Does AI search bring more people into online discussions?
How did participation in Reddit communities change when Google expanded its AI search summaries? Communities eligible to be cited had an estimated 12.4% more daily commenters relative to the comparison group. For a business with a community, eligibility to appear as a search source may be linked to more people joining discussions.
ACADEMIC WORK REVIEWED
The Impact of AI Search on the Online Content Ecosystem: Evidence from Google and Reddit
ResearchersPeibo Zhang, Ruomeng Cui, Dennis J. Zhang
arXiv · 20 June 2026 · v3, peer review not reported · Preprint · Source language: English
English review: 29 October 2026 · The source date appears in the citation above.
Editorial review draft
The source has been checked; the review still carries its original editorial status. DijitalPi did not conduct a customer experiment for this paper.
LET'S READ THE RESEARCH TOGETHER
We first explain the researchers' question, method and findings. We then discuss how to interpret the results, clearly separating DijitalPi's commentary from the source.
01 / WHAT DID THE RESEARCHERS WANT TO UNDERSTAND?
How does participation in communities eligible for AI search citations vary by content type and search interface?
AI Overviews are the AI summaries Google adds to search results. The researchers examine whether Google’s AI summaries replace Reddit discussions or are associated with greater participation. They compare factual content with discussions centred on personal experience and opinion. They also investigate whether this relationship persists when search moves from summaries to a conversational interface.
Explaining the context · DijitalPi commentary
DijitalPi commentary: For content budgeting, appearing in a search answer is only part of the question. Whether people join the conversation matters too. This study provides a reason to assess factual posts and experience discussions separately. Before treating community participation as commercial value, however, a business needs to measure its own outcomes, including clicks, enquiries and sales.
02 / HOW WAS THE RESEARCH CONDUCTED?
How did they test the question?
Zhang, Cui and Zhang examined Arctic Shift records for 105,012 public Reddit communities with at least 10 submissions in December 2023. The dataset covers 1 January 2024-31 July 2025 and contains 60,696,936 community-day observations. The AI Mode analysis extends through December 2025. The researchers used an external change they did not introduce themselves, which they treat as a natural experiment. They compared 70,683 safe-for-work communities with 34,329 communities containing adult or sensitive material. The main event date is 15 August 2024, when AI Overviews expanded to the United Kingdom, India, Japan, Indonesia, Mexico and Brazil. The data are not separated by country, and Türkiye is not on that expansion list.
Understanding the method · DijitalPi commentary
Difference-in-differences compares the change before and after an event across two groups. Both groups can appear in Google’s organic search results, but only safe-for-work communities are eligible for citations in AI Overviews. Eligibility does not mean that every community was actually cited. Assignment was not random. The analysis accounts for stable community characteristics and shared changes specific to each day. The main measures are daily comments and the number of distinct users posting at least one comment that day. Reported percentages divide the estimated effect by the eligible communities’ average before the expansion.
For the content comparison, Gemini 3 Flash classified communities as factual or experience-based using their names, titles and descriptions. It classified 104,975 communities. 37 remained unclassified because of safety filtering or responses that could not be assigned to either category. A manual audit of 1,000 post titles on 2 April 2026 supported the assumption that the control group was excluded from citations. This check took place after the observation period. It is not a direct record of every search result throughout that period.
03 / RESEARCH FINDINGS
The positive association with participation varies by content type and interface
Following the AI Overviews expansion, the researchers estimated daily comments in eligible communities to be 12.0% higher and daily unique commenters 12.4% higher relative to the control group. The pre-expansion averages underlying those percentages were 105.4 comments and 53.6 commenters per day. The authors interpret the similar magnitudes as more people joining discussions rather than existing commenters writing more. They found no statistically significant results for submissions or unique submission authors. Both factual and experience-based communities showed positive, significant results. Estimated effects in experience-based communities were 2.3 times larger for comments and 2.8 times larger for commenters. Google AI Mode is a search service that uses a conversational interface. Following its global rollout on 21 August 2025, experience content’s additional advantage over factual content turned slightly negative for comments and fell by 59% for commenters, while remaining positive. That 59% does not describe a fall in the total number of commenters. The estimated base effect for factual communities also weakened.
Daily comments
12.0%The estimated effect following AI Overviews relative to the control group, expressed against eligible communities’ earlier average comment count.
Daily unique commenters
12.4%The estimated effect relative to the control group. This counts distinct users posting at least one comment that day, not visitors.
DijitalPi commentary: Experience sharing may merit a place in a content plan, but these findings do not support treating it as a permanent search advantage. Tracking factual and experience content separately, counting distinct participants alongside comments, and recording interface changes could make the assessment more informative. Do not use the paper’s percentages as expected returns for your business.
The groups differ at baseline: safe-for-work communities tend to be larger and more discussion-oriented, while many adult or sensitive communities focus on media sharing with less textual interaction. Commenter counts also showed a negative trend before the expansion. A check for an effect at an unrelated event date found a negative, significant result at Reddit’s IPO. The authors say these patterns would reduce the main estimate, but they remain relevant when assessing a causal interpretation. More comments do not directly measure users arriving from Google.
05 / CONCLUSION AND OPEN QUESTIONS
What did we learn, and what do we still not know?
This study is a preprint, a research manuscript that has not yet undergone peer review. It measured participation within Reddit, not website visits or sales. The study shows a positive relationship between eligibility for Reddit citations and discussion participation, with that relationship varying by content type and interface. The reviewed v3 is an arXiv preprint with no reported peer review. Findings from text-based forums cannot be transferred directly to brand websites, video or music, Türkiye or Turkish content; the observation window after AI Mode is also limited. The authors are university academics, with no company affiliation stated. No funding or competing-interest declaration was found in the source. That does not establish an absence of funding or competing interests.
Our open question at DijitalPi: do experience posts in your Turkish-language community attract more distinct participants, and does that pattern persist when the search interface changes?
We created these scenarios to make the topic concrete. They are not cases from the paper or measured client results.
EXAMPLE 01
E-commerce: product facts and usage experiences
An illustrative camping equipment retailer hosts technical product questions and accounts of customers’ experiences in its community.
Which content type is associated with more people joining the conversation?
Open the recommendation for example 01+
Track comments and unique commenters separately for each content type
Use predefined rules to distinguish technical information from customer experience posts. Record daily comments and distinct commenters for each group, marking search interface changes on the same timeline. Assess these results separately from website clicks and orders. The Reddit findings do not guarantee the same outcome for your shop.
EXAMPLE 02
B2B software: feature questions and usage discussions
An illustrative B2B software business publishes feature explanations and accounts of working practices in its user forum.
Do more comments mean that a broader group of people is participating?
Open the recommendation for example 02+
Count distinct participating accounts alongside total comments
Keep daily records of comments and unique commenters for feature questions and usage discussions. Examine both measures to help distinguish more concentrated commenting from changes in participant numbers. These records do not measure enquiries or contracts. The study’s finding needs separate testing in your B2B forum.
EXAMPLE 03
Tourism: travel information and travellers’ experiences
An illustrative tourism business shares transport information and travellers’ route experiences in its community.
Will the participation advantage of experience posts persist when search changes?
Open the recommendation for example 03+
Track factual and experience content separately over time
Use a consistent recording process for daily comments and unique commenters in both content groups. Record campaign and publishing schedules too, so business activities coinciding with search interface changes remain visible. A before-and-after change alone does not establish its cause. The study’s observation window after AI Mode was also limited.
TRY IT WITH YOUR TEAM
Count comments and commenters separately
Plan a shared record for factual and experience content, tracking comments, unique commenters and search interface changes. Assess clicks and sales separately from these measures. This measurement proposal has not yet been implemented by DijitalPi.
Peibo Zhang, Ruomeng Cui, Dennis J. Zhang Goizueta Business School, Emory University (Peibo Zhang, Ruomeng Cui); Olin Business School, Washington University in St. Louis (Dennis J. Zhang).
The academic work belongs to the researchers named above. This page contains DijitalPi's explanatory review and original business examples; it is not a full translation of the paper.
Cookie preferences