An SEO Walks Into A Garden 🪴

An SEO walks into a garden, a yard, a lawn, a back yard, a bed, a field, a greenhouse, a nursery, a patio, a terrace 😄 you know the joke!

I was trying to figure out why my flower pot wasn’t blooming again after cutting, so I went to Perplexity and searched:

"I used miracle grow all purpose liquid fertilizer for flowers after cutting and they didn't grow"

I got the answer I needed — but, being the SEO that I am, I clicked on the "Steps" tab to see how Perplexity generated the response. And what I saw looked a lot like query fan-out.

I don't think it's exactly how Google does it, but the concept is the same. Perplexity broke my query into smaller, related queries and combined information from multiple sources, not necessarily ones ranking for my original question.

In this blog, I’ll explore a few key topics related to query fan-out, like what kind of content works best in this new era, how to find out which of your content is being used to answer AI prompts, and more.

TL;DR

• Query fan-out is when a main query is broken into a number of sub-queries or related queries to generate a better, more complete answer.
• Shorter, more specialized content tends to perform better in this context.
• Use GA4 to see traffic/clicks from LLMs, and use log files to find how many times your content/website was used in answering users prompts in LLMs.
• Measuring traffic and impressions in LLMs is important, since ChatGPT-driven visits are converting at much higher rates.
• SEO and GEO are not the same. Maybe there's an overlap, but they're not an exact match. I mean for starters, we know that traditional search does not include "query fanout" tactics right?
• Vector embeddings are important in a query fanout world, but also handle them with care, as optimizing for them solely may backfire and any gains achieved can be lost.
  

What is Query Fanout?

Query fan-out is when a main, often complex query is broken down into several sub-queries or related queries to better answer the original one.

For example, for the query:

"how to grow flowers in July in Ontario"

a few synthetic queries maybe created to be able to answer this complex query. Here's one way this may look like:

Long form or short form content?

The term "query fan-out" has been trending lately, and a lot of SEOs — myself included — are wondering how to create content optimized for this concept.

Should we still write in-depth, comprehensive content that covers every angle of a topic? Isn’t that the advice we've always followed? And does that advice still hold in the world of AI search and LLMs?

Not according to Olaf Kopp! Olaf mentioned few important things that we may have missed:

1. LLMs struggle with long content, often missing important details in the middle.
2. Specificity increases your content’s chances of being selected by LLMs.
3. LLMs intentionally avoid citing a single source too many times to promote diversity.

So is longer content still better for SEO?

I'm going to say specific specialized content is better for SEO. I still wouldn't go too short. Look at the results from reddit, they're not lengthy in-depth blogs, they're mostly short answers for very specific questions.

However, on the other hand:

Koray 's framework is all about "How to make our web entity cheaper for Google". There are 2 main costs Google and search engine face when trying to return results to your query:

• Cost of finding relevant content
• Cost of assessing the quality and integrity of this content

and my understanding is that Koray prefers long form content, so I'm curious to know his thoughts on Olaf's notes above!

Clicks and Visibility in LLMs

So, “query fan-out,” huh?

A user types a complex, long-tail prompt. The LLM breaks it down into smaller, relevant queries, then pulls and combines content from multiple trusted sources.

Now, the real challenge: How do we track this?
How do we know:

• When our content is being used?
• What content is driving traffic?
• How often LLMs are referencing our site?

In traditional SEO, we track impressions and clicks in the SERPs. But what about ChatGPT or other LLMs?

How to clicks traffic from LLMs and Chatgpt?

You can easily track traffic your website is getting from LLMs in GA4. There are tons of resources online on that, if you're not tracking this, you are definitely not doing SEO reporting for 2025.

Screenshot from my website's GA4 AI traffic report:

How to track impressions in LLMs and Chatgpt?

LLM impressions: how many times your content was used to answer a prompt without the user necessarily clicking through.

But how to track that?

Come in, Log file analysis

Dan Hinckley recommended reviewing logs for the ChatGPT-User agent (using a tool like Screaming Frog’s log analyzer), so you can get real data on:

• How many conversations included your content
• Which specific URLs were pulled in
• What topics you’re being sourced for

Why does this matter?

Because we’re still figuring out:

• How to optimize for LLMs
• How to influence what LLMs say about us
• And how to prove our value to clients in this new search landscape

Dan also mentioned that "ChatGPT-driven visits are converting at much higher rates". That alone makes this worth tracking.

Hyper-personalization

This ties directly to our long vs. short content debate.

Google wants to personalize results. That means it will favor specific, tailored content over generic long-form articles.

From a recent webinar (on my must-watch list for this week!), here are some key insights (via NotebookLM):

• Google uses personal data from Gmail, browsing history, past purchases, etc.
• Vector embeddings help calculate relevance.
• Personalization increases traffic to more sites, rather than concentrating it on a few top-ranking pages.
• The SEO industry is generally not prepared for this shift, often dismissing it as "just SEO" despite third parties defining "GEO" as something different.

To vector or not to vector 😄

This section leans on Marie Haynes’ excellent article "Could optimizing for vector search do more harm than good?"

What are vector embeddings?

They are a way to turn text into numbers. This helps in understanding and comparing different pieces of text. For example, you can use these numbers to see how closely a piece of content is relevant to a keyword.

Why are vector embeddings important?

Vector embeddings are important in this new era of AI search. I've also seen people use this technique in so many useful ways like URL mapping for example!

In the screenshot below, you can see that Google explicitly tells us that "vector search" is the same technology that powers Google search.

So we should optimize for vectors, right?

Marie’s take: Be careful.

Yes, she’s had success ranking with vector-optimized content, but it didn’t always last.

Why?

If you consistently look great to vector systems and get chosen by Google as a relevant result — but user behavior suggests otherwise — your rankings may drop

In short: Don’t trick the system. User satisfaction still rules.

And That’s a Wrap (Almost 😄)

Query fan-out is important, but not that complicated. Break complex queries into smaller ones, find relevant answers, and combine them.

Honestly, we’ve been doing this in SEO for years already! Targeting long-tail keywords and their variations.

Vector embeddings? Super useful, for relevance calculations, for keyword clustering, for a dozen things. In my Python for SEO course, I show how to calculate URL-keyword relevance using vectors. Once you understand the concept, the applications are endless.

But like anything in SEO, handle with care.

Don’t chase trends blindly. Don’t forget the user. And please, don’t rely on Google's most ambiguous advice “create helpful content” without taking the user into account.

That's that for today folks! And see you next newsletter.

*Disclaimer: Chatgpt fine-tuned the wording after I finished all the writing!