Cheaper fact-checking can backfire when recommenders adjust their standards

Making recommendations easier to verify does not necessarily improve the recommendations themselves. In a theoretical preprint, Ying Gao of the University of British Columbia and Yingju Ma of the University of Cambridge identify circumstances in which cheaper checking encourages a self-interested intermediary to recommend more bad products. Their numerical examples show that deteriorating standards can outweigh consumers’ verification savings.

This is a conditional theoretical result, not evidence from an experiment or an analysis of platform data.

Persuasion versus exploitation

The model gives consumers a safe option and a risky product whose value the platform knows. The platform earns more when consumers adopt the risky product, creating an incentive to recommend it even when adoption would give consumers negative value.

Consumers know their own checking costs and understand the recommendation policy. They can accept, reject or pay to learn the risky product’s value perfectly. Verification costs can represent time or attention rather than money. When recommendations are attractive on average, low-cost consumers verify while high-cost consumers accept without checking.

In the main analysis, the platform commits in advance to a recommendation policy. The authors show that an optimal binary policy can take the form of a quality cutoff: recommend the risky product above the cutoff and withhold it below.

Lowering the cutoff admits more bad products but makes checking more valuable. The platform balances additional adoption by non-checkers against the incentive for other consumers to verify and reject bad products.

A “false recommendation” here means recommending a product with negative consumer value. It does not necessarily result in a purchase: checking can prevent adoption.

Why checking-cost differences matter

The authors examine transparency improvements that reduce everyone’s verification costs by the same proportion, preserving relative differences across consumers. Even these proportional improvements can produce opposite responses.

With exponentially distributed checking costs, cheaper verification raises standards in the paper’s example. False recommendations disappear as checking becomes nearly free.

A sufficiently heavy-tailed cost distribution can reverse the response. Such a distribution includes consumers with exceptionally high checking costs. As verification becomes cheaper, the platform may find persuading marginal consumers less rewarding than exploiting those who still accept without checking.

The local mathematical condition concerns how the non-checking share responds proportionally to increases in the recommendation pool’s expected downside. When the safe option earns the platform nothing, declining proportional sensitivity at the optimum produces the adverse response, under the paper’s regularity assumptions. A profitable safe option changes that condition.

Heavy tails are not essential. Mixing two groups with different average verification costs can generate similar behavior even when each group separately has a light-tailed distribution. The authors call these groups “sophisticated” and “naive,” but both make rational decisions in the model.

In one constructed example, cheaper verification triggers a discontinuous switch between recommendation strategies. The mass of false recommendations rises from 0.193 to 0.461, and consumer welfare falls by 6%. These are model outcomes, not measured market effects.

Commitment changes the comparison

Without commitment, the platform chooses whether to recommend after observing quality, taking consumers’ beliefs as given. The authors analyze a threshold equilibrium assuming that the safe option is profitable for the platform and the risky product has nonnegative average consumer value.

In that equilibrium, cheaper verification never increases false recommendations or lowers consumer welfare. Unless every product is recommended, the platform is disciplined until it is indifferent between a false recommendation and the safe option.

Removing commitment is not therefore a prescription for helping consumers. At the same checking costs, commitment supports at least as high standards and consumer welfare as the corresponding no-commitment threshold equilibrium. It improves outcomes in levels while permitting an adverse response to greater transparency.

Limits of the warning

The model assumes binary recommendations, perfect verification and consumers who correctly anticipate the policy. Richer ratings and imperfect checks remain open questions, and the paper does not estimate real consumers’ checking-cost distributions.

Worse recommendations also need not outweigh cheaper verification. In the proportional-cost model, consumer welfare approaches its first-best level as checking costs vanish. The narrower lesson is that assessing transparency requires considering intermediaries’ strategic responses, not just the direct savings from easier access to information.