How does AI search work for your category?
AI engines decompose every buyer question into subqueries before retrieval. The subquery space is different for every industry: different constraints, different persona signals, different engines dominating each stage. This page maps how citation works in nine high-priority categories and shows the modeled subquery breakdown for each.
Your buyer asks one question. The engine asks eight. You are missing seven.
This is not SEO. Retrieval-augmented generation (RAG) selects passages, not pages. The selection event is invisible, and it happens before your buyer sees any results.
The head query arrives
A buyer types one question. The AI engine receives it as a retrieval task, not a search query. The two are architecturally different: a search engine ranks documents; a retrieval engine decomposes the question and assembles an answer from fragments.
Fan-out happens invisibly
The engine expands the head query into 6 to 18 subqueries before any retrieval occurs. These are never shown to the buyer. They encode intent, constraints, persona signals, and the gaps the buyer did not know to ask about. Each subquery is sent to a separate retrieval pass.
Fragments are selected, not ranked
For each subquery, the engine holds multiple candidate passages simultaneously and selects one. This is the selection event. Winning it is not about keyword density. It is about carrying the right evidence, scope, and structure for that specific subquery's intent class.
The answer is synthesized from winners
The engine assembles the selected fragments into a synthesized answer. Citations attach to the source of the winning passage. A brand cited in three of eight fragments appears as a primary recommendation. A brand absent from all eight is invisible, regardless of domain authority, traffic, or brand recognition.
Every engine rewrites differently
ChatGPT strips constraints and generalizes. Perplexity preserves keyword form. Copilot compresses to 6 words. Claude applies an honesty filter. Google AI Mode expands with structural hierarchy. The same head query produces a different subquery space on each engine. That means one content strategy does not cover all surfaces.
Citabld maps the full space
We run the Citabld Decomposition Calculus (CDC) to derive the complete subquery space for your category, conditioned on your buyer personas, across all nine engines. Then we produce the extraction-engineered content that wins the selection events that matter most.
The subquery space is not universal.
A "best X software" query in fintech generates compliance-gated subqueries. The same query structure in logistics generates geography-anchored and integration-specific subqueries. The same head query in EdTech forks into completely different paths based on whether the buyer is an IT director or an individual teacher. Industry shapes the entire decomposition.
Price constraints mutate; compliance constraints almost never do. Industry determines which constraints survive engine rewrites.
Some industries have one buyer type. Others have four. Each persona produces a different subquery tree from the same head query.
High-stakes industries trigger EVIDENTIALIZE subqueries automatically. The engine asks for proof before the buyer does.
Perplexity dominates fintech research. Google AI Mode dominates local and retail. The same brand needs different coverage on different surfaces.
B2B SaaS comparison subqueries are heavily generated. Service businesses generate fewer comparisons and more evidence-seeking rows.
Real estate and logistics queries almost always retain a geographic constraint through rewrites. Most other categories drop it.
Select your category.
Modeled subquery rows for each industry. Every row is MODELED, derived from the Citabld Decomposition Calculus, not captured from live engine logs. The paid library adds measured rows where available.
B2B SaaS
The category every AI engine knows best and defaults to least honestly.
Buyers evaluate 4 to 7 tools using AI before contacting sales. The engine runs comparison, alternatives, and fit-checking subqueries simultaneously. A brand absent from the comparison layer loses the shortlist before the demo.
best crm for mid-market sales teams
salesforce alternatives under $100 per user
hubspot vs pipedrive for 50-person team
crm with built-in revenue forecasting 2026
what crm do series b startups use
crm that integrates with slack and notion
is salesforce worth it for small business
crm migration downtime risk
These rows are MODELED. They are derived from the Citabld Decomposition Calculus, not captured from live query logs. Google's surfaces expose no query data to third parties. The paid subquery library adds MEASURED rows from instrumented surfaces where available.
See the modeled fan-out for your head query in 90 seconds.
Free. No account required. We derive the subquery space from the Citabld Decomposition Calculus and produce one sample content piece for a fragment of your choice.