What exactly did the study measure?

Whether AI platforms cite the same sources when asked the same question. I submitted 15 standardized queries to Claude, ChatGPT, and Gemini through three interfaces each — mobile app, web browser, and API — for 135 total runs and over 750 logged citations. Every frequently cited domain was then cross-referenced against its full SEO profile via Ahrefs' API.

What was the single biggest finding?

The platforms almost never agree. Mean source overlap across platforms was below 5% per query. For one CRM question, ChatGPT and Claude shared exactly zero domains. Three AIs answering the same question are drawing from nearly disjoint pools of sources.

What is the "Two-Tier Citation System"?

The discovery that the same platform behaves differently depending on how you access it. ChatGPT's web interface never cited a domain below Domain Rating 50 — a de facto DR 85+ quality gate. Its API cited a DR 4.5 site with roughly 12 monthly visitors six times in a single answer. The consumer product has quality controls that the API — which powers most AI tools on the market — does not share.

Does this mean SEO is dead?

The opposite. SEO authority is quietly doing GEO's heavy lifting. Every domain reliably cited across all platforms had DR 85+, massive referring-domain counts, and dominant branded anchor text. But DR predicts the floor, not the ceiling: high authority is necessary for cross-platform citation, not sufficient.

Can anyone guarantee my brand gets cited by ChatGPT?

No — and the data is unambiguous about it. AI citation is stochastic: the same query, seconds apart, on the same platform, produces different citation sets. There is no "position 1" to hold. Anyone selling guaranteed citations is selling certainty in a system that is structurally incapable of providing it.

So what actually increases the probability of citation?

Two things. First, genuine domain authority — real link acquisition and brand ubiquity (brand mentions correlate with AI visibility at r = 0.664, far above backlinks at r = 0.218). Second, content structure: across all platforms, citations clustered around three archetypes — "Best X" roundups, product documentation, and step-by-step framework articles. RAG retrieval selects on structural patterns, not just domain authority.

Do the platforms behave differently from each other?

Distinctly. Claude cites like a researcher — 41–56% vendor documentation and primary sources. ChatGPT cites like a librarian — 58% guide-type content. Gemini cites like a journalist — 36% media citations, favoring Forbes, PCMag, and TechRadar. A single "AI visibility" strategy optimizes for a platform that doesn't exist.

Why should a small or indie brand care?

Because product discovery is moving to chatbots, and the citation gate runs on authority signals most small brands don't clear. Below DR 50, the cross-platform citation rate in the data was zero. If your buyers ask AI "what's the best X," the answer is being assembled without you — regardless of product quality.

What are the study's limitations?

Three honest ones. Single collection date (April 4, 2026), so results are a snapshot — and longitudinal data elsewhere shows 40–60% monthly fluctuation in AI citations. ChatGPT's API failed to search on 5 of 15 queries, limiting some API comparisons. And N=15 queries was designed for depth across intent categories, not statistical inference. Replication is invited — the protocol is fully documented.

Where can I read the full study?

The complete PDF — methodology, citation tables, and the DR threshold data — is free just enter your email and get access to your study. If you replicate it, extend it, or find a hole in it, I want to hear about it. That's how evidence works.

I asked the top AI platforms the same question. Here are the results

Here’s a number that should bother you: 0.05.

That’s the average overlap between the sources Claude, ChatGPT, and Gemini cite when you ask them the exact same question. Not 50%. Not 15%. Effectively zero.

I ran a controlled experiment — 15 standardized queries, 3 platforms, 3 interfaces (mobile, web, API), 135 total runs, 750+ citation events. Then I cross-referenced every frequently cited domain against its full SEO profile via Ahrefs’ API. To my knowledge, nobody had done this at the query level before. The aggregate studies (Profound’s 680M citations, Semrush’s 78.6M searches) told us overlap was low. The controlled version shows it’s worse: for a query about CRM software, zero domains were shared between ChatGPT and Claude.

But the finding that actually matters for anyone spending money on visibility is this one:

The same platform doesn’t even agree with itself.

ChatGPT’s web interface never cited a single domain below Domain Rating 50 — a de facto DR 85+ gate. ChatGPT’s API cited a domain with DR 4.5 and roughly 12 organic visitors a month, six times, in one query. Same company. Same model family. Completely different retrieval pipeline.

I’m calling it the Two-Tier Citation System: the consumer-facing web product runs quality controls that the API — the thing powering every chatbot, integration, and “AI research tool” built on top of it — does not share.

Three more findings worth your attention:

Platforms have personalities. Claude behaves like a researcher (41–56% vendor documentation and primary sources). ChatGPT behaves like a librarian (58% guides and how-to content). Gemini behaves like a journalist (36% media citations — Forbes, PCMag, TechRadar). If you’re optimizing one strategy for “AI visibility,” you’re optimizing for a platform that doesn’t exist.

Vendors cite themselves. On commercial and comparison queries, 40–56% of all citations went to the vendors being compared. Ask an AI to compare project management tools, and it quotes asana.com and monday.com back at you.

Academic sources barely exist. Across all 135 runs, .edu and .gov domains appeared in fewer than 2% of citations — even on queries with substantial academic literature behind them.

So what does this mean if you’re buying or selling GEO services?

It means “ranking” in AI citation is a category error. There is no position 1. The same query, seconds apart, on the same platform, produces nearly disjoint citation sets. AI citation is stochastic — the best you can do is raise the probability of being cited, never guarantee it.

Which leads to the only recommendation in the study I’d put in bold: be deeply skeptical of anyone promising guaranteed AI citations. The data says that certainty doesn’t exist. Selling it is selling something that isn’t real.

The full study — methodology, tables, and the DR 85 threshold data — is free, just enter your email to receive it. Because the industry has enough fabricated urgency already; what it’s missing is evidence.

I Asked 3 AIs the Same Question 135 Times. They Almost Never Agreed on Sources.

DOWNLOAD NOW

What you need to know:

▸What exactly did the study measure?

▸What was the single biggest finding?

▸What is the "Two-Tier Citation System"?

▸Does this mean SEO is dead?

▸Can anyone guarantee my brand gets cited by ChatGPT?

▸So what actually increases the probability of citation?

▸Do the platforms behave differently from each other?

▸Why should a small or indie brand care?

▸What are the study's limitations?

▸Where can I read the full study?

Subscribe to our Newsletter.

Or Keep Reading...

Brand Strategy & Consulting: The Complete Guide to Building and Scaling a Strategic Brand

I Asked 3 AIs the Same Question 135 Times. They Almost Never Agreed on Sources.

Cosmetics Terminology Decoded: What Benefit, Ilia, and Merit Beauty Teach You About Building a Brand