Optimising for the Surfaceless Web  ## Metadata - Author: [[Jono Alderson]] - Full Title: Optimising for the Surfaceless Web - Category: #articles - Document Note: --- tags: [AI, Marketing, SEO] author: Jono Alderson source: jonoalderson.com reference: [https://www.jonoalderson.com/conjecture/optimising-for-the-surfaceless-web/] date created: 2025-10-30 00:00 --- # 🏷️ Title: Optimising for the surfaceless web ## 🧭 Table of Contents - [📝 Summary](#-summary) - [📖 Main Content](#-main-content) - [🗒️ Notes](#️-notes) - [📚 References](#-references) - [🗣️ 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) ### 🔗 References - Include links cited in or relevant to the article. ## 📝 Summary - What happened. - The web’s “surface” (pages, clicks, SERPs) is being compressed into AI models’ internal representations. - Why it matters. - Visibility becomes an echo; survival depends on being useful, true, and integral within the model’s map. - Key stakeholders. - Marketers, publishers, brands, SEO/SEM teams, platform builders, AI labs. - Risks and tradeoffs. - Ephemeral campaigns decay; manipulative signals get filtered; overfitting to model priors risks loss of nuance. - Bottom line. - Optimise for persistence in the model: clarity, consistency, connection over surface performance. ## 📖 Main Content - Explain the core arguments in plain language. - AI digests web content into a probabilistic “substrate.” What is shown to humans is downstream of what models already “know.” - Selection pressure: stable, corroborated, connected information persists; noise and contradictions dissolve. - Marketing shifts from seeking attention to earning existence inside the model. - Add sourced context and define terms. - “Substrate”: the model’s internal knowledge graph/distribution of associations. - “Persistence”: recurring recognition across training cycles. - Note uncertainties or contested claims. - Degree of model memory and update cadence varies by vendor; exact weighting of “useful/true/integral” is speculative but directionally sound. ### 🗒️ Notes - “There are no more surfaces – only systems.” - Visibility is “a by-product of how well you exist in the model’s internal map.” - Selection pressure favors clarity, corroboration, connection; “campaign noise burns off.” - Traits that endure: Useful (reduces ambiguity), True (statistically coherent), Integral (structurally connected). - “You don’t persuade a model. You support it.” - Temporal persistence: reassert coherent patterns across time and media; survive model updates. - “The game isn’t about visibility anymore. It’s about viability.” ## 📚 References - Optimising for the surfaceless web (original article) (https://www.jonoalderson.com/conjecture/optimising-for-the-surfaceless-web/) - Google Search Generative Experience / AI Overviews explainer (https://blog.google/products/search/generative-ai-search/) - OpenAI: Models and training overview (https://platform.openai.com/docs/guides/reasoning) - Schema.org (structured data) (https://schema.org/) - W3C: Verifiable Credentials overview (https://www.w3.org/TR/vc-data-model-2.0/) ## 🗣️ Paraphrase - Paragraph A paraphrase. - As AI compresses the web into internal models, the visible layer becomes a residue; relevance is determined before anything appears to users. - Paragraph B paraphrase. - To endure, information must be useful, statistically consistent, and structurally connected, so that removing it would weaken the model’s understanding. ## 📰 Newsletter - The web’s “surface” is fading as AI systems internalise our content. Winning now means becoming part of the model’s scaffolding: publish clarifying, consistent, well-cited resources; align language, data, and presence; and reinforce over time. For WordPress folks: invest in structured data, canonical taxonomies, entity pages, and provenance. Useful links: Schema.org, verifiable credentials, and performance/SEO checklists. ## ✅ Action Items #### DevRel - Developer advocacy: Run “model-first content” talks; demo entity alignment; measure citations/links won. - Education: Workshop on schema, disambiguation, provenance. - Developer experience: Provide SDK/examples for structured data and content signatures. - Developer marketing: Publish canonical explainer series; track inclusion in AI answers. - Community management: Guide on coherent language/terms; moderation for consistency. - Internal/external/cross-project: Set SLAs for taxonomy and schema updates. #### OSPO - Policy & Governance: Contribution guidelines for factuality, citations, and change logs. - License Compliance: Automate license metadata in feeds. - Security Management: Sign releases; publish attestation docs. - Software Inventory: SBOMs per release; freshness SLOs. - Developer Support: Office hours on provenance and data integrity. - Community Engagement: Fund upstream schema and tooling. - Strategic Alignment: OKRs on structured data coverage and entity coherence. #### Community - Maintainers: Roadmap clinics on naming/versioning coherence. - Contributors: First-issue rotations on doc fixes and schema. - Sponsors: Tie funding to measurable knowledge integrity outcomes. - Ecosystem: Cross-project incident drills for misinformation corrections. #### WordPress - Core: First‑class structured data APIs; entity registry; performance budgets. - Core Performance: Metrics for crawlability and content stability. - Core AI: Guardrails for provenance and eval sets for coherence. - Core Program Team: Cadence for cross-team terminology alignment. - Mobile: Offline-friendly canonical content modules. - Playground: Labs to test structured data and crawl scenarios. - Community: Meetup kits on “surfaceless SEO.” - Training: Lessons on schema, entities, citations. - Marketing: Narrative on “from visibility to viability.” - Support: KB gaps index; escalation for factual corrections. - Design: UX for canonical entity pages; design tokens. - Documentation: Examples-first schema patterns; term glossaries. - Photos: Metadata hygiene and licensing signals. - TV: Caption quality and transcripts for model ingestion. - Polyglots: Glossary alignment across locales. - Accessibility: Audits that also improve machine parsing. - Meta: Reliable .org APIs; contributor identity signals. - CLI: Generate schema and provenance by default. - Hosting: Defaults for sitemaps, compression, signatures. - Tide: Signal accuracy and false-positive review loops. - Openverse: Rights/provenance enrichment; ingestion quality. - Themes: Lints for structured data and semantics. - Plugins: Provenance checks; security liaison patterns. #### The WPCC - Programs: Stipends for docs and structured data stewards. - Projects: Governance clinics; release-team documentation sprints. - Funding experiments: Matched grants for citation-winning content. ### FAIR - Technical independence: Local metadata, signatures, verifiable releases. - Policies and covenants: Funding transparency; anti-capture clauses. - Productized governance artifacts: Charters, operator agreements, scorecards. ## 🧪 Hypotheses - Content with higher citation density achieves greater inclusion in AI answers. - Consistent entity naming/taxonomy reduces contradiction penalties across model updates. - Verifiable provenance (signing/VCs) increases model trust and persistence. - Multi-format reinforcement (text/code/data) improves temporal persistence vs. single-format posts. ### 📊 Methodology - Measure AI answer inclusion pre/post structured data rollout (A/B across properties). - Longitudinal tracking of brand/entity mentions across model updates. - Controlled release with signed artifacts vs. unsigned; compare model citations. - Cross-format publishing experiment; assess recall in generative answers. ### ❓ Questions - How often do major models overwrite entity associations? - Which provenan - Summary: The web we built had surfaces. We could design them, decorate them, and fight for space upon them. That web is gone. There are no more surfaces - only systems. - URL: https://www.jonoalderson.com/conjecture/optimising-for-the-surfaceless-web/  When I wrote about [the machine’s emerging immune system](https://www.jonoalderson.com/conjecture/machine-immune-systems/), I argued that AI ecosystems would eventually learn to protect themselves. They’d detect manipulation, filter noise, and preserve coherence. They’d start to decide what kinds of information were safe to keep, and which to reject. That wasn’t a prediction of some distant future. It’s happening now. Every day, the surface of the web is scraped, compressed, and folded into the models that power the systems we increasingly rely on. In that process, most of what we publish doesn’t survive contact. Duplicate content dissolves. Contradictions cancel out. Persuasive noise is treated as waste heat and vented into the void. What remains isn’t the web as we know it – it’s something flatter, quieter, and more internal. A composite understanding of the world, shaped by probability and consensus. A compressed version of reality, organised for machines to reason about rather than for humans to explore. And that’s the new landscape we’re marketing into: one with no pages, no clicks, and *no real surface at all*. We built our disciplines on visibility. On the idea that success meant being seen. But as the machine closes itself off, as the interfaces between human experience and machine understanding grow thinner, visibility starts to lose meaning. The web we built had surfaces. We could design them, decorate them, and fight for space upon them. That web is gone. There are no more surfaces – only systems. So the question isn’t how to stand out anymore. It’s how to stay *inside*. #### [**Visibility is legacy**](https://www.jonoalderson.com/conjecture/optimising-for-the-surfaceless-web/#h-visibility-is-legacy) For most of the web’s history, marketing has been an act of performance. You built something on the surface – a website, a post, a page – and hoped that enough people, and enough algorithms, would look at it. The goal was to be discoverable. Searchable. Indexed. That surface once mattered. It was where meaning lived; a thin film of presentation stretched across a sea of content. We learned how to decorate it, how to manipulate its texture, how to catch the light. But surfaces don’t mean much to systems that don’t look. AI doesn’t browse, or scroll, or click. It doesn’t see the things we publish as artefacts, or experiences, or journeys. It sees them as data: a dense accumulation of words and links and relationships. What was once a web of pages becomes, in its eyes, a map of connections and probabilities. As those systems grow, the visible layer stops being the web’s living tissue and becomes something closer to a residue. A user interface. A translation layer between the model’s internal state and our human need for touchable things. What we still see – the search results, the snippets, the website visits – are just reflections of a deeper process. The model has already decided what matters before anything becomes visible to us. Visibility, then, is no longer a measure of relevance. It’s an *echo*. A by-product of how well you exist in the model’s internal map of the world. Marketing, then, can’t rely on that shadow anymore. You can’t simply perform on the surface and expect to be understood below it. You have to exist in the substrate – in the model’s inner representation of reality – where things are no longer seen, only *known*. The game isn’t about visibility anymore. It’s about becoming part of the organism – being recognised as useful, true, and integral to its functioning. #### [**The machine’s new selection pressure**](https://www.jonoalderson.com/conjecture/optimising-for-the-surfaceless-web/#the-machines-new-selection-pressure) Every cycle of training, pruning, and retraining redraws the boundaries of what the machine believes to be true. Information doesn’t simply disappear; it’s weighed, compared, and either reinforced or allowed to fade. Fragments that align with the model’s broader understanding are retained and strengthened. Those that contradict it, or contribute nothing new, dissolve quietly into statistical noise. This isn’t malice – just maintenance; the system trimming what no longer fits. From a distance, it looks a lot like natural selection. Ideas that are clear, consistent, and corroborated tend to persist. They become anchor points in the system’s internal map of the world. Messy, manipulative, or self-serving signals lose fidelity with each generation until they’re effectively gone. In a web of endless redundancy and synthetic repetition, that selection pressure is profound. Only the most stable patterns survive ingestion, compression, and re-ingestion. Everything else – the campaigns, the slogans, the *noise* – burns off in the friction between training cycles. That’s what survival looks like now. Not attention. Not engagement. But *persistence*. #### [**Survival traits: Useful. True. Integral.**](https://www.jonoalderson.com/conjecture/optimising-for-the-surfaceless-web/#survival-traits-useful-true-integral) If visibility has given way to persistence, the question becomes: what kinds of information survive? What traits make something worth keeping, when the machine is constantly distilling and rewriting the world? Three patterns seem to emerge. They’re not rules so much as instincts – signals that the system tends to reinforce rather than erase. ##### [**Useful**](https://www.jonoalderson.com/conjecture/optimising-for-the-surfaceless-web/#useful) The model keeps what helps it make better predictions. Content that clarifies, contextualises, or resolves ambiguity adds weight. It’s *nutritious* to the system. It strengthens its ability to connect ideas and answer questions. Marketing, by contrast, has often specialised in the opposite – in the art of adding noise to capture attention. In a surface-driven world, that worked. In a closed, self-editing one, it reads as contamination. To persist, information has to improve the model’s understanding of reality. In practice, that means producing things that make the broader web smarter – content that’s cited, contextualised, and consistent wherever it’s encountered. ##### [**True**](https://www.jonoalderson.com/conjecture/optimising-for-the-surfaceless-web/#true) Truth, in this context, isn’t about moral correctness. It’s about statistical coherence – about holding the same shape across many perspectives. Signals that remain consistent across time, context, and corroboration develop gravity. They become stable landmarks in the machine’s internal map of the world. Contradictions, rebrands, and rhetorical pivots weaken that gravity until the entity itself loses definition. If the machine can’t tell what you are, it stops believing you exist. ##### [**Integral**](https://www.jonoalderson.com/conjecture/optimising-for-the-surfaceless-web/#integral) The most resilient information isn’t the loudest or the most original. It’s the most connected. Ideas that are cited, linked, quoted, or built upon become structural. They help to shape the scaffolding that other knowledge depends on. Delete them, and the system’s understanding weakens. That’s what it means to be integral – to occupy a position where absence would create tension. To be so entangled in the model’s sense of how the world fits together that forgetting you would break something. These are the new markers of durability: usefulness, truthfulness, and integration – the traits that make you hard to erase. #### [**Marketing as symbiosis**](https://www.jonoalderson.com/conjecture/optimising-for-the-surfaceless-web/#marketing-as-symbiosis) If the machine is learning what to keep, then marketing becomes an act of coexistence. The challenge isn’t to outsmart the filters or to force attention; it’s to make yourself part of the pattern the system depends on. In a sense, the best marketing has always done this. The most durable brands, ideas, and voices are those that made themselves *useful* to the environment they occupied – not through volume, but through contribution. What’s changed is that the environment is now synthetic, recursive, and self-curating. The audience is a model. You don’t persuade a model. You support it – with clarity, consistency, and connection, the same qualities it’s built to recognise and reward. That means thinking less about *performance* and more about *participation*. Your goal isn’t to win visibility, but to earn residency – to become something the machine recognises as part of its metabolism. The idea of “brand” shifts in that light. A brand isn’t just a promise to people anymore; it’s also a pattern that helps machines resolve ambiguity – a consistent cluster of language, data, and context that supports their confidence. When your language, data, and presence align in ways that improve the system’s accuracy, you stop being external. You become infrastructure. And once that happens, you’re no longer competing for attention at all. You’re competing for existence. #### [**Temporal persistence**](https://www.jonoalderson.com/conjecture/optimising-for-the-surfaceless-web/#temporal-persistence) Nothing inside the machine stays still for long. Each new wave of training reshapes the landscape – compressing, merging, and overwriting its understanding of the world. Entities drift. Meanings decay. Facts are reinterpreted. The machine forgets as quickly as it learns. In that churn, persistence becomes an active process. You don’t exist once; you exist *continuously*, by reasserting your pattern every time the world is rewritten. That doesn’t mean producing endless noise. It means maintaining coherence – allowing the traces of your presence to echo across time, context, and medium. The goal isn’t repetition, but reinforcement. Old marketing rewarded novelty. The machine rewards consistency. It trusts what endures. It preserves what it can still recognise after compression, translation, and paraphrase. To survive, treat your reputation, language, and data as living systems – resilient enough to deform without breaking. Most campaigns capture a moment; few survive a model update. Most campaigns are built to capture a moment. Few are built to survive a model update. #### [**Existence as marketing**](https://www.jonoalderson.com/conjecture/optimising-for-the-surfaceless-web/#existence-as-marketing) The machine immune system doesn’t just defend against manipulation. It shapes what becomes visible, repeatable, and credible across the systems we now rely on. Every model update redraws those boundaries a little – deciding what information stays accessible, and what quietly fades. Our role inside that process is changing. Marketing used to be the art of visibility: we tried to be seen, to be remembered, to be chosen. Now, in a system that no longer looks in any human sense, visibility has no real meaning. What matters is whether we’re included in the model’s internal understanding of how the world works – whether we’re *kept*. That’s the strange new responsibility of marketing: to make our presence valuable enough that deletion would leave a gap. The future belongs to those who remain useful, true, and integral – not as moral virtues, but as traits the system depends on. The rest will fade quietly, not from malice but from optimisation. The machine will keep compressing the world, folding it inward until there are no real surfaces left – only patterns it trusts to describe reality. Our task is to make sure we’re still one of them. Because the game isn’t about visibility anymore. It’s about viability – about whether the machine still remembers you when it dreams.  , Optimising for the Surfaceless Web, Jono Alderson, 01k8v6gfxfc9bw4pj3jn7ydka9.