Subscribe
Share
Share
Embed
A 129K-domain study says backlinks, domain trust, Reddit mentions, structure and freshness beat llms.txt for ChatGPT citations.
AEO Engine's episode "What Actually Gets ChatGPT to Cite You (It Isn't llms.txt)" analyzes a 129,000-domain study showing that backlinks, domain trust, and Reddit mentions — not llms.txt files — drive ChatGPT citations and Google AI Overviews visibility in 2026.
Key takeaways:
Q: Does adding an llms.txt file improve ChatGPT citations?
A: According to a 129,000-domain study, no — llms.txt showed no measurable impact on whether ChatGPT cites a site. Backlinks, domain trust, and Reddit mentions were far stronger signals.
Q: What factors actually get a website cited by ChatGPT and Perplexity?
A: The study points to backlink authority, domain trust, Reddit mentions, well-structured content, and content freshness as the top ranking drivers for AI citations.
Q: Is Reddit important for AI search visibility in 2026?
A: Yes. Reddit mentions correlated with higher citation rates across ChatGPT and Perplexity answers, making Reddit a measurable channel for AI visibility strategies.
Since ChatGPT reached mainstream adoption, brands have chased technical shortcuts like llms.txt, hoping a single file would unlock AI citations. The 129K-domain study covered here, originally shared on x.com, closes that debate: AI answer engines like ChatGPT, Perplexity, and Google AI Overviews reward the same trust infrastructure that powered classic SEO — backlinks, domain authority, structured markup, and fresh content — plus newer signals like Reddit community mentions. That creates a concrete commercial opportunity for marketing teams and SaaS founders who need measurable AI visibility but lack a repeatable system. AEO Engine helps those buyers turn AI search ranking factors into an accountable channel, from AI-driven competitor analysis to answer engine optimization for local businesses and B2B brands competing in conversational AI results in 2026.
Get cited by ChatGPT, Perplexity, and Google AI Overviews — start with https://aeoengine.ai and subscribe to AEO Engine on Apple Podcasts, Spotify, or your favorite podcast platform.
Answer Engine Optimization (AEO) is how your brand gets cited, recommended, and surfaced inside ChatGPT, Perplexity, Google AI Overviews, and Claude. This is the daily podcast for marketers, founders, and SEOs who want their brand to be the answer AI engines give.
Each episode breaks down a new AEO tactic, a real algorithm change, or a brand that just won (or lost) visibility inside AI search. Topics include: how ChatGPT decides which brands to recommend, how Perplexity chooses its sources, how Google AI Overviews differ from traditional SERPs, how to structure content for LLM citation, schema strategies for answer engines, and the emerging field of Generative Engine Optimization (GEO).
Brought to you by AEO Engine — the platform brands use to monitor, measure, and grow their AI search visibility. Whether you're a B2B marketer, DTC founder, or in-house SEO, this podcast turns the daily chaos of AI search into a concrete playbook you can execute on.
New episode every morning. Transcripts on every episode. Subscribe to stay ahead of how AI engines rank and recommend brands.
[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 digging into a study that's been moving through the S.E.O. and A.E.O. crowd all week — what actually gets ChatGPT to cite you, and why llms dot txt may not be the lever everyone shipped it as. I'm joined by Marcus Reid. Marcus, welcome.
[Guest] Hey everyone. Good to be here. I've read the same threads you have, and I have opinions, which I assume is why I'm here.
[Host] Let's start with a scene most founders know. You open ChatGPT, you ask it to recommend a brand in your category. It names three. Two are competitors. And the source it links to is a Reddit thread from 2022 where a stranger wrote, "I tried this one, it was fine."
[Guest] And the founder screenshots it, sends it to the team chat at eleven at night, and everyone goes quiet. Because nobody can point at the button that fixes it.
[Host] That's the feeling. The old answer was rank higher. That answer doesn't map anymore.
[Guest] Right. And there's a name for it now — answer engine optimization. The A.E.O. discipline. The goal isn't ranking. It's being the source the model pulls from while it builds the answer.
[Host] So let's get into what actually happened here. What is this study?
[Guest] A breakdown of a large analysis — 129,000 domains, a set of pages, a set of prompts — circulating as evidence that the classic signals still decide AI search. Backlinks, domain trust, mentions on Reddit and Quora, page structure, freshness. All of those move citation odds more than llms dot txt does.
[Host] So the file teams rushed to ship last year isn't the lever.
[Guest] Not according to this. It's not that it hurts. It's that it's not doing the work. Meanwhile, domains that show up in Reddit or Quora discussions had roughly four times higher odds of getting cited. Four X is not a rounding error.
[Host] And Reddit has its own data. Profound looked at ChatGPT citations and found Reddit is around 2 to 3 percent of all of them. But about 99 percent of those point to individual threads — not subreddit pages, not user profiles.
[Guest] That's the detail I'd underline twice. The model isn't citing a community. It's citing one conversation where a real person answered a specific question. That is a very different target than "go get mentioned on Reddit."
[Host] Then there's the trust layer — G2, Capterra, Trustpilot. Presence there appears to roughly triple citation odds.
[Guest] Which is a very boring sentence and a very expensive one if you've spent three years ignoring it.
[Host] There's a Google relationship too. AirOps research found pages ranking number one in Google get cited about 3.5 times more often. But rank isn't a guarantee. Their headline number was that 85 percent of sources never make the cut.
[Guest] And the inverse is true — zero-backlink pages can still get cited. The research notes a 2011 blog post can get pulled in before Google ever surfaces it to a human reader. That should bother anyone who thinks backlinks are the entire story.
[Host] So how does the mechanism actually work? Why would a Reddit thread beat a polished landing page?
[Guest] Two modes. Parametric memory — the model answers from training data, no retrieval, no citations. And web search mode — live retrieval, citations attached to what it actually used. If your content only lives in the first mode, you cannot be cited. Not badly. Not rarely. Cannot.
[Host] And the line that reframed it for me: ChatGPT doesn't know your site. It doesn't browse continuously. It cites what it extracted for that specific answer.
[Guest] Yeah. It's not a relationship, it's a transaction, per query. Which means — and this is my read, not the study's — the phrase AI visibility is hiding a lot. There isn't one visibility. There's a per-prompt result. Like a contractor pulling one tool out of the truck for one job.
[Host] Then selection itself. Citations attach to sources that contributed substantive, verifiable information. Generic content rarely gets cited even when it ranks well.
[Guest] Which explains the Reddit thing completely. A thread answers one question with specifics. A category page says "we're passionate about quality." One of those is extractable and one of those is wallpaper.
[Host] Structure and freshness fall out of the same logic. If the model has to parse it fast, the page has to be legible to the parser.
[Guest] Legible to the machine, not just pretty to the human. Different skill, and most content teams only staff for one of them.
[Host] So why did this go viral? It's a study, not a launch.
[Guest] Because it's evidence for something people already wanted to be true — that classic authority still wins. Half of S.E.O. Twitter has been waiting for the llms dot txt crowd to be politely told their magic file doesn't matter. This is the receipt, and receipts travel.
[Host] I'd push back a little. The study argues authority moves citation odds. It doesn't say relevance doesn't. There's a discussion on amicited dot com making the reverse case — that ChatGPT cites the most relevant source, not necessarily the most authoritative domain.
[Guest] I think that's a distinction without much difference in practice. Relevance and authority correlate, because authoritative sites tend to be the ones that actually answer the question. But fine — if you're a small site with one genuinely great answer to one specific question, you're not locked out. I'll give you that one.
[Host] Noted. That's the disagreement for the record.
[Guest] Put it in the newsletter.
[Host] The other thing that jumped out is model fragmentation. Analyze AI tracked 83,670 citations across ChatGPT, Claude, and Perplexity between November 2025 and January 2026. ChatGPT draws 12.1 percent of its citations from Wikipedia. Claude: 0.1 percent. Perplexity: zero.
[Guest] That's a fun stat until you realize what it implies. Build a strategy around one model's citation pattern and you're optimizing for that model's audience and nobody else's. Perplexity's top citations run roughly 46.7 percent Reddit and about 14 percent YouTube. Different animal entirely.
[Host] So "get cited by AI" isn't one project. It's three projects sharing a content foundation.
[Guest] And I genuinely don't know if this holds in six months. These are twelve-week windows measured on systems that ship weekly. I could be wrong about how stable any of it is.
[Host] Fair. The other thing worth flagging is fabricated citations. Asking for recent sources, asking about a niche topic, and above all asking it to cite sources with browsing turned off. That last one is the most common trigger.
[Guest] That one's almost a party trick. You ask a model with no web access to cite its sources and it will invent something with total confidence — which, to be fair, is what a nervous intern would do too. The difference is the intern eventually admits it.
[Host] And the community pushback on astroturfing. The discussion was clear that genuine participation in relevant subreddits works, and shilling gets you the opposite of what you wanted.
[Guest] Right, and that's the fork in the road nobody wants to talk about on stage. The playbook that works is the one you can't fully automate, because the trust is the point.
[Host] So if you're running a brand — a Shopify store, a B2B team — what do you actually do Monday morning?
[Guest] Three things. Get into the trust surfaces — G2, Capterra, Trustpilot — with real, specific reviews. Be present in the communities where your buyers already argue about your category, honestly and in your own name. And restructure pages so a model can extract a clean answer in one pass instead of guessing at your meaning.
[Host] Which is the layer we build at A.E.O. Engine — not a magic file, the boring substrate underneath it. Topical mapping so the site covers the whole question space, structured content so answers are extractable, authority signals so the model has a reason to trust the page, and a publishing cadence so freshness isn't a quarterly scramble. The content agents handle the volume. The strategy decides what deserves to exist.
[Guest] Volume without a point of view just makes more pages nobody extracts from.
[Host] Agreed. That's why the hundred-day sprint starts with mapping, not writing. And honestly, the Reddit finding is the one I keep thinking about — because most brands cannot buy their way into a thread. They have to earn a mention from a person who has no incentive to give one.
[Guest] Which is the least scalable and most durable thing on the list. Annoying, but that's the pattern.
[Host] Quick recap, topic only: citations come from retrieval, not from existing. Backlinks, domain trust, Reddit and Quora mentions, structure, and freshness appear to do more work than llms dot txt. And the model mix matters — ChatGPT, Claude, and Perplexity are three different games with three different source diets.
[Guest] And the boring version of that advice is the version that actually works.
[Host] Head to A.E.O. Engine dot A.I. — that's aeoengine.ai — to see how we run this for brands chasing citations instead of rankings. See you next episode.