Carbon Labels Work at the Shelf, Not in Memory

Two pre-registered experiments on what happens when emissions data is visible at the moment of choosing

🏷️ Information you can see, versus information you once saw

Almost everyone reading this already knows, roughly, that beef is worse than beans and that flying is worse than the train. Knowing has never been the hard part. The interesting question is narrower and more practical: does putting an emissions number in front of someone at the instant they decide change what they pick — and does the effect survive once the number is gone?

The study

A 2025 paper in Nutrients by Jonas D. Lenk, Pierre Chandon and Sonalie Doshi ran two incentive-compatible, pre-registered choice experiments in which the researchers manipulated one variable that most label studies leave fixed: the proportion of products in the assortment that carried a carbon label. Not whether a label existed, but how much of the shelf was labelled.

In the first experiment, 715 people shopped for instant meal products in an online grocery store stocked with a complete product category. In the second, 1,233 people were pushed into an explicit trade-off: they ranked cut-fruit products by preference, then made two consecutive choices — one without labels, one with them — so that any shift had to be paid for in preference. Both groups were surveyed again months later to see whether anything had stuck.

📊

Lenk, Chandon & Doshi (2025) — Nutrients

“Which Consumers Change Their Food Choices in Response to Carbon Footprint Labels? The Role of Political Ideology and Other Socio-Demographic Factors”

Reported findings:

Two pre-registered, incentive-compatible choice experiments (n = 715 and n = 1,233)
Increasing the proportion of carbon-labelled products led liberals and centrists to choose lower-emission foods, with minimal or no impact on conservatives
The authors note the labelling “never backfired” in any group
No other individual characteristic moderated the effect once political ideology was controlled for
Younger age, lower subjective socioeconomic position and urban residence improved responsiveness indirectly, by predicting a more liberal ideology
Among liberals, the labelling effects persisted for four months but not longer

View publication →

Nutrients, 17(8), 1321. Published 10 April 2025.

Coverage is a variable, not a detail

The design choice that makes this paper useful is the one about proportion. A single labelled product tells you something about that product. A fully labelled shelf tells you something about the range — it turns an isolated number into a comparison, and comparison is what a choice actually needs. The authors found responsiveness rising as coverage rose, which is a finding about the structure of the information rather than about any particular label design.

That has an obvious implication for anyone estimating their own footprint. A number attached to one meal is close to useless on its own. The same number sitting next to yesterday’s meals, and next to the alternatives you did not pick, is the thing that can be acted on.

Four months, then nothing

The follow-up result is the one worth sitting with. Among the group that responded to labels at all, the effect was still detectable four months after exposure — and had disappeared by the longer follow-up. One encounter with good information moved behaviour for a season, and then the behaviour drifted back.

This is not a criticism of labels. It is a description of how most one-off interventions behave, and it is consistent with a broader pattern in behavioural research: exposure produces a decaying effect, while repeated exposure at the point of decision produces a habit. The practical reading is that the carbon number needs to keep showing up, not to show up once and be remembered.

Who responds, and the honest caveat

The headline moderator in the paper is political ideology: liberals and centrists shifted, conservatives largely did not, and none of the other socio-demographic variables mattered once ideology was accounted for. Age, socioeconomic position and urban residence showed up only indirectly, as predictors of ideology rather than as independent causes.

Two limits are worth stating plainly. These were choice experiments, not months of real grocery receipts, and they were conducted in a specific national context whose political alignments do not transfer everywhere. What generalises is the mechanism — visible comparison at the moment of choice, decaying without repetition — rather than the specific segments.

What this means for tracking your own footprint

A personal carbon estimate has the same two failure modes as a label. It can be isolated — one annual figure, no context, nothing to compare against — and it can be a single event, calculated once and never revisited. The research suggests both are the wrong shape. Emissions information changes choices when it is broad enough to compare and frequent enough not to fade.

That is the argument for daily tracking over annual calculation, and it is why a tracker that shows every logged item against the alternatives, against your own past days, and against a target is doing something structurally different from a label you read once in a supermarket.

Keep the number in front of you

Decarbonr logs food, travel and 12+ lifestyle categories on your iPhone, using emission factors drawn from Poore & Nemecek (2018), the European Environment Agency and the IPCC. No account, no ads, and nothing leaves your device.

General information only. Findings are summarised from the cited publication; refer to the original paper for its full methods and limitations.