The 2003 Framework That Was Already Doing Query Fan-Out via @sejournal, @gregjarboe

I was doing query fan-out by hand for two decades before anyone called it that. I just called it something else: “Russian nesting dolls.”

In August, MJ Cachón published a dataset study. She ran 189 branded prompts through ChatGPT, watched the model fire off 1,797 sub-queries nobody typed. Reading it, I recognized the shape of something I’d been building into press releases since early 2003, long before “fan-out” or “GEO” existed as terms.

When I used to optimize a release, I looked for a four-word phrase with a three-word phrase nested inside it, the way a Matryoshka doll holds a smaller doll inside a bigger one. Say the smaller phrase was “airfare to Philadelphia.” I wouldn’t stop there. I’d find the four-word version, something like “cheap airfare to Philadelphia,” and build the release around that instead. Use the four-word phrase, and the release can be found by someone searching the shorter term or the longer one. Use only the three-word term, and you’re invisible to anyone who types the longer phrase, because it’s missing a word your page never mentions.

That’s not a clever trick. It’s just built for the slice of search that most SEOs undervalue: The queries nobody has typed yet.

The 15% Nobody Can Target Directly

Google first put a figure on it in its 2019 introduction to BERT, writing that 15% of the queries it sees on a given day are ones it has never encountered before. At Search Central Live NYC in March 2025, John Mueller revisited the number and seemed almost bothered by how stubborn it is. “I would have thought at some point most of the searches would have been made; people just ask the same thing over and over again,” he said. “But when we recalculate these metrics, it’s always around 15%.” He’d expected large language models to push the figure higher. They haven’t. It just sits there, a fixed fraction of a growing pile of searches, decade after decade.

Multiply that 15% against the billions of queries Google handles daily, and you get hundreds of millions of searches every single day for phrasing that has never existed before. Most of it isn’t random. It’s the offspring of breaking news, a new product name, a policy that just changed, a phrase a journalist coined on deadline that a thousand people then typed into a search box within the hour. News creates language, and language creates search volume that didn’t exist yesterday.

Press releases were uniquely positioned to catch that traffic, and I don’t think most SEOs, then or now, understood why. A press release is one of the only content formats written and published on the same day the news happens. A blog post takes a news cycle to catch up. A press release IS the news cycle. If the phrasing in that release happened to nest the exact words someone would later type into Google, in either the three-word or four-word version, the release could rank for a query that didn’t exist when the writer sat down to write it.

See also: An Easy Digital PR Strategy For AI SEO

What Cachón’s Data Adds That My Old Trick Never Had

Here’s where the nesting-doll instinct gets a real upgrade. Cachón found that branded ChatGPT fan-out doesn’t behave randomly either. The first sub-query in a run tends to be plain, conversational language. From there, the model narrows with a “site:” operator, then starts pulling exact quoted phrases to check whether a source actually says what it thinks it says. Across her dataset, quote usage climbed 25-fold between a run’s first search and its last.

That’s the nesting-doll principle running in reverse. I used to build outward from a small phrase to a bigger one so a reader’s query, whatever length it happened to be, would still land on my page. The AI systems Cachón studied are narrowing inward, starting broad, and drilling down to a literal phrase it can verify word for word. Either direction, the underlying requirement is the same: your content has to contain the exact wording, at more than one length, or you disappear from part of the funnel.

I’ve already reported that the average AI Mode query in the U.S. now runs triple the length of a traditional search query, based on Google’s own May 2026 usage data. Line that up next to Cachón’s number, that a single branded prompt fans out into sub-queries averaging seven words each. Length isn’t a side effect of AI search. It’s the terrain now, and it has been building toward this for longer than most of the industry noticed.

See also: Data Shows AI Citation Patterns Reveal Strategic SEO Opportunities

My Take

I think the industry spent the 2010s optimizing for the wrong end of the doll. Head terms got all the strategy meetings and all the budget, while long-tail phrasing got treated as an afterthought that Search Console would surface if you were lucky. That was backward even before generative search existed and it’s more backward now that the systems doing the searching, not just the humans, are the ones fanning a single prompt out into a dozen specific phrasings.

How To Apply This To Your Content Strategy

You don’t need Cachón’s API access to use any of this. You need three habits.

Find the nested phrase, not just the seed phrase. Whatever three-word core term you’re targeting, write down the two or three four-word, and five-word phrases that naturally contain it, then build your opening paragraph or H2 around the longer version. Check Google Search Console for queries with high impressions and low clicks first. Those are usually the longer variants already knocking on your door.

Publish at the speed of the news, not the speed of the content calendar. The 15% of queries that are brand new are disproportionately tied to something that just happened. If your organization has a same-day publishing channel, whether that’s a press release, a company blog post, or a rapid-response page, that’s your best shot at owning language before a competitor even knows the phrase exists.

Write the literal answer as a standalone, quotable sentence. Cachón’s data shows AI systems increasingly verify claims by searching for an exact quoted phrase from your own content. If the sentence that answers the question can’t be lifted whole and still make sense, rewrite it until it can.

None of this replaces the fundamentals. It just means the fundamentals were pointed at the long tail before most of us had a name for it.

More Resources:

Featured Image: Rawpixel.com/Shutterstock