Hi 👋🏻
In the last edition, we explored query fan-out — how AI search systems decompose your query into dozens of synthetic subqueries, retrieve passages in parallel, and synthesise a response. If you haven’t read that one yet, I’d recommend starting there, as this edition builds directly on it.
Today, we’re going deeper into the piece that makes fan-out truly transformative — and truly challenging for SEOs and digital marketers: personalisation. Specifically, how AI search systems use memory, context, and behavioural signals to shape which subqueries get generated in the first place.
Remember when we discussed personalisation of traditional search results in #008? We covered how Google uses session context, historical behaviour, information gain scoring, and implicit feedback to re-rank results. The key word there was re-rank. The query stayed the same across users; only the order of results changed.
AI search systems move personalisation upstream. They don’t just re-rank results — they rewrite the questions. And that changes everything.
From Personalised Rankings to Personalised Queries
Here’s the core shift, illustrated with an example from my recent iPullRank deep-dive on this topic.
Two users both search “best running shoes.” One is an experienced urban marathon runner. The other is a beginner preparing for their first half-marathon, training on trails near their home. Same query. Same goal — preparing for a race.
The experienced runner’s fan-out might generate subqueries like: “marathon running shoes carbon plate,” “lightweight racing flats urban marathon sub-4 hour,” “running shoes durability concrete asphalt long distance.”
The beginner’s fan-out might generate: “marathon training shoes for beginners mixed terrain,” “cushioned shoes for long distance beginners dirt paths,” “durable running shoes marathon prep unpaved surfaces.”
Same input query. Different contexts. Different subqueries generated. Different documents retrieved. Different brands recommended. Neither user specified their experience level, terrain, or training context. The system inferred all of it from past interactions, memory, and context signals.
How Each Platform Personalises Fan-Out
Not all AI search systems personalise at the same depth. In my iPullRank research, I mapped each platform on what I call the inference depth spectrum — from shallow (using only what you’ve told the system in this session) to deep (inferring from your entire digital footprint).
Three critical patterns emerge from this comparison:
- The upstream shift is universal. Every AI search platform applies personalisation during query expansion, not during ranking. This means you’re no longer fighting for position within a stable results set — you’re fighting for inclusion in a query set that varies by user.
- Inference depth correlates inversely with transparency. Perplexity shows you the most but personalises the least. Copilot personalises the most but operates as a black box. The platforms where personalisation matters most are the ones where you can observe it least.
- The data source spectrum determines what you’re competing against. On Perplexity, your content competes against other public web content. On Copilot, it competes against the user’s own internal documents. The meaning of “relevance” is now platform-dependent.
The article goes in-depth on the different platforms and the signals they use to personalise responses, alongside experiments I've ran.
What This Breaks
Let’s be direct about the implications. Personalised fan-out fundamentally breaks several pillars of traditional SEO measurement:
Position tracking becomes impossible. There is no stable results page to rank on. Your content either made it into the synthesised response, was used but not cited, or wasn’t part of the consideration set at all. This varies by user context, system memory, and model training.
Impressions become meaningless aggregates. An impression count doesn’t tell you what percentage of relevant queries your content was even eligible for. You might have 10,000 impressions but be invisible to 50,000 users whose personalised fan-outs never generated subqueries your content could answer.
Filter bubbles deepen invisibly. When every query expansion is filtered through the user’s established profile, the system optimises for confirming what it already knows about them. One of the under-appreciated values of generic search was spontaneous discovery — finding things you didn’t know you were looking for. Personalised fan-out erases that. Users don’t see what was excluded. They experience a confident answer and assume it’s comprehensive.
What to Do About It
You can’t control how AI systems personalise fan-out queries. You can’t see user contexts. You can’t optimise for a specific user’s profile.
Accepting this is step one (and it's a tough pill to swallow).
Here’s what you can control:
Think in contextual intersections, not keywords.
- A keyword is single-dimensional: “running shoes.”
- A contextual intersection is multi-dimensional: “running shoes” + “marathon training” + “trail terrain” + “beginner level” + “budget conscious.”
Fan-out generates queries along these intersections. Your content strategy should map the intersections relevant to your business and ensure you address each one. This doesn’t mean creating thousands of hyper-specific pages. It means ensuring your core pages explicitly address the contextual modifiers that your target users’ profiles likely contain.
Invest in user research like never before.
If AI systems personalise based on user context, then understanding those contexts is foundational.
- What does your ideal customer’s AI search profile look like?
- Which platforms do they use?
- What step of their purchase journey do they integrate AI search in?
- What can be inferred from their digital footprint?
Persona mapping isn’t just a brand exercise anymore — it should spill through tone of voice, content strategy, all the way through to technical SEO.
Niche authority beats broad evergreen content.
When the system generates a subquery like “best trail running shoes for first marathon under €150,” it’s looking for content that addresses that specific intersection.
A 10,000-word guide that briefly mentions trail running, briefly mentions marathon training, and briefly mentions budget is less relevant than a 1,500-word piece that deeply addresses exactly that combination.
Specificity wins.
Maintain entity consistency across your entire content footprint.
AI systems build entity associations from everything you publish — not just your website, but social media, forums, YouTube, third-party mentions.
If your brand sends mixed signals about who you serve, fan-out systems receive mixed signals about when to surface you. Decide who you are, who you serve, and what contexts you want to own. Then ensure your entire ecosystem reinforces those associations.
Build comparison content for specific personas.
Not generic “X vs. Y” listicles. Instead, content that explicitly argues why your product is the right choice for a specific user profile, based on the real entity attributes you win on.
This increases your chances of surfacing when fan-out queries include those contextual modifiers, and creates co-citations that introduce your brand as a valid alternative in the right conversations.
The Bottom Line
In the AI search era, context — not content — is king.
For users, their context determines their reality. What the system knows about them shapes which questions get asked on their behalf, which shapes what gets retrieved, which shapes what they see.
For brands, the game is different. You can’t see user context. But you can control how well your content speaks to the contexts that matter for your business. The brands that will thrive aren’t those with the most content or the highest domain authority. They’re the ones with the clearest entity positioning, the deepest niche authority, and the most deliberate alignment between their content and their target users’ likely contexts.
Less “create comprehensive content and build links.” More “understand exactly who we serve, what contexts they carry, and how to be the definitive answer when those contexts shape the queries.”
Learn more on this topic in our Featured Courses 🌟
The concepts in this edition connect directly to what we teach in the Semantic AI-powered Keyword Research course (covering personalisation, intent classification, and synthetic queries) and the AI Search & LLMs: Entity SEO and Knowledge Graph Strategies for Brands course by Beatrice Gamba (covering entity auditing, knowledge graph integration, and brand entity development for AI search).
135+ forward-thinking marketers are already taking our courses 💜
Further Reading 📚
For the complete analysis with platform experiments, patent references, and the full personalisation comparison matrix:
Community discussion 🌟
I recently asked the MLforSEO community about their approach to synthetic queries and whether this new facet of search is incorporated in their keyword research and content strategy. Responses varied, with many users having integrated synthetic queries into their keyword research pipelines, though the majority use them to help cover a greater content depth. favourite tools for query fan-out generation and synthetic queries.
Here are some approaches shared:
This week's discussion topic is related to structured data and it's impact on AI search visibility, namely How has your structured data strategy changed as a result of AI Search growth? - join the discussion on Slack.
Recent resources shared within our Slack community 💬
(If you haven't already...) Join 750+ AI/ML-interested marketers on our Slack community to stay up to date with discussions on AI/ML automation in SEO and marketing.
Happy learning! ✨
Lazarina
