New research shows what happens to advertising performance when AI takes the reins.
Listen to this article:
Generative AI has enabled a wave of AI-integrated business products, making it easier than ever for organizations to automate complex business tasks across IT, marketing, operations, and beyond. But a capability can be impressive without creating measurable business value. That issue sits at the center of the new working paper “Automated Comment Moderation Enhances Social Media Advertising Performance,” co-written by 2025-2026 HBS AI Institute Associate Julian De Freitas. De Freitas and his co-author Jiwoon Park ask what happens after an automated system is deployed around live advertising: Do consumers behave differently? Do campaigns perform better? And which part of the system is responsible? Through large-scale field studies and online experiments, their findings reveal how algorithmic interventions shape consumer trust and bottom-line commercial returns.
Key Insight: The Comment Section is Part of the Advertisement
“[W]e find converging evidence that automated comment management improves ad performance.” [1]
Whenever a brand launches a campaign on social media, the comments beneath it become part of the context through which consumers encounter the brand. Earlier market research had already established that consumers hold brands partly responsible for whatever appears next to them; Park and De Freitas specifically studied automated comment management that can hide harmful comments and respond to others, asking whether that improves performance. Their first field study followed a financial services company running a Trustpilot campaign that drew more than 3.4 million link clicks over ten months. Five months in, they turned on automated comment management, and registration completions per ad click rose from 1.67% to 1.94%. Because this initial test relied on sequential tracking, it provided less rigorous causal evidence than a concurrent one. Therefore, for their second field study the researchers ran concurrent campaigns on Facebook and Instagram, with automated comment management switched on for one and off for the other. With roughly $10,000 in ad spend in each condition, the AI-managed campaigns generated 22 purchases versus 15, producing return on ad spend of 0.678 versus 0.459 without it.
Key Insight: More AI Doesn’t Mean More Value
“To use an analogy, one might still frequent a café where the waiter does not talk much, but not if the windows are broken.” [2]
The AI systems in the field studies incorporated both engagement (replying to comments) and moderation (hiding harmful ones, such as spam or unconstructive brand attacks). Park and De Freitas also used controlled experiments to separate these two activities and see which was driving the gains. In a 2×2 design using a fictitious hair dryer brand on an Instagram-style post, purchase intention averaged 47.2 when moderation was present versus 36.9 when absent. Brand engagement produced no significant effect. The mechanism appears to involve trust. Harmful comments can become signals about the brand itself: a hostile or disorderly comment environment may make the company seem inattentive or poorly managed. By eliminating these visible signs, automated moderation preserves perceived brand trust, which directly drives purchase decisions.
Key Insight: The Transparency Trap
“In short, platform transparency backfires for both brand and platform trust.” [3]
Moderation worked when consumers experienced a cleaner comment environment, but when a platform explicitly revealed that the brand had hidden hostile comments directed at itself, the effect reversed. In that experiment, the purchase intention was 37.5 with moderation versus 45.2 without it. Brand trust also fell, suggesting consumers could interpret the hiding of comments as evidence of self-interested manipulation rather than responsible management. This key contrast is refined by two qualifications. First, the same clear reversal did not occur when the hidden comments contained universally harmful discourse, such as profanity or hate speech. Second, a brand that published its own moderation policy on its profile page kept the full benefit of moderation.
Why This Matters
AI implementation creates value when the surrounding workflows convert capability into better business outcomes. That makes this research especially relevant because it points to a corner of marketing almost nobody funds: by the authors’ count, roughly 19 in 20 brands buy no comment-management tooling at all. The broader lesson for leaders is that some of the most promising AI opportunities may likewise be hiding in places that have historically received less investment or attention. The next stage of AI transformation may be to identify overlooked parts of your business where AI can change the conditions around the work, and then test whether that changes moves an outcome the business actually values.
Bonus
AI doesn’t create value in isolation, making it important to look beyond whether an AI system performs its assigned task and ask what other consequences emerge around that optimization. For a related look at how using LLMs to make more content more engaging can simultaneously make it more polarizing, check out Stop Trading Trust for Traffic.
References & Notes
[1] Park, Jiwoon, and Julian De Freitas, “Automated Comment Moderation Enhances Social Media Advertising Performance,” Harvard Business School Working Paper, No. 27-011 (July 2026): 39.
[2] Park and De Freitas, “Automated Comment Moderation Enhances Social Media Advertising Performance,” 8.
[3] Park and De Freitas, “Automated Comment Moderation Enhances Social Media Advertising Performance,” 29.
This article was created through a human-led, AI-assisted process.
Meet the Authors
Jiwoon Park is a doctoral student in Marketing at Harvard Business School.
Julian De Freitas is an Assistant Professor of Business Administration in the Marketing Unit and Director of the Ethical Intelligence Lab at Harvard Business School, and Associate HBS AI Institute. His work sits at the nexus of AI, consumer psychology, and ethics.