If your keyword spreadsheet assigns one intent label per term, you already know the pain: the SERP shows comparison tables and pricing pages while your brief says "informational." That misalignment is not a one-off mistake. It is what happens when binary labels meet mixed-intent SERPs saturated with redundant sameness.
Fractional search intent classification fixes this by assigning weighted percentages across multiple intent types for a single query. A keyword like "best CRM for startups" might score 45% commercial, 35% informational, and 20% generative, because the SERP, the People Also Ask block, and AI fan-out sub-queries each pull in a different slice of what the searcher needs.
Research from SE Ranking shows that while roughly 70% of traditional Google searches carry informational intent at the population level, individual high-value keywords routinely blend commercial investigation, comparison, and task-generation signals on the same results page. Binary labels cannot represent that mix, and AI search makes the problem worse.
TL;DR
- Fractional search intent uses weighted vectors (summing to 100%) instead of one checkbox per keyword.
- The FSIF model adds twelve sub-intent modifiers on top of six primary vectors for 2026 AI search.
- Query fan-out in Google AI Mode decomposes one prompt into multiple sub-intents you must score separately.
- Binary briefs misalign content when mixed-intent SERPs dominate your cluster.
- Install FSIF scoring at brief stage before drafting, same gate Metaflow uses internally.
What fractional search intent classification means (and why binary labels fail)
If you manage keyword clusters for a B2B SaaS brand, you have seen this failure mode: your spreadsheet says "informational," but the SERP shows product carousels, comparison tables, and pricing pages in the top five. The label was wrong. Not completely wrong, just incomplete.
Search intent classification is the process of inferring why someone typed a query. For twenty years, SEO teams used four buckets: informational, navigational, commercial, and transactional. Tools added local intent. SE Ranking's 2026 guide adds a sixth, generative AI intent, because ChatGPT users now ask tools to produce outputs, not just retrieve links.
That sixth type is necessary. It is still a single label per keyword. And single labels break on mixed-intent SERPs.
The four-type model is a ceiling, not a floor
The classic model was built for search engines routing queries to index sections. Content Harmony's Kane Jamison argued early that marketers need eight intent types (Research, Answers, Local, Transactional, Video, Visual, Fresh/News, and Brand/Entity) because "what Google wants to show" differs from "what the engine thinks the user typed."
Both systems assign one primary type. Fractional search intent classification rejects that constraint. Each query gets a vector.
Mixed-intent SERPs are the norm, not the exception
Take "project management software." The SERP might include:
- A listicle comparing tools (commercial)
- A definition snippet from a vendor blog (informational)
- Pricing pages with buy CTAs (transactional)
- A YouTube walkthrough (visual sub-intent)
Assigning 100% commercial misses the informational PAA questions your page must answer to earn the snippet. Assigning 100% informational misses the comparison table AI Overviews extract verbatim.
Why AI search makes single labels worse
Google AI Mode uses query fan-out, decomposing one prompt into multiple sub-queries before synthesizing an answer. Search Engine Land's fan-out guide notes that a single AI Mode session can branch into dozens of related intents. Your one keyword now carries fractional weight across each branch.
That is the core problem fractional search intent solves: one parent query, many sub-intents, none of them deserving a monopoly label.
The FSIF taxonomy: six primary vectors plus sub-intent modifiers
Metaflow's Fractional Sub-Intent Framework (FSIF) is a sub-intent taxonomy built for 2026 AI search. It has two layers: primary intent vectors (weighted percentages) and sub-intent modifiers (tags with confidence scores).
Primary intent vectors
| Vector | Code | Definition | SERP signals |
|---|---|---|---|
| Informational | I | User wants to learn or understand | PAA boxes, featured snippets, how-to guides |
| Navigational | N | User wants a specific site or page | Brand SERP, sitelinks, login pages |
| Commercial | C | User is comparing before buying | Best/top lists, vs pages, review roundups |
| Transactional | T | User is ready to purchase or sign up | Product pages, pricing, buy/order modifiers |
| Local | L | User wants nearby results | Map pack, near me modifiers, GBP listings |
| Generative | G | User wants AI to produce an output | Prompt-style queries, create/generate/draft modifiers |
Weights across I, N, C, T, L, and G must sum to 100%. A query can legitimately score C:50 / I:30 / G:20. No vector above 80% without at least two sub-intent modifier tags. That rule catches false certainty.
Sub-intent modifier layer
Modifiers describe format and freshness signals the primary vector misses:
| Modifier | Trigger signals | Content implication |
|---|---|---|
| Comparison | vs, best, top, alternative | Side-by-side table required |
| Definition | what is, meaning, definition | BLUF definition in opening 60 words |
| How-to | how to, steps, tutorial | Numbered procedure with H3 steps |
| Pricing | cost, price, pricing, free | Transparent pricing section or table |
| Freshness | 2026, latest, update, news | Date-stamped evidence, recent stats |
| Visual | diagram, chart, infographic | Structured table or visual asset |
| Tool-seeking | tool, software, platform, app | Product category framing |
| Problem-aware | problems, issues, mistakes, wrong | Pain acknowledgment in opening |
| Entity | brand name, product name | Brand-neutral or brand-specific page choice |
| Regulatory | compliance, legal, GDPR, HIPAA | Cited authoritative sources |
| Integration | integrate, connect, API, MCP | Workflow or stack diagram |
| Community | reddit, reviews, opinions | Third-party mention strategy |
A keyword carries one vector row plus zero to four modifiers. Modifiers do not need to sum to 100%. They are tags, not weights.
How vectors and modifiers combine
Example: "best marketing automation for ecommerce"
| Layer | Score |
|---|---|
| C (Commercial) | 55% |
| I (Informational) | 25% |
| G (Generative) | 20% |
| Modifiers | Comparison, Tool-seeking, Freshness |
Your brief now specifies: comparison table (mandatory), definition of marketing automation (opening BLUF), and a generative-ready summary block AI systems can extract. That is fractional search intent classification in practice, not a checkbox that says "commercial."
How to score fractional search intent in four steps
SEO leads and content strategists can run this workflow in roughly twenty minutes per keyword cluster.
Step 1: SERP format audit
Open an incognito window. Search the target keyword. Record:
- Top 5 URL types (blog, product, category, video, forum)
- SERP features present (snippet, PAA, carousel, map, AI Overview)
- Dominant modifiers in titles (best, vs, how, price)
This is your evidence base. Do not trust tool-default labels.
Step 2: Assign primary vector weights
Distribute 100% across I, N, C, T, L, G based on SERP composition:
- 3+ comparison listicles in top 5: C gets at least 40%
- Featured snippet + PAA dominance: I gets at least 30%
- Product/pricing pages in top 3: T gets at least 25%
- Map pack present: L gets at least 20%
- Query reads like a prompt ("write a…", "create a…"): G gets at least 25%
Round to five-point increments. Document your reasoning in the brief.
Step 3: Tag sub-intent modifiers
Apply modifiers from the table above. Minimum two modifiers for any keyword where the top vector exceeds 60%. This prevents overconfident single-intent briefs.
Step 4: Map AI fan-out sub-queries
Run the same keyword in Google AI Mode or ChatGPT. List sub-questions the system generates. Assign fractional weights to each sub-query. If fan-out produces three sub-intents at 40/35/25, your page outline needs three H2 sections, not one generic guide.
This step connects fractional search intent to what AI search visibility means for growth teams: visibility follows sub-intent coverage, not parent-keyword exact match.
Binary vs fractional: what changes in your content brief
Most teams still use a single intent dropdown. Here is what shifts when you adopt FSIF scoring.
| Brief field | Binary workflow | FSIF fractional workflow |
|---|---|---|
| Intent label | One type selected | Six-vector weight row summing to 100% |
| Sub-intent | Optional notes field | Required modifier tags (min 2 if top vector >60%) |
| Outline driver | Primary intent only | Highest-weight vector + all modifiers |
| PAA coverage | Ad hoc | Intent map table linking each PAA to an H2 |
| AI fan-out | Not considered | Sub-query list with fractional weights |
| Ship gate | Subjective editor call | Threshold rules (no vector >80% without modifiers) |
The intent map table ties searcher needs to your outline:
| Searcher need | FSIF signal | Where we answer it |
|---|---|---|
| What is marketing automation? | I + Definition | Opening BLUF + FAQ |
| Which tools compare best? | C + Comparison | H2 with comparison table |
| Can AI draft my workflow? | G + How-to | Generative-ready summary block |
| What changed in 2026? | Freshness modifier | Evidence paragraph with dated stat |
This map is how you close PAA gaps without FAQ filler. It also feeds the information gain content framework. If your fractional scores show heavy informational weight but your draft lacks a named framework or original data, the IG-9 gate blocks ship.
Fractional search intent in AI surfaces: fan-out and generative weight
Platform divergence is extreme. Fractional weights shift by surface, and your taxonomy must account for that.
Profound's ChatGPT intent study, cited by SE Ranking, found generative task intent at 37.5% of ChatGPT queries. Users ask for concrete outputs like "create X" or "draft Y." Traditional Google search remains roughly 70% informational at the population level, per SE Ranking's 2026 statistics roundup.
That gap matters. A keyword that scores G:20 on Google might score G:45 on ChatGPT for the same audience. Fractional search intent classification is surface-aware.
| Surface | Typical top vector | Sub-intent emphasis | Fan-out depth |
|---|---|---|---|
| Google organic | I or C | Comparison, Definition | Low (single SERP) |
| Google AI Mode | I + G blend | How-to, Freshness | High (multi sub-query) |
| ChatGPT | G | Tool-seeking, How-to | Medium (prompt decomposition) |
| Perplexity | I + C blend | Citation, Comparison | High (source aggregation) |
For AI surface optimization, see AEO GEO LLMO best practices for 2026. The short version: score fractional intent per surface, not once globally.
Query fan-out means your parent keyword is a container. Sub-intents inside it compete independently for citation. A page that nails the 55% commercial vector but ignores the 25% informational sub-queries loses AI Overview inclusion, even if it ranks on Google organic.
Installing FSIF scoring in your editorial pipeline
Fractional classification only works if it gates briefs, not if it lives in a strategy deck.
Brief stage
Every keyword in the cluster gets an FSIF vector row and modifier tags before outline approval. Reject briefs where one vector exceeds 80% without documented SERP evidence. Metaflow's batch-03 pipeline, including this post, runs that check in `publish-from-files.mjs` before any draft ships.
Draft stage
Outline H2 count must match the number of non-trivial vector weights above 15%. Three vectors above 15% means at least three distinct H2 sections addressing each slice. Modifiers map to mandatory assets: Comparison means table, How-to means numbered steps, Freshness means dated stat in opening.
Pre-publish stage
Connect FSIF to the content engineering non-commodity framework N3 Stack. Evidence layer: SERP audit documented. Framework layer: FSIF vector row present. Systems layer: intent map table links PAA to H2s. All three must pass.
The outcome your team needs: fewer pages that rank for the wrong slice of intent, and more briefs that survive AI fan-out decomposition because sub-intents were scored upfront, not discovered after a failed draft.
Explore implementation patterns in the Metaflow learning center.
Frequently Asked Questions
What is fractional search intent?
Fractional search intent is a classification method that assigns percentage weights across multiple intent types for a single query instead of one exclusive label. For example, a keyword might score 50% commercial, 30% informational, and 20% generative. Weights sum to 100% and reflect what mixed-intent SERPs and AI fan-out actually show, not what a legacy four-type model assumes.
What is search intent classification?
Search intent classification is the process of determining why a user submitted a query and what outcome they expect. Traditional SEO used binary or single-label systems (informational, navigational, commercial, transactional). Modern classification for 2026 AI search requires fractional scoring because one query can trigger multiple sub-intents across Google, ChatGPT, and Perplexity simultaneously.
How do you classify search intent for AI search?
Run four steps: SERP format audit, assign FSIF primary vector weights summing to 100%, tag sub-intent modifiers from the twelve-modifier taxonomy, and map AI fan-out sub-queries with their own fractional weights. Repeat per surface. ChatGPT often carries higher generative (G) weight than Google organic for the same topic. Document scores in the brief before outlining.
What are sub-intents in SEO?
Sub-intents are secondary intent signals within a parent query that binary classification misses. Examples include comparison (vs, best), definition (what is), freshness (2026, latest), and tool-seeking (software, platform). The FSIF model treats sub-intents as modifier tags layered on top of primary vector weights, so a commercial-dominant query still flags when a definition or comparison sub-intent must be addressed in the outline.
Is search intent still binary in 2026?
No, and it was never truly binary for high-value keywords. Single-label intent was a spreadsheet convenience. Mixed-intent SERPs, AI Overviews, and query fan-out expose the gap daily. Fractional search intent classification replaces one checkbox with a weighted vector row plus modifier tags. Teams still shipping binary labels will misalign content on any keyword where the SERP blends formats.
How does query fan-out affect search intent?
Query fan-out decomposes one AI search prompt into multiple sub-queries, each with its own intent profile. A parent keyword might fan out into a definition sub-query (informational), a comparison sub-query (commercial), and a template-generation sub-query (generative). Fractional weights apply to each branch. Pages that address only the parent keyword's dominant label miss the sub-intents AI systems retrieve against, reducing citation odds in AI Overviews and ChatGPT.
