[Host] Welcome to the A.E.O. Engine AI Search Show, the A.E.O. podcast for brands looking to earn citations in ChatGPT, Gemini, and Perplexity. I'm your host, Vijay Jacob, Founder and CEO of A.E.O. Engine. Today we're getting into something that quietly became a marketing channel and then quietly changed shape on us. Marcus Reid is with me. Marcus, welcome to the show. [Guest] Hey everyone. Vijay, I've been waiting to talk about this one, mostly because nobody's numbers agree. [Host] Let's start where the listener actually lives. It's a Tuesday. A founder opens ChatGPT and types "best CRM for a ten-person startup." They get four names back. Two are competitors. One is a product they've never heard of. Their own product isn't on the list — even though they rank page one on Google and they've spent fourteen months on their blog. [Guest] And the part that stings is that those four names weren't picked by an editor. Somebody argued about them in a Reddit thread, and the model read the argument. [Host] So the founder digs in. The trail leads to a thread from a couple of years ago with eighty comments in it. [Guest] We shipped something like this at my last company. We had a hand-designed comparison page, looked great, and ChatGPT cited a two-year-old Reddit comment from a user named moist underscore towel — I'm not making that up — over our entire site. We put the comparison page in a doc and called it a learning. [Host] There's a name for the thing you just walked through, and it's the whole show today: Reddit threads feeding ChatGPT citations. A citation meaning the specific page the model retrieved and leaned on while it composed the answer. [Guest] That's the mechanical definition. The model pulls pages, uses them to write, then lists the ones it used. [Host] What surprised me is how much of the answer layer is Reddit. A Semrush analysis — a hundred and fifty thousand AI citations across five thousand keywords — found Reddit referenced in 40.1% of AI answers. Wikipedia at 26.3. YouTube at 23.5. [Guest] Forty percent is not a niche. And it isn't just ChatGPT. Same effect shows up in Perplexity, Google AI Overviews, and Google's AI Mode. [Host] Separately, ProFound put Reddit's share of all ChatGPT citations around 2 to 3%. Different denominator, different number. And ProFound flagged that their own earlier filter was stripping YouTube and LinkedIn URLs as false positives, which biased results toward text-first sites like Reddit. So hold the exact figure loosely. [Guest] Which is my favorite kind of statistic. The number is fuzzy and the direction is screaming. [Host] Then this playbook started circulating among founders. The argument inside it: stop treating your own site as the only surface. Reddit Q and A is an answer engine surface in its own right. A.E.O. work, not S.E.O. work. [Guest] I'd push on that framing by about ten degrees. It's not either-or. It's that the thread is a page you can't fully control, and most teams were never staffed to think about pages they don't own. [Host] Fair. [Guest] The thread is the new landing page. You just don't get the admin login. [Host] So walk me through the plumbing. Why is Reddit so over-represented in retrieval? [Guest] Dedicated API integrations. Reddit content arrives on a separate feed sitting on top of whatever the normal web search already returned. So Reddit isn't winning a ranking fight — it's getting a lane of its own. [Host] That explains volume. It doesn't explain which threads. That part is much more specific. An independent analysis across more than four hundred brand queries found the median cited Reddit thread is nine hundred days old. Two and a half years. [Guest] That's the number I'd frame. Not because old is magic, but because citations cluster. Same analysis: in each vertical, roughly a hundred and fifty to three hundred threads do most of the citation work. The same established threads get reused no matter what gets published this week. [Host] So a brand publishing fresh posts isn't competing with a fresh thread. It's competing with a thread that already won. [Guest] It's Groundhog Day. Same small cast, every morning, forever. [Host] The shape matters too. Threads that get cited tend to begin like a prompt, not a post — questions, because that's what users feed the model. [Guest] And substance inside. Statistics, specific data points, detailed answers. A thread full of "I used it, it was fine" is not a knowledge asset. Vague one-liners don't survive retrieval. [Host] There's also a subreddit hierarchy. High-weighting subs reported include r slash S.A.A.S., r slash S.E.O., r slash ChatGPT, r slash Entrepreneur, r slash ArtificialIntelligence, r slash AskEngineers, r slash Finance, r slash Health, and r slash Productivity. Established, moderated, high-traffic. New or tiny subs get deprioritized. [Guest] Which is a deeply boring conclusion dressed up as a conspiracy. Big moderated subs get cited. Same reason journalists quote the Times and not your group chat. [Host] Last piece is associative. When real users describe a product in detail across multiple threads — the use case, the user type, the outcome — the model learns those associations. That feeds citation likelihood later. [Guest] That's the part people skip. It's not one thread. It's a pattern across threads that reads like consensus. [Host] Now the twist, which is where this gets interesting. Around August 13th and 14th, tracking tools — Promptwatch, Trackerly — reported Reddit's share of ChatGPT Search citations collapsing. Roughly 60% down to about 10% over six weeks. Something like an 86% drop in citation share. [Guest] And the internet did what the internet does. r slash aiwars called it a shift worth monitoring. r slash ChatGPT debated what it means for users. r slash technology got nostalgic about the old Reddit — a commenter said ten years ago it had thousands of experts in well-moderated subs. [Host] The most useful finding came out of r slash ChatGPTPro. The point there: ChatGPT didn't stop reading Reddit, it stopped showing it. Still pulling Reddit threads while researching, about one in four pages it consults, same as before. It just stopped listing them in visible citations. [Guest] Which points at a presentation-layer decision, not a sourcing change. And the poster flagged that it was a single-account snapshot — which I appreciated, because that's the level of hedging this deserves. [Host] Here's where I'll push back on the relief. If it's presentation-only, the reading still happens, so the influence still happens. Brands chasing visible citations might be polishing the trophy case while the actual recommendation gets shaped by threads nobody sees. [Guest] I read it the other way, and we can disagree here. If the citation disappears, the feedback loop dies. You can't measure what you can't see, and teams defund Reddit work the moment the dashboard goes quiet. The influence may persist, but the budget won't. My read is that the visibility change is the operational event, even if model behavior didn't move an inch. [Host] That's a fair split. And I could be wrong on this — I genuinely don't know if the reading behavior holds six months from now, or whether this was tested and rolled back. [Guest] I don't know either. There's no official statement from OpenAI or Reddit, so everything about the cause is community inference. [Host] The measurement side has its own problem. One A.E.O. practitioner described tracking three hundred-plus custom prompts for a B2B client, generating thousands of responses. Just two Reddit threads accounted for the vast majority of citations. [Guest] Two. So a headline like "Reddit is 40% of citations" can actually mean "twelve threads are 40% of citations." Those are very different sentences to build a strategy on. [Host] It has a little Big Short energy to it. Everyone trading on the same small bundle of instruments, then reporting it as the whole market. [Guest] Here's where I'd pull it back to what a team actually does on Monday. At A.E.O. Engine we look at whether the brand gets named inside the answer, not just whether a URL shows up. The thread-level view beats the domain-level view — which threads carry the question, who's in them, what they say about you. [Host] And the practical version of that: the assets that win here look like answers. Real use cases, specific numbers, a product described by people who actually used it, in subreddits that are moderated and old enough to matter. That's the same output an A.E.O. content system tries to produce on owned pages — except here the surface belongs to someone else. [Guest] The uncomfortable part is you can't publish your way into it. You have to participate, and you have to be worth quoting. [Host] Which is why I keep telling founders to audit the threads before they audit their pages. If you want the answer engine saying good things about you, you need to know who's already answering the question. That's part of the visibility work we run at A.E.O. Engine — AI search and LLM visibility audits — and the Reddit layer is one of the first places we look. [Guest] And no, I'm not going to pretend the data is settled. It isn't. The direction, though, is not ambiguous. [Host] So, the recap. Reddit threads are a major source layer for AI answers. The citations cluster in a small pool of old, question-shaped, statistics-heavy threads. And the August drop may have changed what's visible rather than what's read. Track both. Head to A.E.O. Engine dot A.I. — that's A.E.O. Engine dot A.I. — and see how your brand currently shows up in AI answers. Marcus, thanks for this. [Guest] Anytime. Go read a thread.