Visibility Is an Asset, but It Is Not a Neutral One

This series asked whether AI search helps community content or quietly replaces it, and what it means when the systems allocating visibility can also be influenced by the text they evaluate. The most defensible answer is conditional. Google AI Overviews increased participation in the Reddit communities eligible for its references, especially where value depended on experience and interaction. Google AI Mode later reduced that advantage by bringing more follow-up work into the search page. A separate controlled study shows that LLM rerankers can be steered by natural-looking candidate text. These findings belong in one model, but they must not be turned into one undifferentiated claim about “AI.”

The relationship is conditional by design

The Google–Reddit evidence changes the starting point for the debate. The relevant question is not whether a summary contains a citation. It is what the summary does to the user’s remaining task. A static AI Overview can answer a factual question directly, reducing the need to visit the cited source. It can also expose a community whose value lies in personal experience, disagreement, advice, or interaction—value that is difficult to compress into a short answer. In that case, the citation can function as a discovery channel.

The study’s main result is evidence for the second path in the observed setting. After the international expansion of AI Overviews, daily comments in SFW Reddit communities increased by 12.0 percent relative to NSFW communities, and daily unique comment authors increased by 12.4 percent. The researchers use the difference between communities eligible for AI Overview references and communities excluded by Google’s policy as a treatment contrast. The estimates are interpreted causally under the study’s difference-in-differences assumptions, and the reported checks support the direction and approximate size of the effect.

Those numbers measure participation, not every stage of discovery. The closeness of the comment and author changes suggests that more people entered discussions rather than only existing commenters producing more messages. But the data do not directly show how many people saw an AI Overview, clicked a citation, read a thread, joined a community, or returned later. A community-level engagement effect is not a click-through rate, a revenue result, a quality judgment, or proof of long-run sustainability.

The distributional pattern makes the visibility asset more specific. Smaller and small-to-medium communities gained more in percentage terms, while the largest communities gained more in absolute activity. That is consistent with AI search widening the discovery frontier beyond the most prominent destinations. A niche community can experience a meaningful change from a small number of additional participants, while a large community can absorb many additional comments without a comparable percentage shift. The result suggests that machine-mediated visibility can redistribute opportunities for attention, not merely intensify the attention already held by incumbents.

The content comparison explains why that redistribution might occur. The study’s experience-good communities show treatment effects 2.3 times the search-good effect for comments and 2.8 times for comment authors. The experience premium is not a rule that every discussion platform benefits or that factual content always loses. Within Reddit, both categories show positive effects. The comparison instead identifies where the discovery channel is stronger: where the source offers interaction-dependent value that a static summary cannot fully deliver.

This distinction is useful because it avoids two symmetrical mistakes. The first is to assume that citations guarantee a healthy relationship between search and source. A citation can remain on the results page even when the summary has removed the reason to visit. The second is to assume that summarization necessarily destroys the source. A summary can answer the portable factual layer while making a relevant community easier to find. The source platform still has to earn the user’s attention after discovery, but AI search may lower the cost of reaching it.

The balance is not fixed, because the interface is part of the mechanism. Google AI Mode provides the clearest evidence in these sources. AI Overviews presented static summaries; AI Mode allowed follow-up questions within the search page. After the August 21, 2025 global rollout, the experience-good premium fell by 59 percent for comment authors and by 102 percent for comments. The comments premium moved slightly negative, while the author premium remained smaller but positive.

The two margins show what it means for an interface to absorb value unevenly. A conversational search page can take over some of the elaboration that would otherwise appear in a discussion thread. It may reduce the depth of redirected engagement without eliminating every reason to participate.