“If you want to boost your AI search presence, just get on Reddit.”
You’ve heard this, or versions of it, I’m sure. It’s the AI search advice equivalent of telling someone to “just network more” to build their biz. Technically true, but facile as hell.
This lovely little nugget glosses over the meat that matters – which subreddit, which threads, which models – entirely. We see the consequences of that error in client data every week.
Via Scrunch, which tracks what’s actually surfacing in AI-generated answers, a more complicated story is revealed. In many cases, a single Reddit thread is doing more work to shape an AI answer than an entire brand website. If you don’t have the right data and understand that, then a Herculean effort to “do Reddit” can go nowhere.
Reddit is infrastructure
First let’s acknowledge the truth: Reddit shows up in AI answers at a rate that should make most brand teams uncomfortable. A June 2025 analysis of over 150,000 LLM citations found Reddit leads all domains at 40.1% citation frequency, with Wikipedia a distant second at 26.3%. It’s the leading cited source for both Google AI Overviews (2.2% of total citations) and Perplexity (6.6%).
As some have observed, there’s a volatility to Reddit citations.
ChatGPT cited Reddit in nearly 60% of prompt responses in early August 2025 before collapsing to around 10% by mid-September – a drop tied to a change in how Google serves search results, not a change in Reddit’s underlying relevance. Reddit’s overall citation share fell roughly 50% between October 2025 and January 2026. The platform remains structurally important, but access mechanisms shift.
The reason Reddit gets cited at all is straightforward: it contains exactly the content LLMs want to work with.
This means a lot of first-hand experience, direct comparisons and natural language that mirrors how real people ask questions. Reddit’s threaded Q&A structure – question asked, multiple answers provided, best answers surfaced – mirrors how LLMs like to present information. The format, in other words, is inherently citable.
Access determines impact
As with a lot of GEO, deals drive citations.
Reddit signed a $60 million licensing deal with Google for AI training. OpenAI pays to access Reddit’s Data API. At the same time, Reddit has moved aggressively against unauthorized scraping and has sued at least one major AI company over alleged unauthorized data use
The practical implication: not every model has the same access to Reddit, and access determines influence. Some models have direct licensed pipelines. Some rely on what gets indexed through search – which is exactly why the September 2025 citation drop hit ChatGPT hard but didn’t affect Perplexity or Google AI Mode the same way. Only 11% of domains are cited by both ChatGPT and Perplexity. The platforms are sourcing answers very differently.
So when someone tells you “Reddit matters for AI search,” the immediate follow-up question should be: for which model? Because the answer changes depending on how that model actually accesses Reddit content.
Reddit as Los Estados Unidos
Most brands see Reddit as a single channel.
It’s better viewed as a federation of independent communities, each with its own rules, tone, and level of authority on a given topic. Sort of like the USA or Mexico.
These “states” function less like sections of one site and more like separate publications. From a citation perspective they don’t perform equally.
Our client and company data shows niche subreddits consistently outperform broad ones. A focused community with consistent language and strong topical alignment is far more likely to be surfaced than a large, noisy general-interest sub.
Threads as the key units
Contrary to a lot of the obtuse advice online, LLMs don’t retrieve “Reddit” in some general sense. Instead, they pull from specific threads. The threads that show up repeatedly tend to share a few characteristics: a clear question that maps to real search prompts, strong engagement, and consistent language across replies. Easy for an LLM to process means easy for an LLM to reuse.
The implication is that you’re trying to get included in a small number of high-performing threads. Our data shows that the types of most valuable threads vary from company to company, so you need to get a close look at what works for you (not everyone).
Language is the leverage
Some brands track mentions and presence in Reddit. It’s a weak metric.
The more useful question is how your brand is described inside the threads that actually get reused – because that language carries through into AI-generated answers. If a product is consistently called durable, good for beginners, or overpriced across multiple threads, those descriptors show up in AI explanations. This is because an LLM is a pattern matching machine.
That’s why narrative control matters, and it’s where most brands aren’t paying attention.
Towards a strategy
We’re in the early days here, so we can give conventional wisdom a bit of a pass. However, data shows it to be deeply unsophisticated.
The default approach fails because there is no understanding of which threads are actually being surfaced, no thought over the primacy of language consistency, and no alignment with how specific models retrieve Reddit content.
Treating this as a data problem, rather than a social media one, changes what you’re actually doing.
Start by mapping which subreddits appear in AI outputs for your category. Analyze which threads get reused and assess why. Test prompts with tools like Scrunch to see which Reddit URLs consistently surface across queries. To the point you can, try and make sure the same descriptors appear across multiple threads. Track how citation patterns are shifting across specific models, because what worked a few months ago may not work now.
Yes, Reddit matters in AI search. But “just get on Reddit” isn’t a strategy. It’s what you say when you haven’t thought carefully about how the system works. The real variables are which subreddit, which thread, what language, and which model. If you’re not operating at that level, you’re wasting your time.