Stop Publishing Markdown for AI. Do This Instead

Stop Publishing Markdown for AI. Do This Instead. 🔗 **Source:** [Stop Publishing Markdown for AI. Do This Instead.](https://twitter.com/pootlepress/status/2026195442535456880/?rw_tt_thread=True) > [!note]+ Document Notes > tags: [ai, wordpress, content-strategy] > author: Jamie Marsland - Head of WordPress YouTube ❤️ > source: Twitter > reference: [https://twitter.com/pootlepress/status/2026195442535456880/?rw_tt_thread=True] > date created: > > --- > > # 🏷️ Title: Stop Publishing Markdown for AI. Do This Instead. >  ## 🧭 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) - [🧱 Refactor Appendix](#-refactor-appendix) ### 🔗 References - Include links cited in or relevant to the article. --- ## 📝 Summary - The author argues that flattening websites into Markdown “meals” for AI is a mistake. - As AI agents read the web, brands risk stripping away presentation, style, and “taste.” - Stakeholders: publishers, WordPress site owners, AI companies, users relying on agents. - Risk: becoming indistinguishable, losing brand, trust, and emotional resonance. - Bottom line: don’t optimise for robots with bland Markdown; invest in brand, taste, and trust. --- ## 📖 Main Content - The author loves Markdown (even built Pootlewriter) but calls it “the beige Volvo of human expression” because it safely moves words but removes personality. - People are rushing to expose Markdown versions of their sites as if AI needs “simpler English,” ignoring that models already read HTML well. - This trend “strips” typography, layout, pacing, and quirky design decisions—the presentation layer humans actually experience. - Using the Mozart analogy: Markdown is the skeleton/notes; the performance, emotion, and timing are the music. Brands live in that performance, not in plain text. - Stephen Wolfram’s point: AI mainly sees the information layer, while humans feel the taste/style layer; that layer is often missing from AI’s input. - “Taste is the last mile”: spacing, restraint, rule‑breaking, and the sense that “a person who cared” made this. Markdown can’t capture silence, pacing, or care. - As agents summarise and recommend on our behalf, they see flattened Markdown—like judging David Bowie from grocery receipts. - The mistake isn’t using Markdown, but believing “the notes are the music” and over‑optimising skeletons for machines. - In an agentic world, information is cheap; taste, style, and trust become the real differentiators. - Recommendation: stop obsessing over perfect Markdown for AI; obsess over brand, trust, and “beautiful things,” because that’s what humans will crave. --- ### 🗒️ Notes - Markdown: fast, portable, “wonderfully boring,” good for frictionless writing, but poor for conveying personality. - “Everyone is suddenly cooking for robots” by adding Markdown feeds, though AI already parses HTML. - This behaviour is “slightly tragic” because it removes the very things that make each site distinctive. - Mozart analogy highlights the gap between data (notes) and experience (music); brands live in the latter. - Presentation/taste layers: typography, layout, pacing, and creative quirks shape human perception. - Wolfram insight: AI currently consumes information, while humans live in taste and style. - Taste defined as spacing, restraint, knowing when to stop, deliberate rule‑breaking, and care. - Agents will increasingly mediate discovery; if they only see flattened content, our brand is invisible. - Future advantage: systems that preserve and transmit style and trust, not just factual accuracy. - Strategic advice: build memorable, trustworthy brands and rich experiences rather than robot‑friendly text dumps. --- ## 📚 References - Pootlewriter – Markdown writer for WordPress: https://pootlewriter.com - Tweet/thread: https://twitter.com/pootlepress/status/2026195442535456880 - Automattic: https://automattic.com - Stephen Wolfram on AI and information layers (general context): https://writings.stephenwolfram.com --- ## 🗣️ Paraphrase - The web is over‑optimising for AI by serving bland Markdown, forgetting AI already understands rich HTML and that humans care about design, pacing, and style. - In a future dominated by agents, information alone won’t differentiate you; carefully crafted taste, style, and trust will, so prioritise brand and beauty over robot‑friendly notes. --- ## 📰 Newsletter In this week’s WordPress‑and‑AI corner, Jamie Marsland argues we’re making a category error: shipping “beige Volvo” Markdown feeds for AI while neglecting the presentation layer humans actually feel. As agents start recommending sites for us, information parity will be the norm—taste, style, and trust will be the moat. For builders: double down on typography, layout, and narrative, and treat AI as a reader of your brand, not just your bullet points. --- ## ✅ Action Items #### DevRel - Clarify guidance for plugin/theme devs on preserving brand and UX in AI‑indexed content. - Create examples and talks showing “tasteful” WordPress sites that still work great with AI tools. - Add docs/quickstarts on exposing structured data (schema, JSON‑LD) without flattening design. - Run campaigns contrasting “flat Markdown” vs “rich brand” sites in agent answers. - Survey developers on how they think about AI as an audience and share patterns. #### OSPO - Establish guidelines discouraging unnecessary “AI‑only” Markdown mirrors in open‑source docs. - Ensure license and attribution metadata survive when agents summarise project content. - Coordinate with security to keep any AI‑targeted feeds aligned with main site governance. #### Community - Host discussions/office hours on “brand, taste, and AI agents” for maintainers. - Provide contributor templates that respect project visual identity as well as content quality. - Encourage sponsors to fund design, UX, and documentation—not just raw features. #### WordPress - Core: avoid core features that encourage unnecessary flattening to robot‑centric formats. - Core Performance: measure UX and perceived quality, not just page speed, in guidance. - Core AI: frame AI APIs/features around preserving site identity and trust. - Core Program Team: align teams on “design and taste as first‑class citizens” in the AI era. - Mobile: maintain brand consistency between app and web views as agents summarise both. - Playground: offer labs that show how agents interpret different WordPress designs. - Community: add sessions at meetups/WordCamps on AI agents and branding. - Training: produce lessons on designing for humans first, AI second. - Marketing: position WordPress as the place to build brands, not just content farms for AI. - Support: document best practices for AI‑related plugins that don’t degrade UX. - Design: champion style, pacing, and “care” as key design review criteria. - Documentation: include examples that show narrative voice and layout, not only syntax. - Photos/TV/Openverse/etc.: maintain strong curation and taste signals in media assets. - Themes/Plugins: set guidelines to avoid “AI‑bait” themes that strip personality. #### The WPOCC - Fund experiments that explore how agents surface WordPress sites with distinctive design. - Support maintainers who invest in accessibility, aesthetics, and trust signals. ### FAIR - Encourage AI tooling that respects provenance and keeps brand context, not just text. - Promote governance artifacts that emphasise user trust, not purely data volume. --- ## 🧪 Hypotheses - H1: Sites that maintain strong visual identity and narrative voice will be recommended more often by AI agents over time than visually generic sites with similar content. - H2: Users will report higher trust in agent summaries that preserve brand cues (names, design hints, tone) compared with purely neutral summaries. - H3: WordPress sites designed with clear taste (layout, spacing, restraint) will have better engagement metrics even when discovered via AI agents. --- ### 📊 Methodology - Compare AI recommendations and click‑through rates between visually rich vs generic sites with matched topics and authority. - Run user studies where participants see different styles of agent summaries (flat vs brand‑aware) and measure perceived trust and recall. - A/B test WordPress themes on similar content to measure time on page, return visits, and sharing. --- ### ❓ Questions - How can AI models be trained to preserve style and brand without hallucinating? - What minimal cues (logo, colour, phrasing) are enough for agents to “carry the performance”? - How should WordPress tooling evolve so creators don’t feel forced to publish special AI‑only feeds? --- ### 🔮 Future Research - Techniques for encoding layout and pacing into representations usable by LLMs. - Agent UX patterns that show users which brand a summary is speaking for. - Longitudinal studies on how agent‑mediated discovery shifts what “brand” means online. --- ### 💡 Implications - Practice: designers and marketers should treat AI as another channel that must carry brand, not as a reason to flatten it. - Policy: platforms and AI vendors may need guidelines for preserving attribution and style. - Education: creators should learn about taste, typography, and narrative as durable skills, not just SEO or prompt tricks. - Funding: support tools and experiments that model style and trust, not just text ingestion. --- ### 🧠 Learning Styles Assessment - Reading level: roughly upper‑high‑school to early college (complex metaphors, abstract reasoning). #### 👀 Visual learners - Present: strong imagery (Mozart, beige Volvos, grocery receipts) but no diagrams. - Improve: add before/after screenshots of sites, and annotated layouts showing “taste.” #### 🔊 Aural learners - Present: rhythmic, speech‑like prose. - Improve: 3–5 minute narrated audio version or podcast segment. #### 🗣️ Verbal learners - Present: metaphors and clear definitions of “taste” and “presentation layer.” - Improve: small glossary for key metaphors and terms. #### 🧩 Kinesthetic learners - Present: mostly conceptual. - Improve: hands‑on exercises redesigning a bland page into a “tasty” one. #### 🧠 Logical learners - Present: clear argument chain from Markdown → stripping → loss of brand → agentic future. - Improve: simple flowchart or table summarising “skeleton vs music” mapping. #### 🧑‍🤝‍🧑 Social learners - Present: shareable thesis for discussion. - Improve: guided prompts for groups to critique each other’s sites for taste. #### 🧘‍♀️ Solitary learners - Present: essay format works well for reflection. - Improve: journaling prompts about one’s own site and brand. #### 🌿 Naturalistic learners - Present: indirect (analogy to cooking for robots). - Improve: extend metaphors about ecosystems, niches, and diversity of “species” (brands). --- ### 📊 Accessibility Assessment #### Screen reader semantics - Present: depends on host (Twitter, Reader); core essay structure is simple prose. - Improve: ensure headings and landmarks if republished on a site; add alt text for any visuals. #### Keyboard navigation - Present: standard web/app behaviour. - Improve: if on a blog, ensure logical tab order and visible focus indicators. #### Contrast and readability - Present: text‑only; contrast depends on theme. - Improve: confirm at least 4.5:1 contrast and comfortable line length/height. #### Text resizing and reflow - Present: generally responsive. - Improve: test at 200% zoom and with reduced motion settings. #### Media and captions - Present: text transcript only. - Improve: if turned into video/audio, provide captions and downloadable transcript. #### Forms and interactions - Present: n/a for the essay itself. - Improve: ensure any related forms meet label and error‑handling best practices. #### Standards reference - Map any gaps to WCAG 2.2 AA, especially 1.4.x, 2.1.x, and 3.3.x. --- ### 🌍 DEIB Assessment #### Cultural breadth - Present: largely Western metaphors (Mozart, Volvos, municipal parking). - Improve: add examples from broader cultural contexts and multilingual web. #### Inclusive language - Present: generally neutral; some playful sarcasm. - Improve: avoid framing people as “municipal parking authority” stereotypes. #### Neurodiversity support - Present: strong structure, repeated metaphors. - Improve: more explicit summaries and section headings to reduce cognitive load. #### Representation - Present: no explicit people examples beyond public figures. - Improve: when adding visuals, use diverse imagery with thoughtful alt text. #### Psychological safety - Present: no sensitive topics, mainly critique of practices/ideas. - Improve: maintain respectful tone towards different skill levels and roles. --- ### ✅ Logical Validity #### Strong reasoning - Uses clear analogies (sheet music vs performance) to distinguish data from experience, supporting the claim that Markdown ≠ brand. - Causal chain: over‑optimising for AI → stripping presentation → weaker brand in an agentic future is plausible and coherent. #### Fallacies - Mild slippery‑slope risk (from “publishing Markdown” to “we are stripping ourselves”), but mitigated by concrete examples; could be tempered by data/examples. #### Tone of voice - Witty, metaphor‑rich, slightly provocative (“beige Volvo,” “cooking for robots”) but ultimately constructive, ending with positive guidance. --- ### 🛡️ DARVO Analysis - Deny: no denial of harm; instead, acknowledgment that Markdown is useful but limited. - Attack: critique is aimed at practices, not individuals. - Reverse victim/offender: none; AI engines and site owners are not cast as victims, just as making a strategic mistake. --- ## 🎯 Conclusion Publishing “clean Markdown for AI” misses the point: machines already read the skeleton, while humans crave the music—taste, style, and trust. In an agent‑mediated web, information is cheap and brand is rare. The practical next step is to design and write for human experience first, ensuring that whatever AI reads still carries your performance, not just your notes. --- ## 🧱 Refactor Appendix ```markdown ### 📚 Literature Split Pad - Purpose: Stage Konik CEE units that become per-source Literature notes. - Split level: **H3**. - Include verbatim quotes under 25 words and a [view highlight] link. ### [H1] 5–10-word present-tense claim > [!quote] [Stop Publishing Markdown for AI. Do This Instead.](https://twitter.com/pootlepress/status/2026195442535456880/?rw_tt_thread=True) by [[Jamie Marsland - Head of WordPress YouTube ❤️]] via [[Twitter]]. Published 2026-02-24. > “We do not need better notes. We need a new score.” Two lines: context and use. Cite data if relevant. [view highlight](https://twitter.com/pootlepress/status/2026195442535456880/?rw_tt_thread=True#H1) #xref #articleseed ID:rw_[MISSING] ### [H2] 5–10-word present-tense claim > [!quote] [Stop Publishing Markdown for AI. Do This Instead.](https://twitter.com/pootlepress/status/2026195442535456880/?rw_tt_thread=True) > “In an agentic world, information becomes cheap.” Two lines: context and use. [view highlight](https://twitter.com/pootlepress/status/2026195442535456880/?rw_tt_thread=True#H2) #xref ID:rw_[MISSING] --- ### 🔖 Atomic Split Pad (Zettel Candidates) - Purpose: Promote reusable claims into Atomic notes. - Split level: **H4**. - Each item: one idea, present tense, specific, testable. #### [Z1] 5–10-word present-tense claim - Scope/condition in one line. - Pointer to page/time or dataset. - Links: [[Parent-Concept]] [[Related-Idea]] #### [Z2] 5–10-word present-tense claim - Scope/condition in one line. - Pointer to page/time or dataset. - Links: [[Parent-Concept]] [[Related-Idea]] --- ### ✏️ Fleeting Queue (Extract by Selection) - Purpose: Capture quick actions or sparks for Fleeting notes. - Extraction: **Select one line** → “Refactor: Extract selection to new note (content only).” - Keep each line short and self-contained. - [F1] Short action — context/https://twitter.com/pootlepress/status/2026195442535456880/?rw_tt_thread=True #fleeting #inbox - [F2] Short action — context/https://twitter.com/pootlepress/status/2026195442535456880/?rw_tt_thread=True #fleeting #inbox - [F3] Short action — context/https://twitter.com/pootlepress/status/2026195442535456880/?rw_tt_thread=True #fleeting #inbox ---