Companies inadvertently fund online misinformation despite consumer backlash

The study addresses how the supply of online misinformation is financially sustained and how to reduce its monetization. Although prior efforts largely target user behavior to reduce demand for misinformation, less attention has focused on curbing financial incentives that sustain its supply. The authors examine the role of advertising—particularly programmatic placements by digital advertising platforms—in financing misinformation websites, assess consumer responses to learning that companies advertise on such sites, and investigate whether company decision-makers are aware of and prefer to avoid such placements. The research aims to quantify the prevalence of advertising on misinformation outlets, identify the contribution of digital advertising platforms, measure consumer backlash in terms of exit (reduced consumption) and voice (signaling concerns), and test whether information interventions could shift both consumer behavior and decision-makers’ platform preferences. The work is motivated by broader societal harms from misinformation and the expectation that generative AI may increase misinformation supply by lowering content production costs.

Prior work has emphasized demand-side interventions (e.g., fact-checking, accuracy prompts, credibility labels, crowd-sourced judgments) to reduce consumption and sharing of misinformation, finding mixed or limited average effects on improving news diets or correcting misperceptions. Media and academic commentary suggests advertising revenue sustains misinformation, with programmatic ad platforms implicated in monetizing such content. Platforms have periodically attempted to reduce ad revenue to misinformation sites, yet prominent brands still appear on them. The authors build on literature examining the infrastructure and ad-tech ecosystems supporting misinformation, the roles of ad servers and retailers, and consumer reactions to corporate behavior (“exit” and “voice”)—including responses to content adjacency, CSR, and political stances. The study contributes by combining large-scale observational evidence on advertising across misinformation and non-misinformation websites with incentive-compatible experiments measuring both exit and voice, and by proposing supply-side, information-based interventions (disclosures and rankings) to reduce monetization, complementing existing transparency and regulation-by-information approaches in other sectors.

Data construction and descriptive analysis: The authors compiled misinformation domains from three sources: NewsGuard, the Global Disinformation Index (GDI), and lists used in prior research. They collected advertiser behavior data from 2019–2021 using Oracle’s Moat Pro, which crawls ~10,000 sites daily. The final dataset includes 5,485 websites (1,276 misinformation; 4,209 non-misinformation), 42,595 unique advertisers, and 9,539,847 instances of advertiser appearances on news websites. For the 100 most active advertisers each year, they also collected weekly data on site appearances and use of digital advertising platforms. Descriptive analyses assessed monetization models (ads vs. paywalls), industry coverage, and relative advertising intensity on misinformation vs. non-misinformation sites. They compared advertisers’ likelihood of appearing on misinformation sites conditional on using digital advertising platforms, including models controlling for industry and time trends. Consumer survey experiment: Using an incentive-compatible design, US participants (n=4,039) chose a gift card from a set of companies. Treatments randomly provided factual information about: (T1) the participant’s top-choice company’s ads appearing on misinformation websites; (T2) the role of digital advertising platforms in placing ads on misinformation sites (without referencing the participant’s chosen company); (T3) both the company’s appearance and platform role; or (T4) a ranking of companies by intensity of appearance on misinformation sites. Primary outcome (exit) measured whether participants switched gift card choice post-treatment (binary). Secondary exit outcomes included switching to a lower-preference gift card and switching product category. Voice was measured by clicks to sign petitions directed at companies or platforms to block ads on misinformation sites. Analyses used OLS with robust standard errors, with and without controls for demographics and behaviors; no multiple-comparison adjustments. Text responses were analyzed to assess motivations behind switching. Heterogeneity analyses were pre-registered along gender, political orientation (Biden vs. Trump voters), frequency of product use, and self-reported consumption of misinformation outlets. Decision-maker survey and information intervention: Executives and managers were surveyed via alumni lists from two executive education programs. Job titles were externally verified for most respondents; 94% held executive/managerial roles. Baseline beliefs were elicited about the prevalence of company ads on misinformation sites and platform roles; respondents could request an “advertisement check” by providing company details. An information-provision experiment updated participants about the role of digital advertising platforms in placing ads on misinformation websites. Outcomes included posterior beliefs about platform involvement and demand for a platform-based solution (choosing to receive information on which platforms least frequently place ads on misinformation sites). Subsamples were analyzed by prior beliefs about whether their company’s ads had appeared on misinformation sites, and by certainty in those beliefs. OLS with robust standard errors was used; some subsample analyses were exploratory due to small n and not pre-registered.

Descriptive ecosystem findings: - Monetization: 89.3% of sampled websites were supported by advertising (2019–2021); 74.5% of misinformation websites were monetized by advertising. Paywalls were rarer for misinformation sites (NewsGuard-rated): US 2.7% vs. 25.0% for non-misinformation; global 3.2% vs. 24.0%. - Industry prevalence and intensity: Companies advertising on misinformation sites spanned many industries; 46%–82% of companies in each of 23 industries appeared at least once on misinformation sites. Relative advertising intensity on misinformation vs. non-misinformation sites was similar overall (mean ratio 1.01; 95% CI [0.945, 1.074]; t(22)=0.311; P=0.759). - Role of digital advertising platforms: Among the top 100 advertisers, in a given week 79.8% of those using digital advertising platforms appeared on misinformation sites vs. 7.74% of those not using platforms (two-sided t-test t(192.12)=93.903; P<0.001; n=144). Platform use substantially increased the likelihood of appearing on misinformation sites even after accounting for industry and time. Consumer experiment (exit and voice): - Exit (switching gift card choice): - T1 (company info): b=0.13; 95% CI [0.10, 0.16]; P<0.001. - T3 (company + platform info): b=0.10; 95% CI [0.07, 0.13]; P<0.001. Under T3, attribution of responsibility to platforms rose by 4 percentage points (b=0.04; 95% CI [0.02, 0.06]; P<0.001) yet switching still increased. - T4 (company ranking): b=0.08; 95% CI [0.05, 0.11]; P<0.001. Among switchers (n=430), T4 induced the largest shifts toward companies that less frequently advertised on misinformation sites (b=0.95; 95% CI [0.19, 1.71]; P=0.015). - T2 (platform info only): small but significant switching b=0.03; 95% CI [0.01, 0.05]; P=0.012. Secondary exit outcomes (switching to lower preference; switching categories) were not significant for T2. - Robustness: Participants also switched to lower-preference options (Cols 3–4) and across categories (Cols 5–6) under T1, T3, T4, indicating real costs to switching. - Voice (petition clicks): - Company-directed petition: T4 increased clicks by 0.04 (P≈0.047); T1, T2, T3 not significant. - Platform-directed petition: T2 increased clicks by 0.05 (P<0.01); others not significant. - Heterogeneity (exit): Stronger effects among women (Δ=+0.05; P=0.011) and Biden voters (Δ=+0.03; P=0.058); weaker among frequent users (Δ=−0.05; P=0.007) and those consuming select misinformation outlets (Δ=−0.04; P=0.097). Biden voters were also 5 percentage points more likely to voice concerns against companies (P=0.04). Decision-maker beliefs and behavior: - Baseline beliefs: Decision-makers overestimated overall prevalence of advertising on misinformation sites (estimated 64% vs. observed 55% among 100 most active advertisers) and underestimated platform role (estimated 44.5% vs. observed 79.8% among top advertisers). Only 41% believed consumers react against companies advertising on misinformation sites. - False uniqueness: Only 20% believed their own company’s ads had appeared on misinformation sites. Among those requesting an ad check where company data existed, ~81% of companies had appeared on misinformation sites. Of those informed their company had appeared, 62% were surprised; none were surprised when told they had not appeared. - Information demand and action: 74% requested an advertisement check; 73% opted to receive information on consumer responses (58% exit, 15% voice); 18% signed up for a 15-minute expert session on avoidance strategies. Information intervention on decision-makers: - Posterior beliefs: Significant increases in beliefs about platform involvement overall (winsorized; Table 3 Col 1) and especially among those who believed their company had not advertised on misinformation sites (Table 3 Col 3). Among the “No” subgroup, increases were 39.98 (certain; P=0.049) and 144.25 (uncertain; P=0.022) in belief measures (Table 4 Cols 1–2). - Platform-solution demand: Overall null effects (Table 3 Cols 4–6), but among participants who were uncertain and unaware of their company’s involvement, demand increased by 36 percentage points (β=0.36; 95% CI [0.11, 0.61]; P=0.008; n=68) (Table 4 Col 4).

The findings show that advertising is the dominant monetization model for misinformation websites and that programmatic advertising via digital platforms substantially amplifies advertisers’ presence on such sites. Consumers penalize companies whose ads appear on misinformation outlets by switching away, even when made aware of platforms’ role in ad placements. Rank-based transparency steers consumers toward lower-misinformation advertisers, and platform-focused information mobilizes voice directed at platforms. Decision-makers are often unaware of their own companies’ inadvertent financing of misinformation yet express high demand for information and solutions. Together, these results suggest that simple, scalable, information-based interventions can reduce the supply-side financial incentives of misinformation: providing advertisers with transparent, ongoing reporting of where ads appear and providing consumers with disclosures or rankings about advertisers’ presence on misinformation sites at decision points (e.g., during shopping or alongside ads). Such measures could complement existing demand-side approaches, reduce the need for episodic boycotts, and create persistent reputational incentives for advertisers to avoid misinformation sites. Platforms are well-positioned to implement these disclosures, though their incentives may be misaligned; thus policy-driven transparency requirements (e.g., EU DSA, US bills like Honest Ads Act and CTDA) may help. Regulation-by-information parallels in other industries suggest such transparency can be effective in addressing externalities. Overall, aligning advertiser and consumer preferences via transparency could reduce inadvertent ad revenue flows to misinformation outlets and, over time, diminish their viability.

This study combines large-scale observational evidence with incentive-compatible experiments to demonstrate that: (1) misinformation websites are predominantly financed by advertising; (2) digital advertising platforms greatly increase the likelihood that companies’ ads appear on misinformation sites; (3) consumers penalize companies for advertising on misinformation outlets, even when platforms are implicated; and (4) decision-makers are often unaware of their companies’ involvement but demand information and solutions once informed. The authors propose two low-cost, scalable interventions: (a) platform-enabled transparency for advertisers on where their ads appear, allowing better control to avoid misinformation sites; and (b) transparency for consumers via disclosures and rankings that can shift demand away from high-intensity misinformation advertisers. Future research should: (i) field-test these interventions over longer horizons to quantify revenue impacts on misinformation outlets and validate generalizability across a broader set of advertisers; (ii) assess advertiser responses when decision-makers hold correct beliefs about prevalence and risks; and (iii) analyze platform incentives and the net societal benefits vs. foregone ad revenues. A multi-pronged approach combining platform tools, advertiser practices, and policy-driven transparency may effectively curb the financial incentives for misinformation supply.

- Experimental context and external validity: Consumer outcomes were measured in a gift-card choice setting; while incentive-compatible, real-market impacts may differ. Longer-term behavioral persistence and field outcomes (e.g., actual sales or ad allocation changes) were not directly measured. - Subsample analyses: Some decision-maker intervention effects rely on small, exploratory subsamples not pre-registered (e.g., uncertain and unaware subgroup), warranting cautious interpretation. - Multiple testing: No adjustments were made for multiple comparisons, raising potential false-positive risk across numerous outcomes. - Measurement and data coverage: Advertising appearances were derived from Moat Pro crawls, which, while industry-standard, may not capture all placements uniformly across sites/times. The misinformation site lists, though curated (NewsGuard, GDI, prior work), may involve classification limitations. - Platform incentives and implementation: While platforms are well-situated to provide transparency, their economic incentives to implement these interventions are uncertain and were not tested. - Generalizability: Executive respondents were drawn from executive education alumni; although most held relevant roles, the sample may not represent all firms or industries globally.