell their data, and publishers sell their media.
But audience extension comes with a problem that publishers' sales teams know well and rarely talk about. It's the gap between:
the volume of audiences visiting the publisher's site (unique visitors in Google Analytics or Piano, for example), which the publisher hopes to sell;
the volume of audiences actually addressable in programmatic tools.
That gap can reach 75%.
So where does all that audience go?
A series of technical constraints can hurt audience tracking. Put end to end, they account for the losses. Call it the audience leakage chain.
The good news: there are technical fixes at every step to cut the loss rate and improve tracking.
Link 1: bot traffic that slips through
Publishers running standard "Accept all" / "Continue without accepting" consent banners have noticed something odd. A significant share of visitors bounce at the banner, even when they arrive from a high-intent path such as a very specific search query.
The reason: security tools such as Cloudflare block most bot traffic, but some bots get through. They then stall at the next step, the consent choice. The result is abnormally high bounce rates on consent banners.
That traffic isn't human, yet the analytics tool still counts it in the site's audience. Improving tracking won't help here. The fix is to block unwanted bots further upstream.
Link 2: the consent rate
Since GDPR took effect, every website has to collect a visitor's consent before assigning an identifier that enables ad targeting. No identifier, no audience extension.
Consent rates swing widely depending on the banner. There are many banner designs today, but the three most common types break down like this:
"OK Consent" | "No Consent" | "I'm leaving the site" (bounce rate excluding suspicious traffic) | |
|---|---|---|---|
Standard "Accept all" or "Reject all" banner | 55%-60% | 35%-45% | 0%-5% |
Dark-pattern banner ("Accept all" is emphasized) | 65%-75% | 15%-30% | 5%-10% |
Consent wall | 85%-95% | 0%-10% | 5%-15% |
Most publishers historically used standard or dark-pattern banners because France's data protection regulator, the CNIL, approved those designs. A late-2025 ruling by France's top administrative court, the Conseil d'État, forced the CNIL to loosen its stance on consent walls.
Half of France's 100 largest sites have since switched to a consent wall, including national dailies Le Monde and Le Parisien and sports daily L'Équipe. The model does push bounce rates up. It also sharply increases the volume of trackable audience on the site.
Recommendations: favor a consent wall to maximize trackable audiences. Publishers can go a step further with a login wall that captures the user's email address, which is very useful for cross-site, cross-browser and cross-device targeting. Some have already made the jump, including L'Équipe on its mobile app.
Link 3: browser anti-tracking protections
Since Google Chrome announced it was dropping its Privacy Sandbox project, cookieless seems to have vanished from the conversation.
Publishers still worry about it. Safari accounts for about 30% of mobile browsing in France and blocks targeting through third-party cookies.
Anti-cookie privacy policies lead to:
lower fill rates;
lower CPMs;
audience extension being blocked for those users.
The good news: alternative identifiers such as EUID, First-ID, ID5, RampID and UtiqID get around browser blocking.
Recommendations:
deploy at least one alternative identifier;
pick an ID with scale in the French market that feeds DSPs' ID graphs, to improve sync rates (see below);
share the identifier with every SSP seat connected to the site.
Link 4: DMP sync rates
The risk at this stage is a badly implemented DMP tag, or a technical issue that stops it from firing. Either makes the audiences unusable.
The main fixes are to check:
that the tagging plan has no gaps;
that tags fire correctly across every section of the site.
Link 5: DSP sync rates
The risk here is a poor sync rate between the DMP and the DSP. Your options are limited, since the problem sits between two external tools.
All you can do is A/B test DSPs to see which one syncs best with your DMP.
Conclusion
Human traffic to publisher sites is expected to keep falling in the near future as AI queries drive more zero-click search. Working on the leakage chain is becoming critical to limit the loss of usable traffic.
Data sellers that put in the work on their leakage chain will weather the crisis best. They'll also be in position to win advertisers' biggest media budgets.
Enjoyed this piece? Read more from Open Garden:
