“I Asked The AI About My Own Company And It Made Stuff Up”
A lot of marketers have tried this lately. You type your brand into ChatGPT, ask “so what do you say about us,” and wait to see your own reflection. Half curiosity, half dread.
And the answer comes back a little off — some stale detail, some forum gripe repeated as fact.
But the thing that actually bugs people isn’t that the info is old. I keep hearing the same complaint: “our site has the right version, and the AI just… didn’t look there.” The mismatch is in where it looked.
So is that real? Does the AI really skip your clean corporate page and grab some anonymous post instead? Let’s look at the numbers, because somebody finally measured this at scale.
AI Pulls From A Tiny Handful Of Sites
Here’s the headline: the sources AI leans on are far more concentrated than you’d guess. A 2026 Contently roundup of several large studies makes the lopsidedness pretty clear.
The names at the top are always the same three.
- Wikipedia — 7.8% of all ChatGPT citations, and reportedly about 22% of ChatGPT’s training data.
- Reddit — up to 1 in 5 of Perplexity’s citations (~20%), and the single most-cited source across major engines in Peec’s analysis.
- LinkedIn — cited in 14.3% of ChatGPT Search answers, and a fast climber: 11th in November 2025, 5th by February 2026 in Profound’s tracking.
Notice what they share. Wikipedia is an encyclopedia, Reddit is an anonymous forum, LinkedIn is a business network — all three are places where other people write, not pages you control.
I’d assumed the AI would weight official announcements more heavily. It doesn’t. It reaches for the spots where third parties are talking.
”Sure, But Is The Data Any Good?”
Fair question. When you see stats like these, the reflex is “some tool measured a tiny sample and called it a trend.” But the underlying studies are big.
- Peec AI analyzed about 30 million sources.
- SEMrush looked at 325,000 prompts.
- Profound tracked 1.4 million citations.
- SE Ranking aggregated 129,000 domains.
That’s hundreds of thousands to tens of millions of data points, and the same pattern — Wikipedia, Reddit, LinkedIn on top — keeps falling out.
So this isn’t a fluke from one small sample.
Getting Talked About Is What Earns The Citation
Here’s the part that matters most if you do marketing. SE Ranking found a striking correlation: domains with lots of Reddit brand mentions averaged 7 ChatGPT citations, while domains with few mentions averaged 1.8. That’s about a 3.9x gap.
So whether the AI cites you tracks less with how slick your own pages are, and more with how much you’re being talked about out there. The mentions pile up in communities and on social, and that pile is what makes you citation-worthy.
Think of the diner down the street. You can paint “best lunch in town” on your own sign all day — but the place new customers actually walk into is the one the regulars keep recommending. AI citations move the same way.
The AI’s Take On Your Brand Is Already Skewed
And it’s not just where it looks. How the AI sizes up a brand comes with its own built-in bias. A study presented at EMNLP 2024 found that several AI models recommend luxury brands to high-income countries 88-100% of the time — a clear thumb on the scale for big global names.
The AI isn’t a neutral referee. It’s a mirror for whatever lopsidedness sits in its training data.
There’s a weirder wrinkle too. SparkToro ran 2,961 trials and found the AI’s brand-recommendation list comes back different more than 99% of the time — but a given brand’s appearance rate across all those runs is statistically stable. So any single answer is noise. The overall pattern of what it cites and which brands it reaches for is pretty steady.
Which tells you something practical. If you want to know how you look inside the AI, don’t agonize over one reply. Watch the trend over time — what it cites, how it talks.
The Real Work Is Off Your Own Site
Pull it together and the lesson is blunt: in the AI era, polishing your own website isn’t enough on its own. The sources AI trusts are the third-party places — Wikipedia, Reddit, LinkedIn — and the mentions stacking up there are what feed the citations.
So the first move for a marketer is to stop chasing your own search ranking as the only scoreboard. The thing worth watching is where the AI pulls you from, and the tone it uses when it does.
You’ve probably heard “GEO” by now. GEO is Generative Engine Optimization — basically, the work of getting cited and recommended by AI. And step one isn’t your ranking; it’s knowing your sources and your story.
Look at whether your brand is actually building up on Wikipedia and Reddit, and whether the picture those places paint matches reality. Keep updating your own site, sure — but keep an eye on the reputation growing in the places you don’t own. AI-era brand management is mostly a shift in where you point your attention.
Sources
- [R1] Contently (2026), “Top Sources LLMs Cite Most in 2026”. Contently
- [R2] Kamruzzaman et al. (2024), “Global is Good, Local is Bad?: Understanding Brand Bias in LLMs”, EMNLP 2024. ACL Anthology
- [R3] Rand Fishkin / SparkToro (2026), “AIs are highly inconsistent when recommending brands or products”. SparkToro
