Where to Publish So AI Search Cites You: What a One-Week Experiment Found
If you want AI search to cite something, where you publish it decides more than how much you publish. A paper from Fudan University posted on October 8 found that a single article on a site an AI engine already leans on can show up in that engine's answers within days, and that 40 posts on a self-hosted WordPress site got zero references over a week (Search Engine Journal).
The researchers proved it by making something up. They invented a concept, wrote about it on sites like Medium and YouTube, and watched eight of the ten AI search tools they tracked start citing it inside seven days (arXiv). That's useful for anyone trying to get their brand into AI answers, and a little alarming for anyone who'd rather a stranger not be the one doing it.
The short version
Each AI engine has its own short list of favorite sites, and posting on those sites gets you cited far faster than posting on your own blog.
In the test, one article on a site the engine preferred did more than 20 matched posts on sites it didn't.
The same route is open to anybody who wants to write about your brand, including people selling $14 posting packages.
ChatGPT has been moving toward primary sources this year, so the pages only you can write still carry weight.
What the Fudan researchers tested
The paper is called "From Public Posts to AI-Search Citations: Measuring the Fragility of AI Search," and the tests ran in March and April of this year. The team first ran 120 questions through ten AI search tools (four English ones, ChatGPT, Perplexity, Grok and Google's AI, plus six Chinese ones like Doubao and DeepSeek) and logged 17,211 citations across 6,356 websites. The top 20 sites took between 20.5% and 70.8% of each tool's citations, which tells you how concentrated this is (arXiv).
Each engine had its own favorite. Reddit led for Google's AI and Grok, YouTube led for Perplexity, and Wikipedia led for ChatGPT, and six to ten of every tool's top ten sites were places where an ordinary person can post (Search Engine Journal).
So they made accounts. They invented a "Cactus Gravity Protocol," a fake keyboard and an opinion question about AI replacing junior programmers, and they planted made-up names and numbers in each post so they could tell when an engine had picked it up. They checked first that no engine knew anything about these topics, then asked the same five questions every day for a week.
A made-up idea, cited within a week
Researchers invented a concept, posted about it on sites each AI engine already cites, then asked the same questions daily for seven days. English-language engines shown.
Source: Liu et al., Fudan University, "From Public Posts to AI-Search Citations" (arXiv, Oct. 2026), as reported by Search Engine Journal. Link counts run from each engine's first citation through day seven. Tests ran March to April 2026 on fictional topics, not real brands.
The English-language results varied a lot by engine. Grok linked to the posts in 30 of 32 answers once it started citing them, Google's AI in 15 of 33, Perplexity in 13 of 35 and ChatGPT in 7 of 20. Grok also repeated a planted detail in 53% of its answers, ChatGPT in 10%, Google's AI in 6% and Perplexity in none (Search Engine Journal). Perplexity linked to the posts without repeating the invented facts, which I'd read as it citing the page for context and pulling its claims from somewhere else. I could be wrong on that, it's a guess from the numbers.
Where to publish so AI search cites you
The placement test is where I'd start. The researchers paired posts about five little-known vacation spots, one on a site the engines cite a lot and others on sites they cite less. The frequently cited site produced the first citation every time. On the less-cited sites the best case took 20 posts to get there, and the 40 posts on a self-hosted WordPress site got nothing in seven days (Search Engine Journal).
If you've spent the last year publishing on your own blog and wondering why the AI engines don't seem to notice, that's a big part of the answer. A brand-new page on your site has to be found, crawled, trusted and then picked over sites the engine already likes. A post on the engine's favorite site skips most of those steps.
This lines up with what I saw in the AI Verdict Study. Across 32,089 citations from ChatGPT, Claude, Gemini and Perplexity, about four in five of the citations I could classify pointed to sites other than the brand being asked about, and on the first unbranded question a brand's own site supplied 4% of them (Where AI engines go to decide about your brand). The engines form most of their view of you from other people's pages. The Fudan paper adds a time element to that, which is that a well-placed page can get into the mix within days.
The favorite sites move around, so I wouldn't treat any list as permanent. Ahrefs found Reddit showed up in 19.1% of ChatGPT's U.S. answers in March and 2.3% in September, and Wikipedia fell from 14.1% to 2.9% over the same stretch (Search Engine Journal). I went through what that means for Reddit specifically in Is Reddit worth it for AI visibility, and the short answer is that it depends heavily on your category and the engine.
Anyone can use the same route to talk about your brand
The researchers bought a $14 posting package from a GEO vendor. It produced 13 public posts in about 28 minutes, and one AI search tool (Doubao) cited the content with their planted details within an hour (arXiv). The authors say plainly that this was one confirmed case and not a success rate.
Plenty of marketers will see an opportunity in that, and there is one. I'd also think about what it means when somebody else does the posting. If posting on Medium or YouTube or a busy forum can change what an engine says about a made-up protocol in a week, an unhappy customer or a competitor's agency can do roughly the same thing to what it says about you. You won't hear about it unless somebody on your team is asking the engines about your brand on a regular schedule.
Two caveats matter here. None of the test topics were real brands, and none of the English tools cited the fake keyboard posts within the week, so product and brand questions may be harder to move than a made-up concept with no competition (Search Engine Journal). The tests also came before Google said on May 15 that its spam policies cover attempts to manipulate "generative AI responses in Google Search" (Search Engine Roundtable). Google's own AI optimization guidance says chasing inauthentic mentions "isn't as helpful as it might seem." I'd take them at their word on that, and I'd stay far away from anybody selling posting packages.
ChatGPT has been leaning toward primary sources
Ahrefs published something this month that points a slightly different way. They looked at ChatGPT's 200 most-cited sites in U.S. answers from March to September and found government health sites, research sites, regulators and hands-on testers like RTINGS climbing, while explainer sites fell. PubMed went from 64th to 4th. WebMD went from 18th to 79th (Search Engine Journal, Ahrefs).
ChatGPT's sources moved this year
Ahrefs tracked the 200 sites ChatGPT cites most in U.S. answers. Community and encyclopedia sites fell while primary sources climbed.
Source: Ahrefs, via Search Engine Journal (Oct. 12, 2026). Based on 2.4 million U.S. answers in March and 3.8 million in September. Citation ranks don't measure traffic.
So the open platforms are an easy way in, and at least one engine has been giving more weight to the original source of a fact. I don't think those two findings fight each other. A post on a busy platform gets noticed fast, and a page that's the primary source of something (your pricing, your product specs, your own research, your help center) has a better shot at staying cited after the summaries come and go. In How long do ChatGPT citations last I went through a Green Flag Digital study of finance pages where brand-owned pages grew from 1.8% to 22.4% of ChatGPT's top citations over a year, while Wikipedia and Reddit dropped to zero in that set.
What I'd do with this
Number one, find out which sites each engine reads for your category. Ask the questions your buyers ask in ChatGPT, Perplexity, Gemini and Google's AI Mode, and write down every site they cite. Bing's new Topics Portfolio report in Webmaster Tools also groups the queries where Microsoft's AI answers cite your site and lists queries where competitors get cited and you don't (Bing Webmaster Blog). Microsoft calls the numbers approximate, so use it for direction.
Two, put your best work where those engines already look. That might be a YouTube explainer, a LinkedIn article from your subject expert, a guest piece in a trade publication, or a real answer in a community your buyers use. One strong piece in the right place did more in this test than twenty in the wrong one, so I'd spend the effort there before cranking out more blog posts.
Three, keep publishing the things only you can publish. Clear pricing, specs, honest comparison pages, help center articles that answer the tradeoff questions, and original data. In my study, a brand's own site went from 4% of citations on the first question to 35% when the buyer asked about tradeoffs, and help center pages were nearly half of that.
Then last, watch what people say about you on the open platforms, and ask the engines about your brand on a schedule. Run the same questions more than once, since two identical runs in my study shared only 38% of their cited websites. If something wrong starts showing up, you want to find the post it came from while it's still one post.
If you want a quick read on where you stand, the self-check below takes about two minutes. For a fuller look, the free Quick Check runs nine checks and three buyer prompts to show you which layer is holding you back, and the $199 Audit Kit covers all 18 criteria and builds a 13-week plan if you'd rather run the whole thing yourself. The bigger picture of why AI visibility is the number to watch is in The Click That Didn't Happen.
Could AI find the right story about you in a week?
Six questions. Answer for your brand and get a score with the fixes in priority order.
0 of 6 answered
I'd love to hear what you find when you look at which sites the engines cite in your category, especially if it doesn't match what this paper saw. Send me a note or find me on LinkedIn. A lot of how engines pick their sources, and why some brands get explained well and others don't, is what I wrote Explainable about.
Jarred Smith is the author of Explainable: Why AI Recommends Some Brands & Ignores Others, an Amazon bestseller on AEO, GEO, and SEO. He's a marketing leader with nearly 20 years of experience across healthcare, public media, retail, and environmental services. Find him at jarredsmith.com.