Keyword Fragmentation: What AI Search Changes for Your SEO
Summarize this article with AI
15 sites per week: all stuck on 2-word keywords
I audit 15 sites per week. Live audit, 45 minutes stopwatch in hand. The finding is always the same. Product pages, blog articles, category pages—everything is built for keywords like « running shoes », « best hotel Rome » or « SEO training ».
Meanwhile, I open Search Console. 42% of organic clicks come from queries longer than 7 words. I’ve verified this across 12 accounts this last quarter. The order of magnitude is stable. Visitors no longer type generic expressions. They describe a situation.
They write: « what type of shoe for running on muddy trail with a 100€ budget ». Or: « where to find a hotel with heated pool in November in Rome for a family of 4 ». Or: « I have an e-commerce site with 800 products, how do I structure my SEO so Google understands my offer ».
Your pages are optimized for 2018. Google has changed.
On April 8, 2025, Liz Reid, VP Search at Google, gave an interview to Bloomberg’s Odd Lots podcast. She described a structural shift. AI Overviews users no longer submit keywords. They pose entire problems. Google must fragment these queries to answer them. Search Engine Journal published a detailed analysis the next day.
What Liz Reid calls « fragmentation » is the process by which AI breaks down your narrative sentence into a dozen underlying questions, then finds pages that answer each fragment. Traditional SEO, which targets only one fragment, becomes invisible.
What Liz Reid actually said: the end of the hammer keyword
I listened to the interview twice. Here’s the core. Reid explains that users have always wanted to express themselves in natural language. But for 25 years, Google trained them to shrink their thinking to 2 or 3 words. « Best restaurants in New York. » When they really meant « vegan restaurant with seating for 5 people tonight in Brooklyn ».
With AI Search, that bottleneck vanishes. People write paragraphs. Reid cites « meaningfully longer » queries. She doesn’t give a percentage, but my ground observation across 12 e-commerce accounts confirms it: narrative queries have doubled in 18 months.
What’s at stake is not trivial. The old SEO method—one keyword = one page—assumed Google ranked the dominant meaning of an ambiguous phrase first. If « restaurant New York » drove 70% of traffic to luxury addresses, your vegan restaurant page never broke through, even with 200 links. Today, Google doesn’t have to choose. It fragments the query, isolates the intent « vegan », « tonight », « Brooklyn », and arranges different content to address it.
The direct consequence: your page can capture traffic even if the global keyword remains dominated by a giant. Provided it’s legible to an algorithm that segments intentions.
« We see longer queries, in natural language. Users describe their situation, their constraints, their budget. Google fragments this sentence to identify the relevant information. » – Liz Reid, VP Search Google
A French e-commerce site, 800 products, 18 months of work
A client calls me on a Tuesday morning. He sells sports equipment. 4,000 organic sessions per month. A catalog of 800 SKUs. Product pages stuffed with keywords like « trail shoes », « waterproof running jacket ». Result? People who land on his site leave. Conversion rate stuck at 0.8%.
We stop producing classically optimized pages. We open Search Console and listen to narrative queries already being captured: 47 long questions identified. We build 14 semantic siloes. Each silo starts from a real user script.
Example script: « I have a 25 km trail planned in November, I run in mud, I have a 110€ budget, I want a shoe that protects me from ankle injuries. » My team and I forge a pillar page « Trail shoe for muddy terrain and controlled budget », surrounded by 7 satellite pages that decompose this narrative: stability, waterproofing, lug soles, comparison of 4 models, reviews, sizing guide specific to this context.
In 6 months, the pages in this silo capture 312 distinct narrative queries. Organic traffic jumps 210%. Conversion rate follows: +37%, averaging 1.1%. All without increasing ad spend.
The DOSE framework: your visitors adopt scripts, let’s reveal them
I rely on the DOSE framework taught by Guillaume Attias at BMO Academy. One of its building blocks is script theory. A script is a recurring mental schema the user activates to solve a situation. Before, Google broke this script into keywords. The visitor adapted. Now, the AI leaves the script intact and fragments its understanding to address it.
Imagine an e-bike buyer. Her script: « I want a bike to commute to work, 15 km per day, hilly city, minimum 70 km range, 2,500€ budget, I don’t want to sweat when I arrive at the office. » Until now, SEO targeted « best city e-bike ». Generic page. Missed the mark. Google AI fragments the script and weighs micro-intentions: range, motor torque, riding position, discrete design.
When I build a silo, I first map the complete user script, with its 6 to 12 sub-questions. Then I explode it into pages that precisely address each fragment. Result: Google no longer has to make risky inferences. It finds exactly what it needs for each piece of the narrative.
From keyword logic to script logic: how to get started
- Audit your narrative queries. Export Search Console, filter for queries longer than 8 words. Group those expressing the same situation (clothing + weather + use + constraint). You get 7 to 15 dominant scripts.
- Map the fragments. For each script, list the sub-questions Google isolates. Use the « Related searches » panel and rich results to confirm the angles the algorithm prioritizes.
- Build a silo. One pillar page addressing the global script. Satellite pages for each fragment. Link them together in downward and lateral fashion. No page should be orphaned.
- Write for humans, structure for the fragment. Each page must autonomously answer a sub-intention. Semantic markup (h2, h3, lists, tables) helps the AI extract information. No stuffing. Precision.
This is not « long-form » content for Google. It’s an architecture reflecting the user’s mental schema. A site that understands what the visitor is looking for before they finish typing.
The old reflex that costs you traffic (and that you’re afraid to drop)
Many e-commerce teams keep creating pages optimized for a generic keyword. « Summer dress women ». Single page, 800 words, 3 photos, buy button. This page competes with 14 other pages on the same site attacking the same keyword from a slightly different angle. The site cannibalizes itself.
Worse: this page doesn’t answer any fragment. While your competitor has structured a silo around the script « summer dress for civil wedding, not too formal, pastel colors, 80€ budget ». Their pillar page captures qualified traffic. You get a click that bounces.
I worked with a women’s ready-to-wear site. 60 « summer dress » product pages scattered everywhere. We regrouped the intentions into 12 siloes corresponding to precise scripts. Monthly organic entries jumped from 6,200 to 19,800 in 8 months. Bounce rate dropped 14 points.
You’re no longer doing keyword SEO. You’re doing intent SEO.
The fragmentation Liz Reid describes is not a nuance. It’s a paradigm shift. AI Search rewards architectures reflecting your customers’ complete narratives. Not isolated pages repeating a keyword.
I’m not selling the method. I’m showing you the pages. My job is to build systems that run without me. Siloes that capture the long tail because they answer every fragment of the need.
When was the last time you looked at the complete sentences your visitors type before landing on you?
Your live semantic audit in 45 minutes
I review your site, the narrative queries you’re already capturing, and show you where to focus your effort to capture AI Search traffic. No commitment.
Book a strategic call — 45 minFrequently Asked Questions
Does fragmentation mean the death of traditional keyword research?
No. Keyword research keeps its full place for identifying major topics. It simply needs to be complemented by listening to narrative queries and user scripts. You keep spotting volume, but now you address it via architecture, not a single page.
How do I know if my site is already impacted by narrative queries?
Go to Search Console, filter for queries longer than 7 or 8 words. If they represent more than 30% of clicks and your pages aren’t structured to address them precisely, you’re losing qualified traffic. My clients often see 40 to 60% of clicks from these long phrases.
Should I delete my old pages optimized for short keywords?
Not necessarily. Better to integrate them into a silo if they add value. A generic page « best e-bike » can become a pillar page if you rewrite it to cover the entire purchase script and link to dedicated satellites.
How many pages per semantic silo should I plan for?
No magic number. A silo covers a complete script. That might be 5 to 12 pages depending on need complexity. The key is each page answers one distinct fragment. No point creating pages for volume: precision wins.
Will AI Overviews take my pages if I build a silo?
Yes, it’s one of the most direct levers. Since AI Overviews fragments the query, it pulls from your satellite pages to feed its synthetic answers. A well-linked architecture increases your odds of appearing as a source, because Google finds all the puzzle pieces at your site.
