The Impact of AI Search on the Online Content Ecosystem: Evidence from Google and Reddit
Pith reviewed 2026-05-21 08:18 UTC · model grok-4.3
The pith
Google AI Overviews raise daily comments and active users by about 12 percent on safe Reddit communities relative to restricted ones, with gains concentrated in experience-based discussions.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
Using a difference-in-differences design that compares safe-for-work and not-safe-for-work Reddit communities before and after the rollout of Google AI Overviews, the study finds that inclusion in the AI summaries raises daily comments by 12.0 percent and commenting users by 12.3 percent in SFW communities relative to NSFW ones. The increase is driven by experience-based content such as personal opinions and advice. Introducing a more conversational AI Mode later removes most of these gains.
What carries the argument
Difference-in-differences comparison between SFW Reddit communities, which appear in Google AI Overviews, and NSFW communities, which do not, despite both being indexed in organic search.
If this is right
- AI search summaries can increase user activity on source discussion platforms when they reference experience-oriented content.
- The positive effects are limited to opinion and advice threads and do not extend to fact-based queries.
- Switching to a conversational AI interface largely cancels the engagement gains seen with static summaries.
- Platform outcomes depend on both the type of content and the specific design choices in the AI search tool.
Where Pith is reading between the lines
- Platforms may gain from emphasizing personal-experience material if AI summaries continue to favor it.
- Fact-oriented sites could face different substitution effects than community forums under the same AI designs.
- Testing the same pattern on other discussion sites or content types would show how general the interface dependence is.
Load-bearing premise
The parallel trends assumption holds between SFW and NSFW Reddit communities, so that any differential change in engagement after AI Overviews can be attributed to the inclusion of SFW content in the summaries rather than to other time-varying differences between the two groups.
What would settle it
Finding that engagement trends in SFW and NSFW communities remain parallel after the AI Overviews rollout, or observing comparable increases in NSFW communities that should not receive the summaries.
read the original abstract
Search engines traditionally complement online content platforms by directing users seeking information to external websites. The emergence of generative AI search tools that summarize answers directly on the results page may disrupt this relationship by making visits to source platforms optional. We study this question using Google AI Overviews and Reddit, one of the largest online discussion platforms. Our identification exploits Google's content moderation policy: Safe-for-Work (SFW) Reddit communities are indexed by Google organic search and surfaced in Google AI Overviews, while Not-Safe-for-Work (NSFW) communities, though indexed by organic search, are prohibited from being referenced in AI Overview summaries. Using a difference-in-differences design, we find that AI Overviews increase engagement in SFW communities: daily comments rise by 12.0 percent and the number of commenting users by 12.3 percent relative to NSFW communities. The effects are concentrated in experience-based discussions (opinions, advice, and personal experiences) rather than fact-based information. However, the subsequent introduction of Google AI Mode, which allows users to interact conversationally with the AI summary, largely eliminates these gains in experience-based content. These results suggest that the effects of AI search depend critically on interface design and types of content.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript investigates the effects of Google's AI Overviews on engagement in Reddit communities by leveraging the platform's content moderation policies. SFW communities are included in AI summaries while NSFW are not, allowing a difference-in-differences comparison. The authors report that AI Overviews lead to a 12.0% increase in daily comments and 12.3% increase in commenting users in SFW relative to NSFW communities, with stronger effects in experience-based discussions that are mitigated by the later introduction of conversational AI Mode.
Significance. Should the identification strategy prove robust, these results would be significant for the field of information retrieval and online platforms. They provide causal evidence that AI-generated summaries in search can increase rather than cannibalize engagement with certain types of user-generated content, particularly experiential discussions. This has implications for how search engines design AI features and how content platforms adapt to AI search, contributing empirical insights to debates on the impact of generative AI on the web ecosystem.
major comments (2)
- The core DiD identification relies on the assumption that, absent AI Overviews, engagement trends in SFW and NSFW communities would have been parallel. Given documented differences in user demographics, moderation practices, and vulnerability to external shocks (e.g., Reddit policy changes or cultural events), this assumption requires explicit validation through pre-trend analyses and robustness checks, which are not detailed in the abstract and must be prominently featured in the results section to support the causal interpretation of the 12% effects.
- The reported effects of 12.0 percent on comments and 12.3 percent on users are central to the claims, but the abstract lacks information on confidence intervals, sample sizes, fixed effects used, or tests for differential pre-trends in experience-based vs. fact-based threads. These details are load-bearing for assessing whether the concentration in experience-based discussions is statistically reliable.
minor comments (1)
- The abstract could more clearly state the time period of the study and the exact rollout dates of AI Overviews and AI Mode to contextualize the findings.
Simulated Author's Rebuttal
Thank you for the valuable feedback on our manuscript. We address the major comments below and have revised the paper to incorporate additional analyses and details as suggested.
read point-by-point responses
-
Referee: The core DiD identification relies on the assumption that, absent AI Overviews, engagement trends in SFW and NSFW communities would have been parallel. Given documented differences in user demographics, moderation practices, and vulnerability to external shocks (e.g., Reddit policy changes or cultural events), this assumption requires explicit validation through pre-trend analyses and robustness checks, which are not detailed in the abstract and must be prominently featured in the results section to support the causal interpretation of the 12% effects.
Authors: We agree with the referee that validating the parallel trends assumption is essential for a credible DiD identification. Although our original manuscript included some robustness checks, we acknowledge that pre-trend analyses were not sufficiently prominent. In the revised manuscript, we have added a new subsection in the Results section that features event-study plots, formal tests for pre-trends, and additional robustness checks addressing differences in user demographics, moderation practices, and potential external shocks. These revisions strengthen the support for our causal interpretation of the effects. revision: yes
-
Referee: The reported effects of 12.0 percent on comments and 12.3 percent on users are central to the claims, but the abstract lacks information on confidence intervals, sample sizes, fixed effects used, or tests for differential pre-trends in experience-based vs. fact-based threads. These details are load-bearing for assessing whether the concentration in experience-based discussions is statistically reliable.
Authors: We appreciate the point regarding the need for more statistical details in the abstract. In the revision, we have updated the abstract to report the 95% confidence intervals around the 12.0% and 12.3% effects, along with sample sizes and the fixed effects specification. We have also included results from tests for differential pre-trends in experience-based versus fact-based threads and highlighted these in the results section to demonstrate the statistical reliability of the concentration in experiential content. revision: yes
Circularity Check
No circularity: purely empirical DiD leveraging external policy variation
full rationale
The paper presents an empirical difference-in-differences analysis that exploits Google's pre-existing content moderation policy (SFW communities indexed and summarized in AI Overviews; NSFW communities indexed but excluded from summaries) as the source of identification. No derivations, equations, first-principles predictions, or fitted parameters are claimed. The central estimates (12.0% rise in comments, 12.3% rise in commenting users) are direct statistical comparisons of observed post-rollout changes; they do not reduce to quantities defined inside the paper by construction. Parallel-trends is an empirical identifying assumption, not a definitional or self-referential step. No self-citation chains or ansatzes are load-bearing for the result.
Axiom & Free-Parameter Ledger
axioms (1)
- domain assumption Parallel trends assumption holds between SFW and NSFW Reddit communities in the absence of AI Overviews.
Reference graph
Works this paper leans on
-
[1]
Journalism, Media, and Technology Trends and Predictions 2026 , year =
work page 2026
- [2]
- [3]
-
[4]
Novet, Jordan and Elias, Jennifer , title =. 2025 , howpublished =
work page 2025
-
[5]
Free Speech Coalition, Inc.\ v.\ Paxton , year =
- [6]
-
[7]
Draft Report on Copyright and Generative Artificial Intelligence: Opportunities and Challenges , year =
- [8]
- [9]
- [10]
-
[11]
Gholami, Samira and Firullo, Cristiana and Cheyre, Cristobal and Acquisti, Alessandro , title =. 2026 , note =
work page 2026
- [12]
- [13]
-
[14]
Burtch, Gordon and Lee, Dokyun and Chen, Zhichen , title =. Scientific Reports , year =
-
[15]
Tai and Lambrecht, Anja and Hollenbeck, Brett , title =
Padilla, Nicolas and Lam, H. Tai and Lambrecht, Anja and Hollenbeck, Brett , title =. 2025 , note =
work page 2025
-
[16]
2025 , howpublished =
work page 2025
- [17]
- [18]
- [19]
-
[20]
Journal of Political Economy , year =
Nelson, Phillip , title =. Journal of Political Economy , year =
-
[21]
Darby, Michael R. and Karni, Edi , title =. Journal of Law and Economics , year =
- [22]
- [23]
-
[24]
Bakos, Yannis and Brynjolfsson, Erik , title =. Management Science , year =
-
[25]
Journal of the European Economic Association , year =
Rochet, Jean-Charles and Tirole, Jean , title =. Journal of the European Economic Association , year =
-
[26]
Parker, Geoffrey G. and Van Alstyne, Marshall W. , title =. Management Science , year =
-
[27]
RAND Journal of Economics , year =
Armstrong, Mark , title =. RAND Journal of Economics , year =
-
[28]
Benkler, Yochai , title =
-
[29]
Chevalier, Judith A. and Mayzlin, Dina , title =. Journal of Marketing Research , year =
-
[30]
Dellarocas, Chrysanthos , title =. Management Science , year =
- [31]
-
[32]
American Economic Review , year =
Zhang, Xiaoquan and Zhu, Feng , title =. American Economic Review , year =
-
[33]
Jakesch, Maurice and Hancock, Jeffrey T. and Naaman, Mor , title =. Proceedings of the National Academy of Sciences , year =
-
[34]
Agrawal, Ajay and Gans, Joshua and Goldfarb, Avi , title =
- [35]
- [36]
-
[37]
Peng, Sida and Kalliamvakou, Eirini and Cihon, Peter and Demirer, Mert , title =. 2023 , note =
work page 2023
-
[38]
and Lifshitz-Assaf, Hila and Kellogg, Katherine C
Dell'Acqua, Fabrizio and McFowland, Edward and Mollick, Ethan R. and Lifshitz-Assaf, Hila and Kellogg, Katherine C. and Rajendran, Saran and Krayer, Lisa and Candelon, Fran. Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of. 2023 , note =
work page 2023
- [39]
- [40]
-
[41]
Lynch, John G. and Ariely, Dan , title =. Marketing Science , year =
-
[42]
Journal of Marketing Research , year =
Edelman, Benjamin and Lai, Zhenyu , title =. Journal of Marketing Research , year =
-
[43]
International Journal of Industrial Organization , year =
Hagiu, Andrei and Wright, Julian , title =. International Journal of Industrial Organization , year =
- [44]
- [45]
- [46]
- [47]
-
[48]
Large Language Models Reduce Public Knowledge Sharing on Online. PNAS Nexus , year =
- [49]
- [50]
-
[51]
Calzada, Joan and Gil, Ricard , title =. Marketing Science , year =
-
[52]
Journal of Economics and Management Strategy , year =
Chiou, Lesley and Tucker, Catherine , title =. Journal of Economics and Management Strategy , year =
-
[53]
American Economic Journal: Microeconomics , year =
Jeon, Doh-Shin and Nasr, Nikrooz , title =. American Economic Journal: Microeconomics , year =
-
[54]
Athey, Susan and Mobius, Markus M. and P. The Impact of Aggregators on Internet News Consumption , journal =. 2021 , volume =
work page 2021
-
[55]
and Parker, Geoffrey and Van Alstyne, Marshall W
Eisenmann, Thomas R. and Parker, Geoffrey and Van Alstyne, Marshall W. , title =. Strategic Management Journal , year =
- [56]
-
[57]
Corradini, Enrico and Nocera, Antonino and Ursino, Domenico and Virgili, Luca , title =. Information Sciences , year =
- [58]
- [59]
-
[60]
Journal of Econometrics , year =
Callaway, Brantly and Li, Tong and Oka, Tatsushi , title =. Journal of Econometrics , year =
-
[61]
Koenker, Roger and Bassett, Gilbert , title =. Econometrica , year =
-
[62]
Quarterly Journal of Economics , year =
Bertrand, Marianne and Duflo, Esther and Mullainathan, Sendhil , title =. Quarterly Journal of Economics , year =
-
[63]
Review of Economic Studies , year =
Rambachan, Ashesh and Roth, Jonathan , title =. Review of Economic Studies , year =
- [64]
-
[65]
Rahwan, Iyad and Cebrian, Manuel and Obradovich, Nick and Bongard, Josh and Bonnefon, Jean-Fran. Machine behaviour , journal =. 2019 , volume =
work page 2019
-
[66]
Brinkmann, Levin and Baumann, Fabian and Bonnefon, Jean-Fran. Machine culture , journal =. 2023 , volume =
work page 2023
-
[67]
Eloundou, Tyna and Manning, Sam and Mishkin, Pamela and Rock, Daniel , title =. Science , year =
- [68]
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.