Discovery
Can the brand be matched to category-level workspace, knowledge and project-management needs?
Independent case study · Notion public-site baseline
SEO/AEO/GEO Optimization Case Study
This independent study uses Notion as a real B2B SaaS research subject. It is not client work and is not affiliated with Notion. The goal is not to award a single visibility score, but to test how well public information supports discovery, retrieval, comparison and answer generation across different search intents.
At a glance / Evidence status
Problem
How can public SaaS content better support discovery, accurate answers and comparisons in search and AI systems?
Method & tools
Public-source audit, 12 versioned prompts, intent mapping, entity analysis, intervention register and a Python analysis layer.
Deliverables
Baseline framework, four documented redesign examples and a controlled measurement plan.
Verified status
Research design and proposals documented. No deployed Notion changes, measured AI visibility gains or client results.
Research question
Which page, entity and evidence signals make a brand easier to retrieve, summarize, cite and compare, and where does a conventional SEO audit fail to capture those signals?
Baseline v1
The public site provides a useful mix of evidence: workspace positioning, project-management capabilities, AI agents, enterprise search, connector coverage and structured pricing. Those signals can be tested separately instead of being collapsed into a vague claim that a page is "AI optimized".
Comparison
Is there enough explicit evidence to compare capabilities and trade-offs without filling gaps by inference?
Pricing & value
Are plan boundaries, included capabilities and AI packaging clear enough to answer buyer questions accurately?
AI capabilities
Can agents, enterprise search and connected-app capabilities be retrieved with precise scope and terminology?
Audit framework
The audit separates conventional search signals from answer-engine and generative-search requirements. GEO is not treated as a relabelled SEO checklist.
1 · Intent & query fan-out
Map the original buyer question to related sub-questions: category, capability, integration, price, implementation and comparison.
2 · Entity clarity
Check whether product, feature, audience and integration entities are named consistently enough to resolve without relying on brand familiarity.
3 · Answerability
Check whether important questions have direct, self-contained answers close to the supporting evidence.
4 · Citability
Look for specific claims, definitions and constraints that can support an answer. Clear evidence matters more than keyword repetition.
5 · Comparison readiness
Test whether capabilities, limits, pricing and integrations can be compared on equivalent dimensions without guessing.
6 · Evidence freshness
Date volatile evidence such as pricing, AI packaging and beta features to reduce stale-answer risk.
From diagnosis to optimization
The baseline produced six intervention candidates. Four are high-priority: clarify category boundaries, make capabilities comparison-ready, separate AI capability definitions, and connect pricing evidence to the questions buyers actually ask.
Example 01 · Entity clarity
Define the category without flattening the product
Retrieval problem: Notion can be described as a workspace, knowledge tool, project-management system, enterprise-search product and AI platform. A discovery answer may have to assemble that definition from several contexts.
Proposed pattern: lead with one direct definition, then name the major jobs and capability entities in a compact, self-contained passage.
Test: D01, D02, C01 and C02.
Example 02 · Capability boundaries
Keep search and agentic action distinct
Retrieval problem: Agent, Enterprise Search and connectors are related but not interchangeable. An answer can overgeneralize what is searched or what an agent can do.
Proposed pattern: answer "What does Enterprise Search search?" separately from "How is that different from an AI agent?", with permission and source boundaries stated explicitly.
Test: U01, U02 and I01.
Example 03 · Comparison readiness
Expose equivalent buyer dimensions
Retrieval problem: broad product claims are difficult to compare when evidence for project work, knowledge, search, integrations and AI is scattered.
Proposed pattern: structure evidence around buyer questions: projects/tasks, team knowledge, connected search, AI functions and plan inclusion.
Test: C01, C02 and R01.
Example 04 · Pricing & value
Answer the decision, not only the seat price
Retrieval problem: a pricing table cannot by itself answer what a 20-person team should choose. Required capabilities, AI packaging and billing assumptions matter.
Proposed pattern: start with the required capabilities, verify the current plan that includes them, state billing assumptions and date volatile packaging.
Test: P01, P02 and R01.
Why these examples are not "results"
The proposed patterns are original redesign examples based on the public baseline; they are not changes made by Notion and they are not presented as verbatim Notion copy. They demonstrate optimization design. Performance can only be discussed after a controlled observation cycle.
Controlled prompt dataset
Baseline v1 contains 12 frozen prompts across category discovery, alternatives, comparisons, capabilities, pricing/value, implementation and recommendation intent. Prompt definitions are versioned before observations are collected, reducing the temptation to rewrite the test after seeing an answer.
Measurement model
Observed public evidence, analysis, proposed intervention and post-intervention measurement remain separate throughout the study. AI visibility is decomposed rather than treated as a ranking: mention coverage, citation coverage, recommendation inclusion, source patterns and consistency are measured independently.
What this case study will not claim
No traffic, ranking, citation, conversion or revenue improvement will be attributed to an intervention unless it is actually observed. Correlation will not be presented as causation, and missing evidence will be recorded as missing rather than converted into a negative product claim.
Current status
The public baseline, controlled prompts, intervention register, optimization examples and Python analysis layer are versioned. Real AI observations and post-intervention results remain unpublished until they can be collected reproducibly and audited.