80% of Reddit posts cited by ChatGPT and Perplexity have fewer than 20 upvotes (Semrush, November 2025). LLMs ignore your engagement signals. Your Reddit strategy must change.
Key takeaways: LLMs heavily cite Reddit, but they don’t weight by upvotes. 80% of cited posts have fewer than 20 votes. Your strategy must target Q&A format and clarity, not virality.
Reddit’s growth in 2024-2025: why the platform dominates AI sources
Reddit has had a meteoric trajectory since 2024. Sistrix measures a 1,328% growth in the domain’s SEO visibility between July 2023 and April 2024 in the US market. The platform moved from 68th to 5th place among the most visible websites globally.
This dominance stems from two simultaneous phenomena. The first is the data partnership signed between Google and Reddit in 2024, giving Google official access to Reddit’s APIs for $60 million annually. This agreement guarantees the massive integration of Reddit conversations into SERPs and AI models.
The second phenomenon directly concerns LLMs. The Semrush study published in November 2025, conducted on 248,000 Reddit URLs cited in 217,000 prompts across Google AI Mode, Perplexity and ChatGPT Search, places Reddit at the top of cited sources:
- Reddit is the #1 source on Perplexity (4% citation share, average position 3.4 in source list)
- Reddit is the #2 source on SearchGPT (13% share, average position 6.7)
- Reddit is the #3 source on Google AI Mode (9% share, average position 8.8)
The platform surpasses Wikipedia, LinkedIn and Forbes in several configurations. Transactional and comparative queries concentrate most Reddit citations, where users seek human experience over institutional opinions.
The Semrush 2025 study: what really determines citation by LLMs
The Semrush study debunks a widespread agency assumption: upvotes don’t determine citation by LLMs. LLMs don’t weight Reddit like a traditional social platform. They treat threads as semantic content, not as popularity signals.
The numbers are unambiguous:
- 80% of cited posts have fewer than 20 upvotes, and 70% have fewer than 20 comments
- The median engagement of cited posts is 5 to 8 upvotes and 11 to 19 comments
- The median age of cited posts is around 900 days (2.5 years), well beyond classic virality windows
- The median length of a cited post is ~80 words, very short
The actual extraction criterion is semantic relevance, not popularity. Semrush measures a cosine similarity of 0.53-0.54 between AI responses and cited Reddit posts, versus only 0.04-0.05 between the user prompt and the Reddit post. LLMs draw from Reddit by meaning, not by keywords.
Operational consequence: a niche, old, low-voted but clearly structured thread can outperform a recent viral post in AI citations. Traditional Reddit virality logic doesn’t apply. LLM extraction logic does.
The 4 characteristics of a Reddit thread cited by ChatGPT, Perplexity or Gemini
The Semrush study identifies four recurring markers in the Reddit threads LLMs cite. These markers define effective Reddit GEO strategy in 2026.
Explicit Q&A format
Question/answer structured threads represent more than 50% of Reddit citations in AI responses, according to the Semrush study. Comparisons (X vs Y) and open discussions complete the 75%. Other formats (announcements, memes, news) total less than 25% of citations.
This Q&A dominance is explained by LLM extraction mechanics. An explicit question followed by a reasoned answer constitutes an autonomous chunk, directly reusable in a generative response. Narrative or conversational threads without clear structure largely escape LLM radar.
Conciseness and clarity
The median length of cited Reddit posts is 80 words. Threads dense in useful information, without digression, capture more citations than long developments. This conciseness echoes the principles of GEO chunking: an autonomous block of 150 to 300 words maximum with one idea per paragraph.
Concrete implication: a short but precise Reddit comment that directly answers the thread’s question has more chances of being cited than a long argumentative thread. The writing reflex to favor on Reddit isn’t extended demonstration, it’s direct answers.
Topical match over virality
LLMs choose threads to cite based on semantic proximity to the user’s question, not on thread popularity. The 0.53-0.54 cosine similarity documented by Semrush proves models paraphrase relevant content, not that they copy popular content.
For a brand, this means investing in a niche thread well-aligned with a specific search intent beats trying to viralize a generic post. The work consists of identifying the precise questions your customers ask LLMs, then covering them via Reddit.
Relative freshness or durable anchoring
The median age of cited posts (~900 days) reveals a counterintuitive fact: Reddit has long memory in LLMs. A 2023 thread can be cited in 2026 if its content remains relevant and its formulation lends itself to extraction.
This durability contrasts with the 30-day freshness observed on web pages cited by ChatGPT (Ahrefs, 2025). Reddit functions as a collective memory indexed by LLMs, not as a news feed. Investing a Reddit thread produces an effect that can last 2 to 3 years, sometimes more.
The 4 Reddit activation levers for GEO in 2026
A serious Reddit GEO strategy in 2026 rests on 4 levers calibrated to the Semrush data. I apply this structure with my agency clients and at Cockpyt AI.
| Lever | Concrete action | Measurable GEO impact |
|---|---|---|
| Expert participation in Q&A threads | Sourced and factual answers in 5 to 10 threads per month on targeted subreddits | Possible LLM citations within 30 to 90 days |
| Official brand subreddit | Creating an owned space with FAQ, guides, regular AMAs | Structured source LLMs can cite as brand authority |
| Continuous mention monitoring | Surveillance of existing threads about your brand, response to false positives | Reduced risk of AI hallucinations based on outdated threads |
| AI Share of Voice tracking | Measuring citation frequency of your brand + Reddit in LLMs | Quantified data to pilot and arbitrate Reddit budget |
The first lever is the most accessible. Expert participation requires no advertising investment, just writing time and verifiable expertise. Targeting 5 to 10 threads per month on 2 or 3 industry subreddits remains realistic for an SMB.
The second lever (official subreddit) demands more commitment. HubSpot, Notion, Stripe and several tech brands have opened their subreddit successfully. Return on investment is measured in years, not months, but constitutes a durable owned asset that LLMs integrate as an authority source.
Attention on the French market is needed: Reddit is significantly less developed in French than in English. Active French-language subreddits cover a fraction of topics available in English. For a French B2B brand also targeting English-speaking prospects, English Reddit investment remains more profitable than French Reddit in 2026.
How to measure Reddit’s real impact on your AI Share of Voice
Most brands pilot their Reddit strategy blind, without measuring what LLMs actually cite. Traditional KPIs (upvotes, comments, karma) don’t measure GEO performance. They measure social engagement, which isn’t correlated to AI citation according to Semrush.
The measurement method I apply at Cockpyt AI:
- Pre-action baseline: capturing AI Share of Voice across 30 to 100 strategic prompts where Reddit is likely to surface (comparative questions, experience-based queries, product recommendations)
- Identifying existing Reddit citations: extracting Reddit URLs that appear in LLM responses containing your brand
- Targeted action: expert participation or thread creation on subjects identified as deficit areas
- Measurement at 30, 60 and 90 days: new panel execution to identify new Reddit citations generated by your actions
Without a dedicated tool, measurement plateaus at around a dozen manual prompts. The natural variability of LLM responses requires several executions per prompt to obtain a robust signal. Cockpyt AI automates this measurement and historicizes evolution, enabling correct attribution of a Reddit action’s impact on your overall AI visibility.
The Cockpyt Score, a synthetic indicator of your authority in LLMs, integrates Reddit citations as one of the third-party sources contributing to your visibility, in line with the Stacker 2026 study documenting that 64% of AI citations come from third-party sources, not the owned site.
FAQ Reddit and AI visibility in 2026
Why do LLMs ignore Reddit upvotes?
LLMs aren’t social platforms. They extract threads by semantic relevance to the user question, not by popularity score. The Semrush 2025 study proves 80% of cited posts have fewer than 20 upvotes, invalidating the assumption that community votes serve as algorithmic filters.
Which Reddit formats are most cited by ChatGPT and Perplexity?
Q&A threads represent over 50% of Reddit citations in AI responses. Comparisons (X vs Y) and open discussions complete 75%. Other formats (announcements, news, memes) total less than 25%. Prioritize explicit questions with reasoned answers.
How long does a Reddit thread remain citable by LLMs?
The median age of cited posts is 900 days, about 2.5 years. Reddit functions as long memory indexed by LLMs. A well-structured thread today could be cited in 2028, provided its content remains topically relevant.
Is French Reddit as effective as English Reddit for GEO?
No. French Reddit is roughly 5 to 10 times less dense than English Reddit. For a French brand exclusively targeting the French-speaking market, Reddit investment remains limited. For a brand also targeting English-speaking prospects, English Reddit is an essential earned media channel.
How many Reddit threads should I create or comment on for measurable impact?
5 to 10 monthly contributions across 2 or 3 targeted industry subreddits is a realistic investment for an SMB. First LLM citations appear at 30 to 90 days depending on model update rhythm. Regularity matters more than volume.
Should I create an official subreddit for my brand?
Yes for brands with an active community and expertise to share (B2B SaaS, tech brands, professional services). The official subreddit becomes a structured source LLMs can cite as brand authority. For a local SMB or a brand without preexisting community, the effort often exceeds short-term benefit.
How do I know if my Reddit actions actually improve my AI visibility?
Without an AI Share of Voice tool, you pilot blind. The baseline/post-action method on a panel of 30 to 100 strategic prompts allows correct attribution of a Reddit action’s impact on your citation by ChatGPT, Perplexity, Gemini and Claude. Cockpyt AI automates this measurement.
Sources
- Semrush, We Analyzed 248K Reddit Posts: What Drives Visibility in AI Search [Study], November 2025, https://www.semrush.com/blog/reddit-ai-search-visibility-study/
- Sistrix, Reddit’s Google Visibility Surge, 2024-2025.
- Stacker + Scrunch, Coverage Breadth Study: The Latest GEO Research on Expanding Brand Visibility Across LLMs, March 2026, https://stacker.com/blog/latest-research-on-expanding-brand-visibility-across-llms
- Ahrefs, Fresh Content and AI Search (analysis of 17 million AI citations), 2025, https://ahrefs.com/blog/fresh-content/
- Official Google Blog, Expanding partnerships with Reddit, 2024.
