Document 03

Volumetric SEO Engine

Combinatorial opportunity and pruning engine

1. Definition

The Volumetric SEO Engine expands an opportunity across independent dimensions, then compresses that opportunity space into a small set of high-value canonical assets.

The engine is a generator + constraint solver + portfolio allocator.

VOLUME ≠ PAGE COUNT

VOLUME
=
number of meaningful combinations inspected
× number of evidence sources
× number of viable solution forms
× number of measurable learning cycles

INPUTS
→ DIMENSIONAL EXPANSION
→ NORMALIZATION
→ INTENT CLUSTERING
→ BOOLEAN GATES
→ REDUNDANCY COMPRESSION
→ VALUE SCORING
→ PORTFOLIO SELECTION
→ ASSET PRODUCTION
→ OBSERVATION
→ MODEL UPDATE

2. Opportunity dimensions

2.1 Business dimensions

offer
revenue model
margin
capacity
sales cycle
qualification threshold
retention value
risk
strategic priority

2.2 Audience dimensions

segment
role
industry
organization size
experience
problem awareness
solution awareness
urgency
budget
geography
language
device/context
accessibility need

2.3 Task dimensions

define
learn
discover
diagnose
compare
calculate
locate
verify
plan
implement
troubleshoot
buy
use
maintain
renew
refer

2.4 Intent dimensions

informational
comparative
commercial investigation
transactional
navigational
local
support
retention
reputation

2.5 Semantic dimensions

parent concept
subtype
synonym
entity
attribute
attribute value
process
prerequisite
problem
cause
consequence
solution
alternative
comparison
objection
risk
evidence
question

2.6 Modifier dimensions

who
what
where
when
why
how
best
cost
price
cheap/premium
near me
reviews
versus
alternative
for [audience]
with [attribute]
without [constraint]
under/over [value]
before/after
current/year
beginner/advanced

2.7 Asset dimensions

service page
product page
category
guide
comparison
calculator
diagnostic
template
checklist
directory
location page
case study
dataset
benchmark
glossary
FAQ
video
image series
interactive
API/data feed

2.8 Evidence dimensions

firsthand experience
original data
controlled test
field observation
customer evidence
expert review
documented case study
primary-source citation
methodology
screenshots
demonstration
limitations

2.9 Search-surface dimensions

web results
images
video
local
shopping/product
news
Discover
generative AI features
site search
third-party marketplace

2.10 Journey and conversion dimensions

awareness → related content
evaluation → tool/comparison
decision → assessment/demo/quote
transaction → purchase/signup
use → successful implementation
retention → renewal/expansion
advocacy → review/referral/citation

2.11 Time dimensions

evergreen
seasonal
event-driven
newly changed
periodic
decaying
historical
real-time
forecast

2.12 Measurement dimensions

query
page
cluster
country
device
search appearance
audience
offer
change ID
date
cohort
value event

3. The theoretical opportunity volume

A modest generator with:

creates:

This is not a publishing target. It demonstrates why SEO needs a grammar and constraint system. Without compression, “programmatic SEO” easily becomes duplication, thin content, maintenance debt, and measurement noise.

The desired outcome might be only 20, 100, or 1,000 canonical assets depending on the business—even though the system evaluated a much larger possibility space.

8 audience states
× 12 tasks
× 8 intents
× 12 semantic node types
× 10 modifier families
× 12 asset formats
× 6 search surfaces
× 8 journey states
× 8 evidence modes
× 6 geographies
× 6 next actions

15,288,238,080 theoretical combinations

4. Candidate generation

4.1 Candidate sentence

CANDIDATE
=
AUDIENCE
× CONDITION
× TASK
× INTENT
× CONCEPT
× MODIFIER_SET
× GEOGRAPHY
× TIME_STATE
× ASSET_FORM
× EVIDENCE_MODE
× NEXT_ACTION

4.2 Generation pseudocode

FOR EACH approved offer:
  FOR EACH plausible audience:
    FOR EACH observed problem or desire:
      FOR EACH task:
        FOR EACH semantic relation:
          GENERATE query hypotheses
          ATTACH possible formats
          ATTACH possible evidence
          ATTACH next actions
          ATTACH measurement plan

4.3 Evidence enrichment

Each hypothesis must be enriched with observations:

The more sources agree, the higher the demand confidence.

first-party query data
OR customer language
OR result-page evidence
OR competitor visibility
OR sales/support evidence
OR market data
OR paid-search evidence

5. Candidate normalization

Before clustering:

Do not erase meaningful distinctions. “Audit” as a noun, “audit software,” and “hire an auditor” may belong to different tasks.

lowercase for comparison
normalize punctuation
normalize singular/plural where meaning is stable
expand or resolve abbreviations
map synonyms to concepts
identify named entities
identify geography and time
identify modifiers
detect language
detect likely sense
extract task verb
extract transaction state
attach exclusions

6. Intent equivalence test

Calculate a conceptual equivalence vector:

EQUIVALENCE =
sense_match
+ task_match
+ result_type_match
+ answer_structure_match
+ audience_state_match
+ next_action_match
+ current_SERP_overlap

Merge gate

IF sense_match = true
AND task_match = true
AND answer_structure_compatible = true
AND next_action_compatible = true
AND separate_page_value = low
THEN assign to same cluster.

Split gate

IF sense differs
OR task differs
OR result class differs
OR local/product inventory differs
OR legal context differs
OR combined answer harms usability
THEN create separate cluster candidate.

Test gate

IF evidence is ambiguous
THEN:
  keep one provisional cluster
  AND observe query/page behavior
  OR run a controlled asset-format test
  BEFORE multiplying pages.

7. Boolean publication gates

A candidate proceeds only if it passes all hard gates.

APPROVE
=
DEMAND
AND DISTINCTNESS
AND BUSINESS VALUE
AND INFORMATION ADVANTAGE
AND TECHNICAL FEASIBILITY
AND MAINTAINABILITY
AND MEASURABILITY
AND POLICY / ETHICAL COMPLIANCE

Demand gate

PASS IF:
observed impressions
OR customer evidence
OR sales/support frequency
OR stable result ecosystem
OR defensible emerging-demand thesis

Distinctness gate

PASS IF:
the candidate owns a distinct task
OR needs a materially different answer
OR has real local/product data
OR requires a different conversion path

Value gate

PASS IF:
direct conversion value
OR assisted conversion value
OR retention value
OR authority value
OR strategic learning value

Information-advantage gate

PASS IF at least one:
original experience
OR proprietary data
OR better methodology
OR better utility
OR stronger evidence
OR clearer decision support
OR unique local coverage
OR better accessibility

Maintainability gate

PASS IF:
owner exists
AND refresh triggers are defined
AND data can be kept accurate
AND total approved volume fits capacity

Measurement gate

PASS IF:
primary outcome
AND baseline
AND change ID
AND observation method
AND decision rule
are defined.

8. Soft scoring model

Candidates that pass the gates are prioritized.

Use a 0–5 rating or calibrated probability for each factor.

Avoid zero in a multiplicative score by using a minimum floor such as 0.2 for uncertain but nonzero factors, or use a weighted log model.

VALUE NUMERATOR =
DemandConfidence
× BusinessFit
× TaskValue
× InformationAdvantage
× ConversionValue
× Attainability
× ReusePotential
× LearningValue

COST DENOMINATOR =
ProductionEffort
× TechnicalRisk
× MaintenanceBurden
× TimeToLearning
× OpportunityCost

PRIORITY SCORE =
VALUE NUMERATOR / COST DENOMINATOR

Confidence adjustment

ADJUSTED SCORE =
PRIORITY SCORE
× EvidenceConfidence
× MeasurementConfidence

Existing-asset multiplier

IF existing URL has:
impressions
OR links
OR conversions
OR stable indexing
OR strong internal position
THEN multiply by ExistingSignalFactor.

9. Redundancy compression

The goal is maximum demand coverage with minimum canonical assets.

Compression steps

1. Group lexical variants.

2. Merge same-sense queries.

3. Group same-task queries.

4. inspect result-format compatibility.

5. assign one canonical owner.

6. place remaining questions as:

- sections;

- FAQs;

- supporting assets;

- exclusions;

- future experiments.

Group lexical variants.
Merge same-sense queries.
Group same-task queries.
inspect result-format compatibility.
assign one canonical owner.
place remaining questions as:

Compression ratio

A high ratio can be healthy when it reflects disciplined clustering. It is unhealthy if valid distinct tasks are being forced into bloated pages.

COMPRESSION RATIO =
generated candidates / approved canonical assets

Coverage ratio

Optimize both compression and coverage—not one alone.

VALUABLE DEMAND COVERAGE =
value-weighted approved cluster demand
/
value-weighted validated demand

10. Page-volume governance

Capacity formula

SUSTAINABLE PAGE VOLUME
≤
(editorial capacity × quality throughput × refresh capacity)
/
(average maintenance burden × volatility)

Scale rules

IF approved_page_count exceeds maintenance capacity
THEN:
  raise value threshold
  OR consolidate clusters
  OR automate data quality
  OR reduce geography/product scope
  OR stage publication.

IF a template produces mostly repeated text
THEN the template is not a sufficient information product.

IF local pages lack unique local proof
THEN consolidate to a broader service-area asset
OR collect valid local data first.

IF generated assets cannot be individually measured
THEN use representative cohorts
AND prevent uncontrolled expansion.

IF pages decay faster than they can be refreshed
THEN reduce scope
OR change the asset class.

11. Volumetric content architecture

11.1 Parent–subtype volume

Example:

PARENT
→ SUBTYPE
→ ATTRIBUTE
→ VALUE
→ TASK

WEB DESIGN
→ ECOMMERCE WEB DESIGN
→ PLATFORM
→ SHOPIFY
→ COMPARE / HIRE / PRICE / MIGRATE

11.2 Entity–relation volume

Examples:

ENTITY A
→ RELATION
→ ENTITY B

WORDPRESS → COMPARED_WITH → WEBFLOW
WCAG → APPLIES_TO → ECOMMERCE
CALGARY → CONTAINS → SERVICE AREAS
PAGE SPEED → AFFECTS → CONVERSION

11.3 Problem–solution volume

AUDIENCE
→ PROBLEM
→ CAUSE
→ DIAGNOSTIC
→ SOLUTION
→ PROOF
→ OFFER

11.4 Journey volume

One cluster system should cover the journey without forcing a linear user path.

PROBLEM AWARENESS
→ DIAGNOSIS
→ OPTIONS
→ COMPARISON
→ COST
→ PROOF
→ SELECTION
→ IMPLEMENTATION
→ MAINTENANCE

11.5 Format volume

For each approved task, test alternate information products:

The best format is the one that completes the task and creates measurable value—not the one easiest to publish.

article
OR table
OR video
OR calculator
OR template
OR map
OR dataset
OR interactive

12. LAKA volumetric expansion

For every approved opportunity, generate five change classes.

Then apply the ten internal variables to each level:

Then attach the fourteen change descriptors:

This is a volumetric analysis space, not a requirement to execute 700 actions.

5 change levels × 10 internal variables = 50-cell intervention map

50 cells × 14 descriptors = 700 analytical observations per opportunity

Level, Expansion question, Typical candidates — 5 rows
Level	Expansion question	Typical candidates
Baseline	What is happening now?	measurement, inventory, diagnosis
Minor	What small reversible change can improve response?	title, intro, CTA, link
Major	What richer answer or format would serve the task?	rewrite, tool, video, evidence
Structural	What relationships or systems cause the limitation?	merge, graph, template, pipeline
Paradigm	What new solution class could redefine the result?	data product, diagnostic, agent, benchmark

13. Volumetric experiment design

Variation dimensions

query cluster
title proposition
search snippet
direct-answer style
information depth
format
evidence
visuals
internal links
next action
offer
technical implementation

Controlled generation

FOR EACH high-value cluster:
  GENERATE minor, major, structural, paradigm hypotheses
  SCORE by mechanism, value, risk, reversibility, learning
  SELECT a non-confounded test
  ASSIGN change_id
  OBSERVE primary metric + guardrails
  UPDATE priors

Rule against combinatorial confusion

IF multiple high-impact variables change simultaneously
AND the goal is causal learning
THEN split the change
OR explicitly classify it as a package test.

IF the goal is simply recovery
AND delay has high business cost
THEN a package intervention may be valid,
but causal attribution will be lower.

14. Feedback-driven generation

The engine updates its weights based on outcomes.

IF a task family produces qualified value repeatedly
THEN increase its business-fit prior.

IF a format improves task completion across clusters
THEN increase its format prior for similar tasks.

IF a query family generates traffic but poor qualification
THEN lower its value prior
AND inspect audience/intent classification.

IF generated local pages decay or duplicate
THEN tighten local distinctness gates.

IF original data earns relevant citations
THEN increase evidence and reuse scores for related assets.

IF a paradigm asset creates branded demand
THEN create supporting and conversion pathways around it.

15. Output types

The engine should generate these outputs, not merely pages:

opportunity inventory
semantic graph
intent clusters
canonical registry
page specifications
internal-link graph
technical requirements
evidence backlog
authority campaigns
experiment queue
measurement plan
refresh calendar
retirement/merge queue

16. Minimum volumetric workflow

This produces scale through disciplined thought, not indiscriminate publishing.

1. Generate 100–1,000 candidates internally.
2. Normalize and cluster them.
3. Apply eight hard publication gates.
4. Score survivors.
5. Select the top portfolio under capacity.
6. Assign one canonical owner per dominant intent.
7. generate five LAKA intervention levels.
8. execute the smallest intervention capable of affecting the diagnosed mechanism.
9. measure business and diagnostic outcomes.
10. update the generator.

This document as JSON: GET /v1/documents/volumetric-seo-engine · raw markdown

← All documents