When people try to get their companies to appear in AI answers they often look at existing citations by AI and think "If I write an article just like the ones being cited I can rank for that prompt." I decided to explore this theory a bit.
I opened ChatGPT and Google AI Mode in incognito windows and asked both the same intentionally vague question:
"What are the best solutions for hosted database platforms?"
ChatGPT gave me six recommendations, in this order: Neon, Supabase, AWS Aurora PostgreSQL, PlanetScale, MongoDB Atlas, and CockroachDB Cloud.
Google AI Mode gave me ten: Supabase, MongoDB Atlas, PlanetScale, Neon, Amazon Aurora/RDS, Amazon DynamoDB, Google Cloud, Azure SQL, Snowflake, and BigQuery.
Four brands appeared on both lists. That means 66% of ChatGPT's recommendations overlapped with Google, while 40% of Google's overlapped with ChatGPT. Not perfect agreement, but not chaos either.
The citations were much less aligned.
ChatGPT cited 56 sources. Google AI Mode cited 30. They had only six sources in common: Reddit, YouTube, LinkedIn, Medium, GeeksforGeeks, and Northflank.
That is 11% of ChatGPT's citations and 20% of Google's.
In other words, the engines had some agreement on which brands belonged in the answer, but very little agreement on which sources helped justify the answer. The overlap that did exist came largely from social, community, and developer-oriented sites.
I ran the test again with a much more specific query: "What are the best solutions for a hosted database if you're a small company (under 10 people) with a current tech stack of Node, PostgreSQL and AWS Lambda, and a nascent user base?"
ChatGPT recommended six brands: Neon, Supabase, Amazon RDS, Amazon Aurora, Crunchydata, and Railway. Google AI Mode recommended four: Neon, Amazon Aurora, Supabase, and Railway. Every one of Google's picks appeared in ChatGPT's list. Brand overlap went up sharply, which is what I expected.
What I did not expect: citation overlap barely moved.
ChatGPT cited 47 sources. Google cited 27. Only five appeared on both lists, roughly 11% of ChatGPT's citations and 19% of Google's, and most of these are the big social media sites. Nearly identical to the vague-query result.
So specificity pulled the brand recommendations closer together, but the underlying evidence bases stayed almost completely disjoint. The engines converged on what to recommend while continuing to disagree on why.
My test was only two queries, so I would not read too much into it by itself. But it lines up with the larger Semrush research that Eli Schwartz shared on LinkedIn.
According to that data, ChatGPT and Google AI Mode share roughly 69% of their most-mentioned brands, but only 57% of the sources they cite.
The source mix also looks very different by engine. ChatGPT gets more than half of its citations from two sites: Reddit (26.5%) and Wikipedia (24.2%). Google AI Mode's top two sources are YouTube (20.5%) and Facebook (17.2%).
Category matters too. In Food and Drinks, Google AI Mode pulls 77% of its citations from Yelp alone. ChatGPT spreads its citations across Yelp (22%), RestaurantJI (17%), and Uber Eats (12%).
In finance, Semrush found Reddit appearing in ChatGPT's answers at a rate of 176.89%, meaning more than one Reddit citation per query on average. Google AI Mode, by contrast, leans more heavily on Bankrate (86.61%) and NerdWallet (75.07%).
There is also a big difference in citation volume. Semrush's expanded 2026 AI Visibility Index, built on 126 million prompts, found that ChatGPT cites about 15 sources per response. Gemini cites about three.
That difference alone makes "AI visibility" difficult to compare across platforms. A brand could look well-supported in ChatGPT and barely sourced in Gemini for the same broad category.
A lot of AEO conversation treats brand mentions and citations as if they are interchangeable. This doesn't appear to be true.
There are at least two separate problems.
The first is when your brand gets mentioned, but your site is not cited.
Gemini is a good example. Semrush's Ghost Citations study with Kevin Indig found that Gemini names brands in 83.7% of answers but cites sources only 21.4% of the time. So the engine may recommend your product while giving the user no direct path back to your site, documentation, comparison page, or owned content.
That can still create awareness, but it gives you very little control over what the answer is based on.
The second problem is the reverse: your site gets cited, but your brand is not named.
Semrush found that 62% of AI citations are "ghost citations," where a page is used as a source but the brand does not appear in the answer. ChatGPT cites something 87% of the time, but names brands only 20.7% of the time. The same study found that AI engines disagreed on whether to name the same brand in 22% of cases.
Zapier is the clearest example from the Semrush data. It ranks as the #1 most-cited source in digital technology, but only #44 in brand mentions.
That is a strange position to be in. Your content may be helping an AI engine explain the category, while the final recommendation points somewhere else. A citation can be valuable, but citation volume alone does not tell you whether the brand is actually winning demand.
ChatGPT, Gemini, and Google AI Mode retrieve and cite differently. Their source preferences vary by category. Their behavior changes as the products, indexes, partnerships, and retrieval systems change.
The honest conclusion is that nobody — including the platforms themselves — can tell you with confidence why a specific citation gets selected.
My hosted database query would probably return a different citation set if I ran it again today, or next week.
That is why the best AEO advice still looks a lot like good content and distribution strategy. If you are writing about a topic, make the piece original enough to be worth referencing and specific enough to help someone make a decision. Then get it distributed in the places where your category is actually discussed.
Across both tests, citation overlap stayed weak — hovering between 11% and 20% regardless of how vague or specific the query was. But the sources that did appear on both lists skewed consistently toward social, community, and developer-oriented sites: Reddit, YouTube, LinkedIn, Medium, GeeksforGeeks, and Northflank from the vague query; dev.to, AWS, and Neon from the specific one.
Those are places where people explain, compare, debate, and troubleshoot. If your best thinking never leaves your blog, you are asking AI engines to find and trust it in isolation. If it gets discussed, cited, challenged, and reused across the web, it has a much better chance of becoming part of the evidence layer these systems draw from.
The one thing that does change in the AEO era is you have to write a lot more original content for long-tail queries, which is far more prevalent than in the day of search/SEO. And having a scalable process to do is crucial; more on that in a future post.