An analysis of 4,915 AI answers shows that AI-written and AI-assisted content can earn significant ChatGPT citations when it is structured for LLMs, backed by evidence and built around the questions buyers ask.
If I was a consumer tech CMO, I’d be worried about AI-written content and its implications.
Why?
Primarily because so many expert voices are claiming it does not work for GEO, and might even be morally reprehensible.
Well, dear CMO, let me show you some real numbers, from my company that will give you perspective on AI-written content’s impact on GEO, specifically what LLMs cite in answers. Yes, I am offering just one data point, but I am quite sure it will assuage your fears.
The chart below shows citation data from ChatGPT (I picked this LLM because it is, in aggregate, generally viewed as a platform people consult when shopping).
You can see that across 177 unbranded prompts and 4915 AI answers that span the funnel for a tech PR agency, Proper Propaganda is ChatGPT’s favorite source.
Our lead as ChatGPT’s most cited source is not small. We more than double the next closest property, and over the last 84 days we are well ahead of Reddit and the scholarly website arXiv – two places often extolled as important for GEO programs.
The point is not to chest thump, but to tell you about the counterintuitive way we got here, because there’s a lot of broad generalizations made about GEO on the web.
A few common refrains:
- Earned media matters way more for GEO than owned media
- Get on Reddit and you win
- ChatGPT loves Wikipedia
And, my personal favorite:
- AI written content is all slop and will get you penalized. Write content by humans, for humans.
I think the little chart I shared at the top of this post addresses the first three generalizations du jour adequately.
The fourth bit of generic advice is the interesting one.
Again, Proper Propaganda’s data is useful.
The lion’s share of our most-cited-by-ChatGPT content is written primarily by bots – with a human slightly in the loop – and is intended for a bot audience.
The above screencap shows the top 5 pieces of Proper Propaganda content in terms of ChatGPT citations. Those with a red line fit into the “written by bots, for bots” category.
You could go and read them if you really wanted to.
- They aren’t complete shit (though they aren’t winning any writing prizes).
- They aren’t really done in our company’s voice.
- But, mostly, as I said, they are not meant for a human audience.
Does that mean we should change our name to Sloppy Slopaganda?
No.
I have detailed our content strategy – which we call Content Fast and Slow – before. We see non-humans as an important customer archetype and the mediator to buying our services.
And non-humans, aka RAG bots, have fundamentally different preferences around written content. For example, they don’t seem to give much of a damn about voice, and pizazz despite what Google and a range of authorities tell us. It seems the Mara RAGvatrucha like clarity, structure, explainability and evidence. They want to be able to extract easily and seem (overly) preoccupied with methodology and (the appearance of) authority. These are all elements AI can be prompted to create, and create well.
All the faux outrage about using AI to produce slop misses the point. Our data shows, unequivocally, that there IS a place for AI written, and AI-assisted content in a GEO strategy.
Deploy it when and where bots are your audience. More importantly, ignore the ideologues seeking to reduce everything to binary interpretations.