← CHI 2026 map

CHI '26 · Honorable mention · full-paper review · confidence medium-high

Good for the Planet, Bad for Me? Intended and Unintended Consequences of AI Energy Consumption Disclosure

Michael Klesel , Uwe Messer

DOI PDF Program page Page JSON

This is a solid CHI empirical paper because it does more than show that energy disclosure nudges greener choice; it also surfaces a counterintuitive downside in user experience. The main value is the paired behavioral and perceptual evidence, though the claims should stay anchored to the paper’s single low-stakes online experiment.

Axes Lens

Rare contribution shape, typical evidence profile. The point here is not a score. It is to show what kind of claim the paper makes, and whether the evidence pattern is unusual or baseline in this 268 -review set.

Review Summary

This paper’s strongest contribution is that it treats energy consumption disclosure as more than a simple sustainability prompt and tests both intended and unintended consequences in the same study. The abstract and contribution framing indicate a controlled experiment with 365 participants, and the reported results are clear: disclosure substantially increased selection of the energy-efficient SLM, yet did not change subsequent prompt behavior, and it also produced a placebo-like reduction in satisfaction and perceived quality among those who chose the eco-friendly option. That combination makes the paper interesting for CHI because it complicates a common design intuition: transparency or sustainability framing is not automatically benign, even when it succeeds behaviorally. The novelty is therefore primarily empirical rather than conceptual; the paper appears to be among the first to test ECD in this AI model-choice setting and to measure the downstream subjective cost of the nudge. At the same time, the evidence base is bounded. The study is a one-factorial between-participants online experiment, apparently in a low-stakes and short-term setting, with a UK sample and a specific disclosure design. Those constraints matter because they limit how far one can generalize from this interface to real-world AI products, higher-stakes decisions, or longer-term use. The paper is therefore best read as a strong task-level causal finding with clear design implications, not as a universal statement about all sustainability disclosures. Its practical message is valuable: if designers deploy ECD to steer users toward greener AI choices, they should anticipate possible expectation effects and consider how to avoid making the sustainable option feel worse after the fact.

What Changed

Canon before

Energy consumption disclosure is usually treated as a straightforward sustainability nudge: tell users a model is costly, and they will choose the greener option. The paper complicates that canon by showing the nudge can alter choice without improving downstream behavior and can even depress subjective experience through a placebo-like effect.

Departure from common sense

The surprising result is not just that disclosure shifts choice, but that the greener choice can make users feel worse about the same interaction. That runs against the intuitive assumption that pro-environmental framing should be psychologically benign or even rewarding; here, the disclosure appears to create a negative expectation effect.

Actual novelty

The paper’s novelty is an empirical test of AI energy consumption disclosure as a behavioral nudge in model selection, paired with measurement of unintended perceptual consequences. It contributes evidence that ECD can strongly steer choice while also producing a placebo-like drop in satisfaction and perceived quality among those who chose the eco-friendly option.

Evidence

The paper reports a one-factorial between-participants online experiment with 365 participants and frames three contributions, including an empirical test of ECD effectiveness. The abstract reports that ECD increased the odds of choosing an energy-efficient SLM by more than 12, while subsequent prompt behavior did not differ significantly. It also reports a placebo effect in which SLM choosers reported lower satisfaction and perceived quality.

Limits

Method limits

The evidence comes from a single online between-participants experiment, so causal inference is limited to the tested interface and task. The study also appears to rely on one low-stakes choice context and a specific disclosure framing, which constrains claims about broader behavioral or perceptual effects.

Deployment limits

Deployment in real systems may differ because the study uses a fictional choice interface and a short-term session rather than sustained use. The reported placebo-like dissatisfaction could matter for adoption if disclosures are surfaced in production, but the paper does not test mitigation strategies or long-run user adaptation.

Boundary conditions

The findings are bounded by a UK participant sample, a low-stakes online task, and a disclosure design that aggregates energy information at the model level. The paper itself notes that effects may differ in high-stakes situations and that the study does not examine whether the perceptual bias can be reduced.

Manual check: No additional manual check required beyond standard merge QA; evidence is sufficient for the stated claims, but the novelty claim should remain modest and tied to the reported experiment.

Position in field

This paper sits at the intersection of sustainable HCI, transparency/disclosure design, and AI interaction. Its main contribution is to show that sustainability disclosures can have both intended behavioral effects and unintended subjective costs, making it relevant to debates about how to design pro-environmental nudges without harming user experience.

Abstract

To address the high energy consumption of artificial intelligence, energy consumption disclosure (ECD) has been proposed to steer users toward more sustainable practices, such as choosing efficient small language models (SLMs) over large language models (LLMs). This presents a performance-sustainability trade-off for users. In an experiment with 365 participants, we explore the impact of ECD and the perceptual and behavioral consequences of choosing an SLM over an LLM. Our findings reveal that ECD is a highly effective measure to nudge individuals toward a pro-environmental choice, increasing the odds of choosing an energy efficient SLM over an LLM by more than 12. Interestingly, this choice did not significantly impact subsequent behavior, as individuals who selected an SLM and those who selected an LLM demonstrated similar prompt behavior. Nevertheless, the choice created a perceptual bias. A placebo effect emerged, with individuals who selected the "eco-friendly" SLM reporting significantly lower satisfaction and perceived quality. These results highlight the double-edged nature of ECD, which holds critical implications for the design of sustainable human-computer interactions.