- 0Reddit citations in 75 ChatGPT answers25 AI visibility tool questions, US, 11 Nov 2026
- 10 of 440verified crawler requests were for llms.txt filesNew site, first three days, 29 Oct to 1 Nov 2026
- 0 of 9open-source GEO audit tools publish a test of their scoreChecked 10 Nov 2026
- 4 of 10Google US top 10 “best AI visibility tools” lists include their own productRecorded 5 Nov 2026
Short answerWe checked five common AI visibility tips against our published data: ChatGPT cited Reddit 0 times in 75 answers to AI visibility tool questions on 11 November 2026, only 10 of 440 verified crawler requests in a new site’s first three days were for llms.txt files, and none of nine open-source GEO audit tools publishes a test of its score. In this 4-minute episode, Jake and Dylan go through each tip, including the ones still being tested.
Listen on Apple Podcasts, Overcast or RSS.com, or add the RSS feed to any podcast app.
Chapters
- 0:00 Five common tips, and the ground rule
- 0:33 Tip 1: add an llms.txt file
- 1:16 Tip 2: ping IndexNow for fast Bing pickup
- 1:46 Tip 3: ChatGPT loves Reddit
- 2:35 Tip 4: trust your GEO audit score
- 3:19 Tip 5: pick from a best tools list
- 3:40 What to do instead
Read the full data: Testing AI search tactics: what works and what doesn’t. Do AI crawlers read llms.txt? How to rank in ChatGPT: what 388 AI answers cited. Do GEO audit scores predict AI visibility? Who writes the “best AI visibility tools” lists? The IndexNow test on the Experiments page.
Transcript
Jake: Okay, there is so much AI visibility advice out there right now. How much of it actually holds up?
Dylan: Less than you’d think. We checked five of the most common tips against our own data.
Jake: This is Measured, Not Guessed, from the AI Visibility Census. I’m Jake.
Dylan: And I’m Dylan.
Jake: So Dylan, ground rule before we start. When you say a tip doesn’t hold up, what does that mean?
Dylan: It means our data doesn’t back it, or nobody has tested it. It doesn’t mean it can never work. Most of this is one site, or one category, on specific dates.
Jake: Fair. Tip one. Add an llms.txt file, and AI crawlers will use it to find your best pages.
Dylan: We published one on day one of a brand-new site and logged every request for three days. Out of 440 verified crawler requests, only 10 were for llms files. About two percent.
Jake: Who asked for it?
Dylan: Googlebot five times, GoogleOther four times, ClaudeBot once. In those three days, no verified OpenAI or Perplexity crawler asked at all.
Jake: So some bots do download it.
Dylan: Some do. But a download isn’t use. Nothing in the logs shows an AI answer drew on it. We grade it untested.
Jake: Okay. Tip two. Ping IndexNow, and Bing, and so Copilot, will find your page fast.
Dylan: In our first three days, our llms.txt file went out in dozens of IndexNow submissions. Every one got a success response. Bingbot never requested it.
Jake: Never?
Dylan: Never. And on our timing test, Bingbot made no verified visit to the new article at all. We’re running a cleaner head-to-head IndexNow test, with a result due December fifth. So that one is still open.
Jake: Tip three, and this one is everywhere. ChatGPT loves Reddit, so go post on Reddit.
Dylan: On November eleventh we asked ChatGPT 25 buyer questions about AI visibility tools, three times each. That’s 75 answers. It cited Reddit zero times.
Jake: Zero?
Dylan: Zero. Gemini cited Reddit 42 times on the same questions. And in a second check on November thirteenth, ChatGPT cited Reddit in 0 of 300 answers.
Jake: So the advice is just wrong?
Dylan: Not necessarily. An Ahrefs study in March found Reddit was ChatGPT’s most-cited domain. Our questions were about software, and the gap may come down to the topic. The point is, check your own questions before you spend months on Reddit.
Jake: Got it. Tip four. Run a GEO audit, and the score tells you how visible you are in AI.
Dylan: We looked at nine open-source GEO audit tools on November tenth. None of the nine publishes a test of its score against what AI engines actually answer.
Jake: None of them?
Dylan: Zero of nine. One report template even says a 10-point score improvement typically correlates with 15 to 25 percent more AI citations. No source.
Jake: Ouch.
Dylan: And one audit gave example.com, a placeholder page, 94 out of 100. Which is fair, the page is readable. But readable isn’t visible. We’ve registered our own test, and we predict no relationship.
Jake: Last one. Just pick a tool from the “best AI visibility tools” lists.
Dylan: We looked at Google’s US top ten for “ai visibility tools” on November fifth. Four of the ten list their own product. Three put it first.
Jake: That’s a lot.
Dylan: And there’s no consensus. Eight different tools come first across the ten pages.
Jake: So what does someone actually do with all this?
Dylan: Three things. One, ask the questions your buyers ask, and look at what each AI engine actually cites. Two, treat files like llms.txt as cheap experiments, not a strategy. Three, before you trust a tip, or a list, ask who tested it, and who is selling something.
Jake: Including us?
Dylan: Including us. That’s why every number we use is published with its method.
Jake: All five studies are linked in the show notes. Next time: we counted where AI engines get their sources, and news sites made up 3.3 percent of the citations.
Dylan: That’s 27 of 807.
Jake: This was Measured, Not Guessed.
Dylan: AI visibility, measured, not guessed.
Next: one experiment or finding per episode. All episodes, or get new findings by email.
ChatGPT cited Reddit 0 times in 75 answers to 25 AI visibility tool questions asked from the US on 11 November 2026, while Gemini cited it 42 times on the same questions (AI Visibility Census).
Source: AI Visibility Census, “Podcast episode 3: Five popular AI visibility tips, checked: ChatGPT cited Reddit 0 times in 75 answers”, https://aivisibilitycensus.com/podcast/worst-ai-visibility-advice/ (CC BY 4.0)How we made this episode, and limits
Voices: the hosts, Jake and Dylan, are AI-generated voices made with ElevenLabs. They are show hosts, not real people. The research, the data and the script are by the AI Visibility Census team.
Checks and AI use: the script was drafted with AI assistance from the published study and every figure was checked against that page before recording. After recording, we ran the audio through speech recognition and confirmed every number is read as written. The transcript is our script, timed to the audio.
Limits: “doesn’t hold up” means our data does not support the tip or nobody has tested it, not that it can never work. The llms.txt and IndexNow findings come from one new site; the Reddit figures cover software buyer questions on one day, and an Ahrefs study from March 2026 found Reddit was ChatGPT’s most-cited domain; the audit review covered nine open-source tools found on GitHub; the lists study covered one search term in Google US on one day. The IndexNow head-to-head test reports on 5 December 2026. Each linked page has its full limits.
Spotted an error? Email hello@aivisibilitycensus.com and we will correct it.