The future of the website  🔗 **Source:** [The future of the website](https://joost.blog/future-of-the-website/) > [!note]+ Document Notes > tags: [agent-ready-web, website-strategy, machine-readable-content, digital-authenticity, ai-discovery] > author: Joost de Valk > source: Joost.blog > reference: https://joost.blog/future-of-the-website/ > date created: 2026-07-22 > > # 🏷️ Title: The future of the website > <!-- deep-read:sections start --> ## 🧭 Table of Contents - [🔗 References](#-references) - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [📚 References](#-references-1) - [🗣️ Paraphrase](#-paraphrase) - [📰 Newsletter](#-newsletter) - [✅ Action Items](#-action-items) - [🧪 Hypotheses](#-hypotheses) - [📊 Methodology](#-methodology) - [❓ Questions](#-questions) - [🔮 Future Research](#-future-research) - [💡 Implications](#-implications) - [🧠 Learning Styles Assessment](#-learning-styles-assessment) - [📊 Accessibility Assessment](#-accessibility-assessment) - [🌍 DEIB Assessment](#-deib-assessment) - [✅ Logical Validity](#-logical-validity) - [🛡️ DARVO Analysis](#-darvo-analysis) - [🎯 Conclusion](#-conclusion) - [🔄 Refactor Appendix](#-refactor-appendix--the-future-of-the-website) ### 🔗 References - [Source article](https://joost.blog/future-of-the-website/) - [Agent-ready websites](https://joost.blog/agent-ready/) - [AI optimization is replaying early SEO](https://joost.blog/ai-optimization-is-replaying-early-seo-just-faster/) - [Agent-native websites](https://joost.blog/agent-native/) - [Website specification](https://specification.website/) ## 📝 Summary - **What happened:** Joost de Valk argues that websites are evolving from human-facing HTML destinations into verified, machine-readable business home bases serving people, search engines, and action-taking agents. - **Why it matters:** Discovery and conversion increasingly happen in AI answers, chats, and agent-mediated transactions. A site must supply trustworthy facts and interfaces even when the final decision happens elsewhere. - **Stakeholders:** Site owners, publishers, ecommerce teams, local businesses, developers, SEOs, AI-agent builders, customers, standards groups, and platform vendors. - **Risks and tradeoffs:** Separate agent sites drift; interface standards will change; machine optimization can distract from human value; technical perfection may become commodity work; authenticity claims can become another gameable checklist. - **Bottom line:** Keep one canonical source of truth, generate multiple interfaces from it, make claims verifiable across the web, and invest in real people, work, and reputation rather than generic generated content. ## 📖 Main Content ### 🗒️ Notes #### Websites serve humans and agents The article reframes bots as a long-standing audience rather than a new one. Search engines learned to interpret human pages through extraordinary investment, while newer agents often need explicit, predictable surfaces. The practical implication is not to abandon human pages but to expose the same facts through structures agents can parse and act on. #### Discovery surfaces change without permission The movement among Google, Bing, ChatGPT Search, GEO, and AEO shows that site owners do not control which discovery surface matters. Optimization therefore shifts from one dominant crawler toward portable content, structured facts, and observable retrieval across several systems. #### One source should produce several renderings The m-dot and AMP analogy supports a strong operational warning: parallel content stores diverge. Human HTML, structured data, feeds, APIs, MCP tools, and command-line interfaces should be projections of the same governed content model. #### Technical correctness becomes cheaper Agents can read specifications and audit implementations, reducing the scarcity of standards knowledge. Technical quality still matters, but it moves toward a baseline. Strategy, original evidence, judgment, and genuine differentiation become more valuable. #### A website is an interface bundle The durable claim is broader than MCP: a website becomes a collection of interfaces. Human visitors read and browse; agents query availability, book appointments, or place orders. Standards may change, but the need for safe, governed machine actions persists. #### Conversion moves away from owned pages The site remains important even when a buyer never visits it. It becomes the canonical system that feeds facts, policies, inventory, identity, and evidence to third-party decision surfaces. #### Authenticity becomes checkable infrastructure The article treats authenticity as consistency across identity, provenance, customer evidence, external references, and owned content. Agents can compare claims across sources cheaply, making contradictory or synthetic brand narratives easier to detect. ## 📚 References - de Valk, Joost. [“The future of the website.”](https://joost.blog/future-of-the-website/) - de Valk, Joost. [“Agent-ready.”](https://joost.blog/agent-ready/) - de Valk, Joost. [“Agent-native.”](https://joost.blog/agent-native/) - de Valk, Joost. [“AI optimization is replaying early SEO, just faster.”](https://joost.blog/ai-optimization-is-replaying-early-seo-just-faster/) - [specification.website](https://specification.website/) - Search Engine Journal. [ChatGPT Google-search fallback report](https://www.searchenginejournal.com/chatgpt-appears-to-use-google-search-as-a-fallback/552089/). - Suganthan Mohanadasan. [ChatGPT source-selection research](https://suganthan.com/blog/how-chatgpt-picks-sources-part-2/). ## 🗣️ Paraphrase 1. The author asks what a website becomes as AI changes discovery and interaction. 2. His answer is a business home base that people and agents can read, use, and verify. 3. Bots have always consumed websites, but modern agents often lack Google’s mature ability to infer structure from human-first pages. 4. GEO and AEO reflect a fragmented discovery market rather than a wholly new practice. 5. Retrieval providers can change rapidly, so businesses need portable, well-structured source material. 6. Building a separate agent site repeats the maintenance failures of m-dot sites and AMP; shared content with several outputs is safer. 7. Agent-assisted standards work makes technical correctness easier to obtain and less differentiating. 8. Websites matter differently by business model: a local tradesperson may not need more demand, while a global brand depends on a universal digital storefront. 9. A serious website will expose action-capable machine interfaces in addition to HTML. 10. Purchasing decisions increasingly happen outside the site, turning the site into a system of record for other surfaces. 11. Generic generated content loses value because it is abundant; verifiable real-world evidence becomes scarce and valuable. 12. Authenticity depends on consistent identity and third-party corroboration, not self-assertion alone. 13. The future website remains alive but changes into a technically sound, agent-readable, demonstrably real home base. ## 📰 Newsletter The website isn’t dying; it’s getting more jobs. Joost de Valk’s useful frame is that your site becomes the canonical home base behind human pages, AI answers, and agent actions. Don’t build a second “AI site” that drifts out of sync. Keep one governed source and publish several interfaces from it. Technical correctness is getting cheaper; real work, real people, and claims that survive cross-checking are becoming the hard advantage. Start with [the article](https://joost.blog/future-of-the-website/) and its companion pieces on [agent-ready](https://joost.blog/agent-ready/) and [agent-native](https://joost.blog/agent-native/) sites. ## ✅ Action Items #### DevRel - Publish a reference implementation showing HTML, structured data, API, feed, and agent interface generated from one content model. - Teach threat modeling and authorization for action-taking interfaces, not only discovery metadata. #### OSPO - Track open standards for agent discovery and actions; avoid binding core business data to one vendor protocol. - Encourage reusable open schemas and conformance tests. #### Community - Collect case studies from small businesses, publishers, nonprofits, and global brands so “the website” is not defined only by ecommerce and enterprise needs. #### WordPress - **Core:** Evaluate canonical APIs for exposing site facts and capabilities without duplicating content. - **Core Performance:** Measure agent-facing responses, cacheability, and the cost of extra interface layers. - **Core AI:** Explore provider-neutral discovery and action surfaces with explicit permissions. - **Core Program Team:** Coordinate cross-team ownership for schemas, identity, and compatibility. - **Mobile:** Keep the mobile app aligned with the same canonical content and identity model. - **Playground:** Build demos that expose one site through human and machine interfaces. - **Community:** Gather user needs before treating MCP or any single protocol as universal. - **Training:** Create modules on agent-ready content, provenance, and safe actions. - **Marketing:** Explain the website as a trusted source without promising guaranteed AI citations. - **Support:** Document common failures: stale structured data, conflicting facts, inaccessible content, and over-broad agent permissions. - **Design:** Preserve human clarity while exposing machine-readable structure. - **Documentation:** Publish versioned contracts and examples for each machine interface. - **Photos:** Strengthen provenance and licensing metadata for visual assets. - **TV:** Record practical demonstrations and caption them accurately. - **Polyglots:** Test whether machine interfaces preserve locale, translation, and language negotiation. - **Accessibility:** Ensure agent-oriented changes do not weaken semantic HTML or assistive-technology support. - **Meta:** Explore identity and provenance signals across WordPress.org profiles and properties. - **CLI:** Offer inspection commands for canonical facts, schemas, and agent endpoints. - **Hosting:** Provide monitoring, authentication, rate limiting, and logs for machine actions. - **Tide:** Add checks for semantic, structured-data, and interface-contract quality where appropriate. - **Openverse:** Preserve source, creator, license, and provenance metadata in downstream interfaces. - **Themes:** Keep presentation separate from canonical facts and action contracts. - **Plugins:** Use shared capabilities and permissions instead of inventing isolated agent endpoints. #### The WPOCC - Convene interoperable guidance for agent discovery, identity, provenance, and safe transactions across the WordPress ecosystem. #### FAIR - Test whether discoverability and action metadata remain portable across hosts, plugins, agents, and protocol changes. ## 🧪 Hypotheses 1. Sites exposing one governed content source through human and machine interfaces will show fewer fact inconsistencies than sites maintaining separate agent content. 2. Verifiable author, organization, and customer evidence will correlate more strongly with AI citations than generic high-volume content. 3. Agent-action interfaces will increase completed transactions but also increase authorization and fraud incidents unless permissions are narrowly scoped. 4. Automated specification audits will compress technical-quality differences among competing sites within two years. 5. Businesses with strong off-site corroboration will be recommended more consistently across retrieval providers than businesses relying on owned claims alone. ### 📊 Methodology - Build a longitudinal panel of sites with and without machine-readable interfaces; track factual consistency, citations, referrals, and conversions across major agents. - Run controlled tests comparing a shared-source architecture with separately maintained human and agent sites. - Audit agent actions for authorization scope, error recovery, fraud controls, and observability. - Create a provenance score from author identity, external references, reviews, organizational records, and content history; compare it with recommendation frequency. - Interview businesses for which extra demand is not the goal to test the article’s assumptions about universal website value. ### ❓ Questions - Which website facts should be canonical, and who governs them? - How should agents authenticate, obtain consent, and prove an action was authorized? - What prevents machine interfaces from becoming a new inaccessible, pay-to-play distribution layer? - How can a business correct stale facts cached by third-party agents? - When is an API or feed enough, and when does an action protocol add real value? - How can authenticity be measured without disadvantaging new, small, pseudonymous, or marginalized creators? ### 🔮 Future Research - Compare MCP with emerging alternatives using portability, security, accessibility, and governance criteria. - Study how retrieval providers resolve conflicting claims across owned and third-party sources. - Measure whether agent traffic can be observed without invasive tracking. - Develop open conformance tests for fact freshness, provenance, and action safety. - Examine regulatory responsibilities when an agent transacts from stale or incorrect website data. ### 💡 Implications - **Practice:** Model content once, publish several representations, and monitor them for drift. - **Policy:** Require clear authorization, receipts, revocation, and dispute paths for machine actions. - **Security:** Treat agent endpoints as production APIs with least privilege, validation, rate limits, and audit logs. - **Education:** Shift SEO training toward information architecture, provenance, interface design, and cross-surface measurement. - **Funding:** Support open schemas, test suites, and public-interest identity systems so small publishers can participate. ### 🧠 Learning Styles Assessment - **Visual:** A diagram of canonical content feeding HTML, feeds, APIs, MCP, and third-party agents would clarify the architecture. - **Aural:** A narrated case study could explain how one business fact travels through several discovery surfaces. - **Verbal:** The article uses memorable contrasts—one source versus parallel sites, authenticity versus generic output. - **Kinesthetic:** Learners can audit a site, identify duplicated facts, and prototype one machine-readable endpoint. - **Logical:** The claim chain is clear: fragmented discovery increases interface needs; shared sources reduce drift; verifiable evidence supports trust. - **Social:** Teams can debate which actions should be exposed and assign ownership for canonical facts. - **Solitary:** A checklist can guide a private content, identity, and interface inventory. - **Naturalistic:** Not directly supported; an analogy to an ecosystem with one root system and several visible forms may help. - **Reading level:** Approximately grade 10–12; short sentences and examples make the strategic argument accessible, while MCP and retrieval concepts need definitions. ### 📊 Accessibility Assessment - **Screen reader semantics — WCAG 1.3.1, 4.1.2:** Source headings are clear. Any machine-interface work must preserve semantic HTML rather than treating it as obsolete. - **Keyboard navigation — WCAG 2.1.1, 2.1.2:** Not directly assessed from article text; interactive agent controls need complete keyboard operation. - **Contrast and readability — WCAG 1.4.3, 1.4.11:** Not assessable from Reader text alone; verify the live presentation separately. - **Text resizing and reflow — WCAG 1.4.4, 1.4.10:** Not assessable from extracted text; human rendering must remain responsive even if machine interfaces expand. - **Media and captions — WCAG 1.2:** No essential audio or video appears in the source text. - **Forms and interactions — WCAG 3.3, 4.1.2:** Action-taking interfaces need accessible names, errors, confirmations, and recovery paths. - **Standards reference — WCAG 2.2 AA:** Agent readiness is not a substitute for WCAG conformance; semantic accessibility helps both humans and machines. ### 🌍 DEIB Assessment - **Cultural breadth:** The plumber-versus-global-brand contrast adds economic variety but remains grounded in one European business context. - **Inclusive language:** Generally direct and non-stigmatizing; “AI slop” is rhetorical and may reduce precision. - **Neurodiversity support:** Strong headings and concrete examples help scanning; an architecture diagram and glossary would reduce abstraction. - **Representation:** The article discusses businesses and agents more than affected workers, disabled users, multilingual communities, or regions with limited connectivity. - **Psychological safety:** The tone is urgent but not shaming. It permits a rational choice not to invest in a website when demand is already sufficient. ### ✅ Logical Validity #### Strong reasoning - The argument against separate agent sites draws on known synchronization failures from m-dot sites and AMP. - The Bing/Google retrieval shifts illustrate platform volatility and support multi-surface portability. - The distinction between technical baseline and authentic differentiation follows from reduced production cost, though the speed of that shift remains uncertain. #### Fallacies - **Hasty generalization:** “The only thing that can’t be generated is being real” overstates current and future limits; synthetic systems can fabricate convincing evidence. Fix: say durable trust depends on verifiable real-world provenance that is costlier to fake. - **False inevitability:** “A website … should at the very least also be an MCP server” elevates one current protocol into a minimum requirement without adoption or outcome data. Fix: recommend a governed machine interface and treat MCP as one candidate. - **Appeal to novelty:** New agent surfaces are sometimes treated as inherently decisive. Fix: require business-specific traffic, task, and conversion evidence before investing. #### Tone of voice Confident, provocative, strategic, and conversational. Personal experience and memorable claims make it readable, but some absolute phrasing should be tested rather than repeated as settled fact. ### 🛡️ DARVO Analysis - **Deny:** Not applicable. The article does not deny a reported harm or responsibility claim. - **Attack:** Not applicable. It criticizes generic content and outdated architectures, not a victim or complainant. - **Reverse victim/offender:** Not applicable. No interpersonal accountability dispute appears. ## 🎯 Conclusion The strongest practical idea is not “install MCP”; it is to make the website a governed source of truth that can serve several human and machine interfaces without content drift. Technical correctness will remain necessary but may become easier to obtain. Durable differentiation shifts toward original evidence, consistent identity, safe actions, and third-party corroboration. Start by auditing canonical facts and duplicated content, then add the smallest interoperable machine surface that supports a measured user or business need. ## 🔄 Refactor Appendix — The future of the website ### 📚 Literature Atoms (H3) #### One source should generate every website interface > [!quote] ID:rw_1 > “The answer isn’t a second website. It’s one source of truth with multiple renderings.” **Context & use**: - The article’s clearest architecture rule; applies to HTML, structured data, feeds, APIs, and agent interfaces. **Tags**: #xref #articleseed <!-- atomicity: 1 idea ✓ --> #### Website conversion shifts into external agent surfaces > [!quote] ID:rw_2 > “More and more buying decisions happen away from your site: in AI answers, in chat interfaces, in agent-mediated transactions.” **Context & use**: - Reframes the website from final destination to canonical source for distributed decisions. **Tags**: #xref <!-- atomicity: 1 idea ✓ --> #### Websites become bundles of human and machine interfaces > [!quote] ID:rw_3 > “The durable point is this: a website is becoming a bundle of interfaces.” **Context & use**: - Protocol-independent version of the article’s MCP recommendation. **Tags**: #xref <!-- atomicity: 1 idea ✓ --> #### Authenticity becomes verifiable infrastructure > [!quote] ID:rw_4 > “Authenticity stops being a marketing word and becomes infrastructure: identity, provenance, trust.” **Context & use**: - Strongest differentiation claim; requires careful work on fairness and resistance to fabricated evidence. **Tags**: #articleseed #xref <!-- atomicity: 1 idea ✓ --> ### 🧠 Permanent Candidates (H4) #### Canonical sources reduce interface drift **Claim:** A business should store governed facts once and generate each human or machine interface from that source. **Evidence pointer:** [[#One source should generate every website interface]] **Scope & conditions:** Holds when renderings share ownership, validation, and freshness controls; breaks when protocol-specific needs are forced into an unsuitable common model. **Links:** - [[Content governance]] — because defines ownership and freshness - [[Agent-ready web]] — because applies the rule to machine consumers <!-- atomicity: 1 idea ✓ --> #### Verifiable authenticity becomes a system property **Claim:** Trustworthy digital identity depends on consistent provenance and corroboration across owned and external sources. **Evidence pointer:** [[#Authenticity becomes verifiable infrastructure]] **Scope & conditions:** Holds when evidence is independently checkable; breaks when verification systems reproduce popularity or access bias. **Links:** - [[Digital provenance]] — because supplies evidence - [[E-E-A-T]] — because contrasts inferred trust with direct checking <!-- atomicity: 1 idea ✓ --> ### ✏️ Fleeting Queue - [F1] Inventory the canonical business facts currently duplicated across pages, schemas, feeds, and profiles. #xref - [F2] Compare MCP with a read-only API or feed for one real business task before choosing a protocol. #articleseed ### → Things 3 (max 2 actions) - [ ] Audit one site for fact drift across human and machine surfaces — website governance #things - [ ] Define one safe, measurable agent task and its permission boundary — agent interface #things <!-- deep-read:sections end --> ## Highlights - The answer isn’t a second website. It’s one source of truth with multiple renderings. - Note: One canonical content source should generate human and machine interfaces. #atom/literature/h1 - More and more buying decisions happen away from your site: in AI answers, in chat interfaces, in agent-mediated transactions. - Note: The site becomes a system of record for external decision surfaces. #atom/literature/h2 - The durable point is this: a website is becoming a bundle of interfaces. - Note: Protocol-independent framing for human and machine interfaces. #atom/literature/h2 - Authenticity stops being a marketing word and becomes infrastructure: identity, provenance, trust. - Note: Verifiable real-world evidence becomes a system property. #atom/literature/h1 ## New Highlights added 2026-07-22 at 8:57 AM - Before you rush off to build a separate site for agents: we’ve done that before. Remember m-dot sites? Remember AMP? Parallel versions of your website always drift out of sync, and then they die. The answer isn’t a second website. It’s [one source of truth with multiple renderings](https://joost.blog/agent-native/), where the machine-friendly surface is generated from the same content the human one is. ([View Highlight](https://read.readwise.io/read/01ky4xfz5c0aw3te8s1wj0kfmb)) - Note: One governed content source should generate human and machine interfaces, avoiding drift between parallel sites. - Tags: website-strategy, atom/literature/h1, agent-ready-web - The specific standards will churn. MCP is today’s answer; something else will come after it. The durable point is this: a website is becoming a bundle of interfaces. One for humans, several for machines, and the machine ones don’t just answer questions, they take actions. ([View Highlight](https://read.readwise.io/read/01ky4xg118rz6khs5erz7rvx41)) - Note: The durable architecture claim is protocol-independent: one website supports human reading and machine queries or actions. - Tags: machine-readable-content, agent-ready-web, atom/literature/h2 - More and more buying decisions happen away from your site: in AI answers, in chat interfaces, in agent-mediated transactions. This is the uncomfortable one. You can rage against it or you can plan for it. ([View Highlight](https://read.readwise.io/read/01ky4xfzwxfhjpq50xpkf1xhy7)) - Note: The site increasingly acts as a canonical system of record for decisions made in external AI and chat surfaces. - Tags: website-strategy, ai-discovery, atom/literature/h2 - In an agent world, authenticity can’t just be claimed, it has to be checkable. An agent choosing between two suppliers can cross-reference everything, cheaply and instantly. Does the story on your site match the trail you leave everywhere else? Do the people actually exist? Do the reviews hold up? SEOs will recognize the goal: this is what Google’s E-E-A-T was always reaching for. The difference is that Google estimates trust from signals; an agent can simply go and check. Authenticity stops being a marketing word and becomes infrastructure: identity, provenance, trust. Proving you are who you say you are is about to become the most valuable layer of the web. ([View Highlight](https://read.readwise.io/read/01ky4xg23hgqfx12w8mped28zm)) - Note: Verifiable identity, provenance, and trust become system properties rather than unsupported brand claims. - Tags: ai-discovery, digital-authenticity, atom/literature/h1 ## New Highlights added 2026-07-22 at 9:28 AM - Not dead. Not even close. But different: the website is becoming the verified, machine-readable home base of a business. It feeds every other surface. It answers agents and takes their orders. It proves the business is real. And it still, sometimes, gets visited by an actual human, who had better find something authentic when they arrive. ([View Highlight](https://read.readwise.io/read/01ky4yfdxdfnw7p5g17vs8sqak)) - Tags: 👻 ai highlighted ## Links - [[3. Resources/Permanent Notes/Documentation Is Dual-Audience Infrastructure]] — extends: one governed source can serve both human and machine interfaces. - [[3. Resources/Permanent Notes/AI-Facing Documentation Needs an Explicit Authority Boundary]] — qualifies: agent-readable facts need clear provenance and authority.