The Search Everywhere Canvas

The canvas

The Canvas is intentionally question led. It should not begin with "we need 500 pages" or "we need to appear in ChatGPT." Those are proposed outputs or channels.

The Canvas first clarifies the user need and the business reason for solving it.

1. Business Outcome

Define one primary outcome in terms the organization already understands.

Useful outcomes include:

• Help investors understand a financial concept before making a product decision
• Reduce support contacts caused by incompatible product purchases
• Increase completion of a public service application
• Help employees find the current policy without contacting operations
• Improve qualified discovery for a complex B2B product

The outcome should be observable. "Improve AI visibility" is too narrow and too vague. Visibility may support the outcome, but it is not usually the business result.

2. User Questions

List the questions that lead to the outcome.

Use language from search data, support conversations, product research, sales calls, internal search logs, and subject matter experts. Include follow up questions and ambiguous wording.

Then group the questions by the type of answer they require.

Questions that look similar may require different sources and integrity rules. The Canvas makes those differences visible before implementation.

3. Priority Entities

Identify what the questions are about.

Entities may include companies, securities, products, models, people, locations, policies, events, metrics, or concepts.

For each entity type, record:

• Stable identifier
• Common names and aliases
• Parent or category relationship
• Important neighboring entities
• Known ambiguity
• Canonical presentation

This step prevents a common mistake: designing pages before resolving what the system must distinguish.

4. Knowledge Objects

Define the reusable units required to answer the questions.

Do not create one object for every page. Create objects around differences in source, ownership, time, and validity.

For a company research experience, the Canvas may require:

• Company identity object
• Tradable security object
• Current market snapshot
• Historical price observation
• Reported earnings event
• Analyst estimate
• Financial metric definition
• News or event object

A page can combine several objects. The object model should not simply mirror the navigation tree.

5. Integrity Rules

For each important object, define the minimum conditions for reuse.

This section is where the Canvas becomes an engineering and product tool rather than a content brief.

6. Presentations

Only after the knowledge design is clear should the team choose outputs.

Possible presentations include:

• Entity landing page
• Comparison table
• Definition or educational article
• Application component
• Search result snippet
• Structured data representation
• API response
• Internal assistant answer
• Email or notification

For each presentation, define what it adds beyond exposing the object. A page may provide explanation and conversion. An application component may support a decision. An API may enable another product. A comparison table may normalize several objects.

Do not create a presentation only because a channel exists.

7. Discovery Surfaces

Record where the audience is likely to encounter the information.

This may include traditional search, AI search, social discovery, internal search, marketplace search, browser agents, application navigation, or direct API access.

Then record the requirements and limits of each surface.

The framework does not assume that one optimization works identically across every system. It aims to preserve the same reliable knowledge while adapting delivery to the surface.

8. Retrieval Risks

Use the Retrieval Pipeline to identify likely failure points before launch.

Ask:

• Can the source be discovered?
• Why should it be selected for this question?
• What could be misinterpreted?
• Which qualification could disappear during summarization?
• Which page should receive attribution?
• What happens when the object is stale or unavailable?

This creates a practical risk register tied to the information system.

9. Evaluation Set

Turn the user questions into a stable test set.

For each query, record:

• Expected entity
• Minimum correct answer
• Required time context
• Preferred source class
• Required qualification
• Surfaces to test
• Review cadence

Include negative tests. Some questions should return "not available," direct the user to a professional, or avoid combining incompatible values.

10. Success Measures

Select a small set of measures across four categories.

Avoid building a composite score until the individual measures are understood. A single number can hide the difference between broad reach and reliable answers.

11. Owners

Assign ownership at two levels.

Object owner: accountable for source authority, definitions, freshness, and correction.

Experience owner: accountable for how the object is presented and whether the user can complete the task.

Other roles may own implementation, analytics, compliance review, or platform operations. One role should still be accountable for the final decision.

12. Review Cadence

Set separate cadences for operational quality and discovery learning.

• Real time or daily for high risk data failures
• Weekly for material source and presentation issues
• Monthly for controlled discovery evaluation
• Quarterly for framework scope, ownership, and object design

The Canvas should end with named decisions, not a list of open themes.

A completed example

The following fictional example shows how a team might use the Canvas for a product compatibility problem.

The implementation may produce pages and content, but the Canvas reveals that the central problem is a compatibility knowledge model with clear ownership and reuse.

How to run a Canvas workshop

1. Invite decision makers. Include product, engineering, data, search, content, analytics, and a subject matter or risk owner.
2. Choose one discovery problem. Do not map the entire company in the first session.
3. Complete boxes in order. The sequence prevents the team from jumping directly to pages or tools.
4. Mark unknowns. An honest unknown is more useful than an invented answer.
5. Select three decisions. End with a scoped object, an owner, and an evaluation date.
6. Version the Canvas. Record changes as sources, products, and discovery surfaces evolve.

A first session should produce a testable slice. The Canvas is not an invitation to redesign the entire information architecture before shipping value.

What a completed Canvas should produce

• A prioritized user question set
• A defined group of Knowledge Objects
• Integrity requirements for the highest value objects
• A map of controlled presentations and external discovery surfaces
• A retrieval risk list
• An evaluation set with expected facts
• Named owners
• A review date and a short implementation backlog

The filled worksheet is not the final deliverable. The decisions and operating plan are.

Common Canvas mistakes