I don't want to say i was listening...watching over your e-shoulder or anything...but...agents, you know? LOL ;) Kidding aside, glad the timing was good and that you enjoyed it! Thank you!
This was an amazing article, thank you. The one thing I keep on coming back to is the hyper-personalization of AI prompts and what you wrote in your last paragraph. I don't think marketers are talking about this enough. The inconsistency is a huge marketing blackbox. Anyone here for a minute knows AEO/GEO (or whatever we want to call it) is built on core SEO foundations, but I wonder in the effort to try and gain "AI visibility," we will just come back full circle to good ole' SEO. I keep seeing keyword research be dismissed for topical authority, which of course as an SEO I agree with, but and also - when you can't really account for personalization on the user-level and inconsistency from the platform level, we might find ourselves going back to the fundamentals like "what do people actually search for that they might include in a prompt?" Maybe I am thinking about this too simply, but your article got me thinking either way.
You're not thinking too simply at all, Nina. You're identifying something real. The personalization and inconsistency layer genuinely does make this a measurement black box, and the industry is largely pretending otherwise right now.
On the keyword question: your instinct is right, but I'd frame the reason slightly differently. Prompt language borrows heavily from search query habits because people have been training themselves to query machines a specific way for 25 years. That muscle memory is real and observable, and understanding how people phrase questions remains useful signal.
Where I'd push back gently is on the mechanism. Keyword research in traditional SEO was useful because it mapped to a ranking system that matched phrase patterns to documents. That specific mapping doesn't exist the same way in LLM responses.
Topical authority isn't replacing keyword research because keywords became unfashionable. It's because topical authority more accurately describes the underlying mechanism; how well and consistently a model represents your brand within a knowledge domain. That's a different job than query-to-document matching, even if the content work looks similar from the outside.
The black box problem you're naming is exactly why measurement infrastructure matters so much right now. We don't yet have the practitioner tools that translate the underlying mechanics into actionable signals the way SEO tools eventually did. That gap is where a lot of the real work is happening.
I love this article mostly for the citations. Outstanding.
It also points out some major issues with AI. Silos are a HUGE problem. Bias is also a HUGE problem based on the LLMs used. Recency has its place. But therein lies a huge error. Timeliness and recency are not always the right context.
For example, Gestalt Principles may be “old” in Internet years. But they are still applicable now.
My favorite part is that his article makes me think…as a retrieval scholar. I do like Perplexity. But it absolutely needs a human-in-the-loop.
Great write up Duane! I was just having this exact conversation today so the timing is impeccable.
I assume my privacy is being invaded at all times.
Politely. ;) Demurely. With great respect...and an expectation of monetization, of course... LOL
I don't want to say i was listening...watching over your e-shoulder or anything...but...agents, you know? LOL ;) Kidding aside, glad the timing was good and that you enjoyed it! Thank you!
This was an amazing article, thank you. The one thing I keep on coming back to is the hyper-personalization of AI prompts and what you wrote in your last paragraph. I don't think marketers are talking about this enough. The inconsistency is a huge marketing blackbox. Anyone here for a minute knows AEO/GEO (or whatever we want to call it) is built on core SEO foundations, but I wonder in the effort to try and gain "AI visibility," we will just come back full circle to good ole' SEO. I keep seeing keyword research be dismissed for topical authority, which of course as an SEO I agree with, but and also - when you can't really account for personalization on the user-level and inconsistency from the platform level, we might find ourselves going back to the fundamentals like "what do people actually search for that they might include in a prompt?" Maybe I am thinking about this too simply, but your article got me thinking either way.
You're not thinking too simply at all, Nina. You're identifying something real. The personalization and inconsistency layer genuinely does make this a measurement black box, and the industry is largely pretending otherwise right now.
On the keyword question: your instinct is right, but I'd frame the reason slightly differently. Prompt language borrows heavily from search query habits because people have been training themselves to query machines a specific way for 25 years. That muscle memory is real and observable, and understanding how people phrase questions remains useful signal.
Where I'd push back gently is on the mechanism. Keyword research in traditional SEO was useful because it mapped to a ranking system that matched phrase patterns to documents. That specific mapping doesn't exist the same way in LLM responses.
Topical authority isn't replacing keyword research because keywords became unfashionable. It's because topical authority more accurately describes the underlying mechanism; how well and consistently a model represents your brand within a knowledge domain. That's a different job than query-to-document matching, even if the content work looks similar from the outside.
The black box problem you're naming is exactly why measurement infrastructure matters so much right now. We don't yet have the practitioner tools that translate the underlying mechanics into actionable signals the way SEO tools eventually did. That gap is where a lot of the real work is happening.
I love this article mostly for the citations. Outstanding.
It also points out some major issues with AI. Silos are a HUGE problem. Bias is also a HUGE problem based on the LLMs used. Recency has its place. But therein lies a huge error. Timeliness and recency are not always the right context.
For example, Gestalt Principles may be “old” in Internet years. But they are still applicable now.
My favorite part is that his article makes me think…as a retrieval scholar. I do like Perplexity. But it absolutely needs a human-in-the-loop.
🙏🙏👍👍