Been rolling my eyes at AEO hype for a while, but the train keeps going, so finally ran some experiments. 😅

First finding: ChatGPT and Google AI Mode mostly agree on which brands to recommend, but they pull from almost entirely different sources to back those picks.

opened both in incognito and asked the same vague question: 'What are the best solutions for hosted database platforms?'

ChatGPT suggested Neon, Supabase, AWS Aurora PostgreSQL, PlanetScale, MongoDB Atlas, and CockroachDB Cloud. Google AI Mode went with Supabase, MongoDB Atlas, PlanetScale, Neon, Amazon Aurora/RDS, DynamoDB, Google Cloud, Azure SQL, Snowflake, and BigQuery.

brand overlap was decent: about two-thirds of ChatGPT's picks showed up in Google's answer, and 40% of Google's appeared in ChatGPT's.

Then looked at the actual citations. ChatGPT cited 56 sources, Google cited 30. Only 6 overlapped: , , LinkedIn, Medium, GeeksforGeeks, and Northflank. That's 11% of ChatGPT's citations and 20% of Google's. Honestly, expected more overlap given how similar the recommendations were.

Tried a much more specific query-small company, Node/PostgreSQL/AWS Lambda stack, early user base. brand overlap jumped way up. Every brand Google recommended also appeared in ChatGPT's list. but citation overlap barely moved, still around 11% and 19%.

So specificity pulled the recommendations together but had almost no effect on the underlying sources.

My test is just a couple of prompts, but a larger dataset from Semrush shows the same pattern. Across 126M prompts, ChatGPT and Google AI Mode had roughly 69% brand overlap but only 57% source overlap. And that's before we even talk about citation drift over time.

then there's the ghost citation issue: AI cites your source but never mentions your brand. What's the value of a citation then? semrush found that 62% of AI citations use a page as a source without naming the brand in the answer. Zapier is apparently the #1 most-cited source in digital tech, but only #44 in brand mentions.

All this makes me wonder if trying to reverse engineer how to get cited in AI answers is just... pointless? 🤔

Curious if anyone else has run similar tests, especially in B2B or dev tool categories. Would love to see if the citation-vs-recommendation gap holds up across different query types.