This week I ran a keyword gap study for a client. Their site has real authority. Hundreds of pages, a healthy backlink profile built over years, and top-three positions on the terms that matter most to them.
Then I pulled apart the AI answers on those same terms, claim by claim. I did it several times, such as using anonymized searches and external SEO tools, to ensure my search and browsing histories weren’t interfering. And across their three best terms, Google’s AI Overview made 20 distinct claims.
It only cited my client on 5 of them.
On one term, they rank third organically and got cited zero times. The surprise came when I found that the sites that took those claims had less authority, fewer backlinks, and thinner content. The difference? They just answered the exact sub-question Google asked. My client answered a related one.
That gap, between ranking well and barely existing inside the answer, is what this piece is about.
What a fan-out query actually is
When you ask Google (or any AI search tool) a question now, it doesn’t match your query to a page anymore. It takes your one prompt and expands it into multiple related sub-queries to gather broader evidence before producing an answer.
The industry calls these fan-out queries.
For example, ask about consulting hourly rates and it quietly generates four sub-questions. What’s the overall range? What do rates look like by experience? By industry? How do big firms compare to independents? It then answers each one from whichever source it judges best for that specific claim, and it will happily use four different websites in a single answer.
For content strategy, that means you need to cover the likely subtopics and follow-up intents around a question, not just the exact keyword phrase.
If you’re not new to SEO, and topical clusters and “People Also Ask” feel familiar here, that’s because this is their descendant. Same instinct. New enforcement.
The unit is what changed
The old SEO keyword-driven playbook worked. One page per keyword, build out the library, and whoever covered the most queries with the best pages won. I ran that playbook for two decades and the results were real.
That logic still holds. Coverage still wins.
What changed is the unit you count it in.
You used to count keywords. Now you count claims. One page can win five claims inside a single answer. Or it can win none at all.
And here’s the part that trips up established sites (myself included): expertise (or experience), authority, and trust (E-E-A-T) still matter, but they behave differently now. They decide whether you’re eligible to be cited at all. But they don’t decide which claim you win.
That’s why my client, with more expertise, authority, and trust than every site beating them, still lost 15 of 20 claims. Eligibility isn’t ownership.
What keywords fail to answer
Traditional SEO was built on keywords. And then it switched to topics. After all, natural language and semantic search allowed for a better understanding of context and intent. But now, AI takes it even one step further.
For example, one of my clients is a training firm for consultants. Their buyer is the established firm owner already making six or seven figures, not the beginner just starting out. And for many years, they ranked extremely well for the term “consulting fees”.
But “consulting fees” alone is the same string whether a beginner types it or a seven-figure firm owner does. There’s a difference between “how do I set my consulting fees” and “how do I raise my fees with existing clients”.
The full question gives the asker away. As buyers have shifted from typing keywords to asking questions, the query itself tells the machine who’s asking.
Your page has to match that reader, or it loses to one that does.
If this is your kind of thinking, subscribe. Medium-form pieces on positioning, revenue architecture, and the craft of consulting.
The recent shift I had to make
For a long time I’ve used a mental model I call the Triple-P when assessing a client’s organic visibility and to keep reports honest. They’re three layers read as a chain, where one builds on top of the other.
Presence, prominence, and performance.
Presence asked whether you showed up. How many pieces of content you had, how many keywords they covered, how many impressions they earned.
Prominence asked whether it counted. Of everything you published, how much of it actually ranked, for which queries, and how visibly (i.e., how high they ranked on the search results page).
Performance asked whether it paid. Did the visibility align with real searches, earn the click, and drive traffic that turned into pipeline.
The layer that breaks tells you where the constraint sits. For example, strong Presence with weak Prominence means you’re visible to the wrong people. But strong Prominence with flat Performance means attention that doesn’t convert, which is usually a positioning problem and not a search problem.
Essentially, rankings were always the middle layer. They were never the goal. Which is why the model survives the AI shift intact. Only the units change.
Same three Ps but with new units
Presence is now the claims you own. Not pages. Not keywords. Specific, defensible statements your content makes that a machine could lift and attribute. Last week I wrote about expertise being packaged to be cited. The claim is the unit of packaging. A thousand pages that make no liftable claims is a library with no inventory.
Prominence is now coverage. Claims won over claims made. My client’s 5 of 20 is a Prominence score, and it’s countable in a way “average position” never honestly was. Brand mentions inside AI answers sit in this layer too, because being named in the answer is the new version of ranking in the blue links.
Performance is now demand. Here’s the data point you need to pay attention to: one page I checked was the most-cited source in its answer, quoted four times, and it earned two clicks. Citations don’t produce visits. The machine answers the question and offers to keep helping. But on that same site, brand searches are up 28% and direct traffic is up 13% while organic clicks fall. Graded on sessions, the work looks like failure. Graded on demand, it’s working.
Claims are the new keywords. Coverage is the new rankings. Demand is the new clicks.
And if you’re wondering why clicks can’t carry the Performance layer anymore, it’s because most searches no longer produce one. SparkToro puts zero-click at 68% of US Google searches this year. The AI answer absorbs the visit. (I broke down the full numbers in an earlier piece.)
But the buyer didn’t vanish with the click. They resurface later as a brand search, a direct visit, or a “we saw you mentioned somewhere” on a sales call. That’s the demand I’m telling you to grade.
And that’s part of good positioning work.
What to count now
Here’s the exercise I recommend, and it takes an afternoon.
Pick five money questions, plus five questions only your real buyer would ask. Pull the AI answer for each (use a private browser to roughly anonymize your results — it’s not perfect, but it’s better than your current browser).
Break every answer into its individual claims. Count two numbers: claims made, claims won. That ratio is your new scoreboard, and you rerun it monthly, because these answers reshuffle constantly even when their substance holds.
Then look at the claims you lost. In my experience, most of them aren’t lost to better expertise. They’re lost to better formatting. The claims you win sit in liftable modules with clear labels, many with a number in them. The ones you lose are buried in prose.
That's structural work, where it used to be just writing work, and it’s exactly what my SOME framework exists for. SOME stands for Summarize, Organize, Modularize, and Emphasize.
Google even ships the receipts now. Search Console added a generative AI performance report at the end of August. The measurement excuse is gone.
Know a marketing leader still reporting average position to their board? Send them this.
The scoreboard is the strategy
This year, my client (and many other firms I’ve worked with or heard from) is seeing some dramatic shifts in organic visibility and resulting traffic.
My client ranks third on a term where the answer barely knows they exist. Six months ago that would have read as a win in every report they ran. Nothing about their site changed. The scoreboard did.
Count what the machine counts. Claims, coverage, demand. The logic you already know still wins the game. It just got a new unit of measurement.
If your rankings look healthy while your traffic and pipeline drift apart, that’s the Triple-P read I run in fractional engagements. Let’s have a chat.
About the author
Michel Fortin is the creator of Power Positioning and fractional CMO, CRO, CSO, or CGO for growth-stage firms and expert-led practices. Over 35 years and 200+ industries, his work has influenced over $3 billion in revenue by combining deep positioning expertise with AI-amplified marketing strategy. He’s the author of Power Positioning and a recognized voice on organic visibility, revenue architecture, and authority-driven growth.
If your growth is stalled and you’re not sure whether the constraint is functional or architectural, book a diagnostic call. Learn more about his approach at michelfortin.com.