From SEO to KEO: WebGraphs, .keg and the Knowledge Layer for AI Agents

The web is evolving from understanding information to acting on structured knowledge.

That is the idea behind WebGraphs: an open proposal for helping people and businesses describe what they are, what they know, what they offer and how those facts relate, in a format that machines can discover and verify.

This is not an established web standard. It is a working hypothesis, built from two connected observations:

1. The web is becoming increasingly autonomous.
2. Visibility is moving from pages, to answers, to knowledge, to decisions and eventually actions.

I explored the first observation in Web 6.0: The Autonomous Internet. This article explores the missing knowledge layer beneath it.

A speculative map of the web

The numbered versions of the web are not governed by one official roadmap. Different organisations use the labels differently, particularly beyond Web 3.0. The model below is therefore not a prediction dressed as certainty. It is a way of seeing the pattern.

Stage	Capability
Web 1.0	Read
Web 2.0	Read / Write
Web 3.0	Read / Write / Own
Web 4.0	Read / Write / Think
Web 5.0	Read / Write / Feel
Web 6.0	Read / Write / Own / Think / Feel / Act
Web 7.0	… / Create
Web 8.0	… / Control
Web 9.0	… / Optimise
Web X	Human + machine symbiosis

The important point is not whether the numbers prove correct. It is the direction of travel.

The web began as information people could read. It became participatory, transactional and increasingly intelligent. The next major shift is from systems that understand information to systems that use it to decide and act.

The visibility stack after SEO

Search engine optimisation was designed around pages. The page was the unit being crawled, indexed and ranked.

Answer engines changed the unit. A system could extract an answer without requiring the user to visit the page. Generative systems changed it again by synthesising information across multiple sources.

Discipline	Primary unit
SEO	Pages
AEO	Answers
GEO	Generated knowledge
KEO	Structured, attributable knowledge
DEO	Decision shaping
AAO	Agent selection
LAO	Agent execution

AEO and GEO are already useful labels. The later terms are proposals, not recognised disciplines. Their value is as a sequence:

First machines find information. Then they form answers. Then they make decisions. Eventually they take action.

If that sequence is even roughly correct, the next battleground is not merely ranking a page. It is making knowledge sufficiently clear, attributable, current and machine-readable to participate in a decision.

What is Knowledge Engine Optimisation?

Knowledge Engine Optimisation, or KEO, is the practice of structuring and publishing attributable knowledge so that search engines, generative systems and autonomous agents can interpret it with less ambiguity.

I originally described KEO as optimisation for “structured truth”. That is memorable, but technically too confident. A publisher can structure a claim. It cannot make the claim true by formatting it.

A credible knowledge system therefore needs to separate:

• the entity making a claim;
• the claim itself;
• the source supporting it;
• the date and version;
• the relationship to other entities;
• the method of verification;
• and the confidence or status attached to it.

KEO is not better copywriting for robots. It is knowledge architecture.

The missing layer

The foundations already exist.

Schema.org provides a shared vocabulary for describing entities. JSON-LD provides a W3C-standard way to express linked data in JSON. Sitemaps give crawlers a predictable way to discover pages.

These are powerful ingredients, but they solve different parts of the problem.

• Schema.org helps define meaning.
• JSON-LD helps serialise linked data.
• Sitemaps help discover URLs.
• Structured-data validators check syntax and selected conventions.

What is still missing is a simple, open publishing workflow that lets a non-developer create an entity graph, attach provenance, validate it, package it and expose it for machine discovery.

The WebGraphs proposal

WebGraphs.org would be an open protocol, visual builder, validator and public index for machine-readable knowledge graphs.

A user would create a radial graph resembling a mind map:

• nodes represent entities, claims, products, people, organisations, places or capabilities;
• branches represent relationships;
• sources and evidence can be attached to individual claims;
• the visual graph translates into standards-based JSON-LD;
• the published graph can be validated and listed in a public index.

The aim is not to replace Schema.org or JSON-LD. That would be pointless reinvention. The aim is to make them easier to create, package, verify and exchange.

Two meanings inside KEG

The acronym needs a clean distinction:

• Knowledge Entity Graph: the graph of entities, attributes, claims and relationships.
• Knowledge Exchange Graph: the distributable package containing that graph and the information required to interpret and verify it.

The proposed file extension for the exchange package is .keg.

What is a .keg file?

A .keg is a proposed compact, standardised container for publishing structured knowledge about entities and their relationships.

It would normally contain JSON-LD, but the value is not compression alone. Compression is easy. Trust, provenance, versioning and interoperability are the hard parts.

A sensible first version could be a ZIP-compatible container with a predictable structure:

webgraph.keg
├── manifest.json
├── graph.jsonld
├── sources.json
├── signatures/
├── proofs/
└── assets/

The package could provide:

• Compression: reducing the payload where graphs become large;
• Containment: packaging graph data and supporting evidence together;
• Integrity: hashes and signatures showing whether content has changed;
• Versioning: making updates and superseded claims explicit;
• Provenance: connecting claims to publishers and evidence;
• Distribution: providing one predictable machine-readable resource.

A website might expose the package at:

https://example.com/webgraph.keg

It could also advertise the file in the page header using a standard link relation, and potentially through a well-known discovery endpoint if the protocol matures.

Crucially, publishing a .keg file would not force Google, OpenAI or any other system to ingest it. Sitemaps do not guarantee indexing either. The goal is more defensible:

Lower the cost of discovering, interpreting and verifying useful knowledge.

Why a visual graph builder matters

Most business owners will never hand-author JSON-LD, RDF or graph queries. Nor should they need to.

The visual builder is not decorative. It is the adoption mechanism.

A bakery could describe:

• the organisation;
• its locations;
• products and ingredients;
• opening hours;
• delivery areas;
• certifications;
• ownership;
• relationships with suppliers;
• and the evidence behind important claims.

The owner sees a comprehensible map. The system emits structured data.

That gives smaller organisations a better chance of being understood by machines without requiring an internal semantic-web team. It does not guarantee selection by AI, but it reduces one structural disadvantage.

From KEO to decisions and agents

KEO is only the first layer.

Decision Engine Optimisation

DEO asks whether the published knowledge helps a system make a particular decision. An agent choosing a supplier may care about price, availability, distance, certification, delivery reliability and cancellation terms. Visibility becomes contextual rather than universal.

Autonomous Agent Optimisation

AAO asks whether an agent can identify an organisation or capability as a suitable participant in a task. This requires machine-readable constraints, permissions, policies and capabilities, not merely persuasive content.

Logic and Agent Optimisation

LAO concerns execution. Can the agent understand what action is available, what inputs it needs, what rules apply and what result to expect?

This is where knowledge graphs meet APIs, permissions, identity, payments and accountable automation.

The minimum viable standard

The wrong way to begin would be to spend years designing a perfect universal ontology in private.

The useful first version is smaller:

1. Publish a public WebGraphs v0.1 proposal.
2. Define a narrow .keg package structure using JSON-LD.
3. Build a visual graph editor for organisations, people, products, services and claims.
4. Create a validator that checks syntax, identity, provenance and broken relationships.
5. Publish a public index of opt-in graphs.
6. Build WordPress and Shopify exporters.
7. Demonstrate one independent agent consuming the graph to make a constrained decision.

The final step matters most. A standard without a working consumer is a beautifully labelled drawer.

What could kill the idea?

There are several serious risks:

• No consumer incentive. AI companies may prefer their own crawling and internal graph systems.
• Publisher gaming. A new optimisation discipline attracts manipulation almost immediately.
• False certainty. Structured claims can look authoritative while remaining wrong.
• Vocabulary sprawl. Reinventing concepts already covered by Schema.org would create friction rather than remove it.
• Governance. An open standard needs transparent stewardship, versioning and dispute processes.
• Cold start. Publishers will not create graphs without consumers; consumers will not integrate without graphs.

These are not reasons to abandon the idea. They define what must be solved.

The real opportunity

The original thought was a service resembling sitemap submission: instead of giving a search engine a list of pages, give intelligent systems a map of knowledge.

The stronger version is not another submission portal. It is an open layer:

• a protocol anyone can implement;
• a visual tool anyone can use;
• a container any compliant system can parse;
• a validator anyone can inspect;
• and a public index no single AI company controls.

That is the broader WebGraphs idea.

Schema helped the web describe things. Sitemaps helped crawlers find pages. A future knowledge exchange standard could help agents understand how entities, evidence, capabilities and actions fit together.

The web is moving from documents to decisions.

The question is whether knowledge remains trapped inside pages, or becomes portable infrastructure.

---

Read next: Web 6.0: The Autonomous Internet, then explore the active builds and working hypotheses inside The Lab.