We Said Spend-Based Overstated Ad Emissions by 450%. GMSF v1.3 Changed the Math.

In early 2026, we made a strong claim on this blog: spend-based carbon measurement overstates digital advertising emissions by around 450%. The comparison was real, the arithmetic was sound, and the conclusion held, under the methodology of the time.

Then GMSF v1.3 arrived, and it changed the math.

This is the honest update. We are retiring the “450% error” headline, and the lesson that replaces it is sharper than the one it corrects.

What we said, and why it was true then

The original comparison ran the same campaigns through two methods at once.

Spend-based methodology multiplies advertising spend by a generic factor from economic input-output databases: spend a million euros, multiply by a sector average, get an estimate. GMSF v1.2 activity-based methodology calculated emissions from what actually happened: impressions served, data transferred, video seconds encoded, supply paths traversed.

Across 83 campaigns measured in Q4 2025, spend-based calculations averaged 2.62 tonnes CO2e per million euros of spend. GMSF v1.2 averaged 0.47 tonnes. Spend-based looked roughly 5.5 times higher, and the explanation was simple: spending reflects commercial value (what publishers charge), not operational intensity (what infrastructure actually runs).

That gap was the whole article. It no longer exists.

What GMSF v1.3 changed

GMSF v1.2 measured the use phase: the electricity consumed while an ad is selected, delivered, and displayed. That was a real step forward over spend-based proxies. But it undercounted in two ways. It ignored embodied emissions, the carbon of manufacturing the servers, networks, and above all the user devices that run the ads. And it approximated the true depth of the programmatic supply chain.

GMSF v1.3 closes both gaps. It moves to a bottom-up life-cycle assessment that counts use and embodied emissions across the three stages of an impression (inventory selection, creative delivery, device usage), using official reference data (ADEME “Numerique 2.0” 2025, Ember 2024 grids).

The consequence is large. For the same campaign, v1.3 lands roughly 5 to 6 times higher than v1.2. Run our 0.47-tonne figure through that and you get something close to 2.6 tonnes per million euros, which is almost exactly the spend-based number we spent an entire article criticizing.

So the “450% overstatement” no longer holds. At the portfolio level, spend-based and a complete life-cycle measurement now land in the same ballpark.

So was spend-based right all along?

No. And this is the part that matters.

Spend-based now lands near the right total, but by coincidence, not by measurement. Its sector-average factors happened to drift toward the truth as the industry’s real footprint, embodied included, turned out higher than v1.2 suggested. A stopped clock is right twice a day. That does not make it a clock.

Being roughly right at the aggregate still cannot do three things that matter.

It cannot tell two campaigns apart

Consider two campaigns that each cost 500,000 euros.

Campaign A runs premium video on major publishers through direct deals, with first-party targeting and creative served from edge caches near users. Campaign B runs the open programmatic exchange through fifteen supply-chain hops, with heavy uncompressed video and third-party data from multiple providers.

Same spend. Campaign B emits several times more. Spend-based methodology reports them as identical, because it only ever sees the invoice. A complete measurement sees the infrastructure.

It cannot drive optimization

You cannot reduce what you cannot attribute. Spend-based gives you one number for the whole portfolio, with no breakdown by channel, format, supply path, device, or geography. v1.3 traces emissions to each of them, so you know which lever actually moves the number, and by how much.

It cannot survive an audit

The Corporate Sustainability Reporting Directive requires assured reports, audited by third parties who verify both the data and the method. An auditor looks at a spend-based figure and asks one question: what operational data supports this number? Spend-based has no answer by design, because it substitutes an economic proxy for operational data. v1.3 traces every figure back to impression logs, transfer volumes, device splits, and grid factors. One is unsubstantiated. The other is audit-ready.

The real lesson: completeness, not just accuracy

Think of it as three tiers.

• Spend-based is a proxy. Right by luck at best, useless for action, and impossible to substantiate.
• Activity-based (v1.2) was real, but use-phase only. It flattered you with a low number.
• Full life-cycle (v1.3) counts use and embodied. It is the rigorous standard: a bigger number, but the true one, and an actionable one.

The uncomfortable takeaway: if your “good” carbon number came from a spend-based proxy or a v1.2-style use-phase model, your real footprint is almost certainly larger than you reported. That is not a reason to look away. It is the first honest baseline you can actually reduce.

What this changes for you

Reported tonnage will rise, not fall. A v1.3 baseline is higher than a v1.2 one, so do not expect the headline number to drop. It will likely go up. The point is no longer “measure accurately and pay for fewer offsets.” It is “measure honestly and stand behind the figure in front of an auditor, a regulator, or an investor.”

Optimization still works the same way. The real wins have not changed: consolidate supply paths, cut creative weight, shift format and device mix, favour low-carbon grids, drop made-for-advertising inventory. v1.3 simply measures those gains honestly, embodied included, instead of crediting reductions that were partly an artefact of an incomplete model.

Migration is straightforward. Most teams can run v1.3 alongside their existing measurement. On Carbon Intelligence, the dual-engine keeps your familiar baseline and adds v1.3 as an option, so you can compare the two stage by stage and switch at your own pace.

Why we are publishing this correction

We could have quietly edited the old post and moved on. We did not, because we hold our own methodology to the standard we ask of our clients. When better data arrives, you update the number, even when the new number is less flattering, and even when it revises something you said yourself.

The “450% error” made for a good headline. The truth underneath it is more useful: spend-based was never measuring anything, v1.2 was measuring half the picture, and the real footprint of digital advertising is larger than the industry has been comfortable admitting. Measuring it honestly is the only way to bring it down.