The Third Audience 🔗 **Source:** [The Third Audience](https://dri.es/the-third-audience) > [!note]+ Document Notes > tags: [ai, seo, web-development] > author: Dries Buytaert > source: dri.es > reference: [https://dri.es/the-third-audience] > date created: 2026-01-16 00:00 > > # 🏷️ Title: The Third Audience > <!-- deep-read:sections start --> ## 🧭 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. - Dries exposed Markdown versions of his blog posts via content negotiation and `.md` URLs, then hinted their existence in HTML. - Why it matters. - Within an hour, major AI crawlers aggressively switched to the Markdown, showing there is now a “third audience”: AI agents. - Key stakeholders. - Web publishers, CMS/platform maintainers (e.g., Drupal/WordPress), AI companies, and SEO/GEO practitioners. - Risks and tradeoffs. - Cleaner content aids models but may further reduce human visits and weaken the creator–AI value exchange. - Bottom line. - It’s technically easy to optimize for AI agents, but the long‑term incentives and fairness for creators remain unresolved. ## 📖 Main Content - Explain the core arguments in plain language. - Historically, sites served humans and search engines. With AI agents consuming content to power answer engines, there is now a third audience. - Dries’ blog content is authored in Markdown, but traditionally served as HTML that includes layout, navigation, and other chrome. AI crawlers had to sift through this noise. - By adding HTTP content negotiation (`Accept: text/markdown`) and `.md` endpoints for every post, the site can serve clean Markdown with metadata. - A small HTML `<link rel="alternate">` tag advertises the Markdown version (“Markdown auto‑discovery”), similar to RSS auto‑discovery. - Once deployed, bots like ClaudeBot, GPTBot, and OpenAI’s SearchBot quickly discovered and started requesting the Markdown instead of HTML. - This suggests AI agents strongly prefer structured, noise‑free formats, and will adopt them immediately when available. - However, it’s unclear whether this helps the site owner (e.g., more citations in AI answers) or just helps AI companies extract value without sending traffic back. - Dries acknowledges the ethical tension: humans are improving machine access while machines teach humans not to visit the original sites. - He frames this as an experiment he’ll continue, despite unresolved questions about fairness and the future of the open web. - Add sourced context and define terms. - SEO (Search Engine Optimization): practices to rank higher in traditional search result pages. - GEO (Generative Engine Optimization) / AEO (Answer Engine Optimization): emerging practices to be included or cited in AI‑generated answers rather than just link lists. - Content negotiation: HTTP mechanism where clients specify formats they accept; servers select an appropriate representation (e.g., HTML vs Markdown). - RSS auto‑discovery: pattern where HTML pages advertise their RSS feeds using `<link rel="alternate" type="application/rss+xml" ...>`. - Note uncertainties or contested claims. - Impact on traffic and visibility from being more AI‑friendly is unknown and may vary by domain. - Long‑term economic models for compensating content creators in AI ecosystems are unsettled. - Whether adopting GEO/AEO practices early yields durable advantage is speculative. ### 🗒️ Notes - Dries’ blog posts are stored as Markdown in Drupal but historically served as HTML to everyone, including bots. - Problem: AI crawlers had to parse through navigation menus and wrapper `<div>` elements to find the actual article copy. - Simple solution: - Add content negotiation so that an HTTP `Accept: text/markdown` header returns raw Markdown instead of rendered HTML. - Enable `.md` suffix on any post URL (e.g., `https://dri.es/principles-for-life.md`) that returns Markdown plus metadata (title, date, tags). - Discovery mechanism: - Add an HTML `<link rel="alternate" ...>` pointing to the `.md` URL for each page, mirroring RSS auto‑discovery. - This “Markdown auto‑discovery” allows bots to parse one HTML response, see the alternate link, then switch to Markdown for future fetches. - Observation: - Within about an hour, hundreds of requests from AI crawlers (ClaudeBot, GPTBot, OpenAI SearchBot, others) started hitting the Markdown endpoints. - This implies bots actively parse `<link rel="alternate">` and quickly adopt cleaner feeds when offered. - Conceptual framing: - First audience: humans (UX, readability, design). - Second audience: search engines (SEO, structured data, sitemaps). - Third audience: AI agents (LLMs, answer engines) that ingest structured content at scale. - Strategic trend: - From SEO to Generative Engine Optimization (GEO) and Answer Engine Optimization (AEO): shifting from ranking in SERPs to being cited in AI answers. - Making AI‑friendly representations (Markdown plus metadata) is one possible GEO/AEO tactic. - Ethical and ecosystem concerns: - More machine‑readable content could accelerate the shift from open web browsing to AI‑mediated answers. - The economic “value exchange” is unbalanced: creators invest in content and formatting; AI companies benefit disproportionately. - Dries is uneasy but chooses to experiment rather than disengage, planning to monitor and report future results. ## 📚 References - Dries Buytaert, “The Third Audience” – https://dri.es/the-third-audience - Drupal (CMS used for the blog) – https://www.drupal.org - On RSS auto-discovery pattern – https://www.rssboard.org/rss-autodiscovery - Background on Answer Engine Optimization (AEO) – https://moz.com/blog/answer-engine-optimization ## 🗣️ Paraphrase - Paragraph A paraphrase. - The author enabled his Drupal‑powered blog to serve Markdown versions of every page via content negotiation and `.md` URLs, then advertised those in HTML using an `alternate` link. Major AI crawlers rapidly switched to fetching the Markdown, revealing how eager AI agents are for structured, noise‑free content and framing them as a distinct “third audience” alongside humans and search engines. - Paragraph B paraphrase. - While the implementation was technically simple, the broader consequences are complex. Cleaner feeds might increase citation in AI‑generated answers but could also help AI companies exploit content without returning traffic, accelerating the erosion of the open web’s value for creators. The author acknowledges this tension and treats the change as an ongoing experiment with uncertain outcomes. ## 📰 Newsletter AI isn’t just visiting your WordPress sites—it’s reading them as a first‑class audience. Inspired by Dries Buytaert’s latest post, “The Third Audience,” this week we’re looking at how simple tricks like content negotiation and Markdown (or JSON) alternates can make your posts radically more legible to AI crawlers. For WordPress, think REST API, headless setups, and feed hygiene as emerging “Generative Engine Optimization.” If you’re experimenting with GEO/AEO, share your plugins, patterns, and pitfalls—we’ll feature the most useful recipes in an upcoming issue. ## ✅ Action Items #### DevRel - Document patterns for AI‑friendly representations (Markdown/JSON, alt feeds) across example apps, including WordPress. - Run workshops on GEO/AEO basics: how answer engines differ from classic SEO and how to measure impact. - Publish sample code for content negotiation and `rel="alternate"` discovery in popular stacks. - Create a campaign around “designing for the third audience,” with talks and demos at web and OSS events. - Survey developers on concerns about AI traffic vs. human traffic to inform guidance. #### OSPO - Draft guidance on exposing machine‑readable content while preserving attribution and license requirements. - Ensure scanning and license checks cover feed/alt endpoints used by crawlers. - Coordinate with security on bot traffic patterns, rate‑limiting, and abuse prevention. - Maintain inventories of machine‑readable endpoints (feeds, APIs, `.md`) across projects. - Explore contribution to standards or MoUs around fair AI use and attribution for open‑source docs. #### Community - Offer maintainers templates for “third audience” support: alt links, feeds, and structured metadata. - Host contributor sprints to add AI‑friendly endpoints to project docs and blogs. - Provide sponsors with transparent reporting on AI‑related traffic and how it affects value. - Coordinate with adjacent ecosystems for shared incident drills if bots overload infrastructure. #### WordPress - Core: Evaluate whether core should expose clearer alternate representations or improve existing feeds for AI consumption. - Core Performance: Track performance impacts of increased bot traffic on `.json`/`.feed`/`.md`‑style endpoints. - Core AI: Define responsible limits and expectations for AI crawlers accessing .org properties. - **Core Program Team:** Align roadmap items (REST API, feeds, performance) around “third audience” needs and risks. - Mobile: Ensure apps handle content originating from AI‑optimized sources consistently. - Playground: Provide labs showing how AI crawlers might interact with different WordPress configurations. - Community: Add sessions on GEO/AEO and AI‑aware content strategy to meetups and WordCamps. - Training: Produce lessons on exposing structured formats and evaluating AI‑mediated traffic. - Marketing: Shape narratives on “WordPress in an AI‑mediated web” with balanced optimism and caution. - Support: Track support tickets related to bot traffic, performance, and AI scraping. - Design: Explore UI patterns to disclose AI‑specific endpoints and opt‑outs to site owners. - Documentation: Add sections on “Designing for the third audience” with concrete code examples. - Photos/TV/Openverse: Clarify AI‑usage expectations and machine‑readable licensing/meta. - Polyglots: Ensure language‑specific feeds and endpoints remain machine‑readable and consistent. - Accessibility: Keep AI‑oriented changes from degrading human accessibility (e.g., alt text, semantics). - Meta: Consider how wordpress.org exposes structured content to crawlers and AI tools. - CLI: Add commands or scaffolds for enabling AI‑friendly feeds or negotiation. - Hosting/Tide/Openverse/Themes/Plugins: Coordinate best practices and guardrails for AI‑oriented scraping and scoring. #### The WPOCC - Design fellowships and microgrants around researching and prototyping AI‑friendly yet creator‑respecting content formats. - Fund documentation and governance experiments on fair use, attribution, and AI traffic. - Pilot transparent reporting on AI crawler activity and its impact on funded projects. ### FAIR - Promote local logs and metadata to understand which AI agents are accessing which content. - Encourage funding transparency from AI consumers and clear policies about acceptable crawler behavior. - Package charters and operator agreements that embed AI‑era rights (attribution, rate limits, opt‑outs). ## 🧪 Hypotheses - Sites that expose Markdown (or similar structured formats) via `rel="alternate"` will see rapid and sustained adoption by major AI crawlers. - Providing cleaner machine‑readable content increases the likelihood of being cited in AI‑generated answers. - Increased AI crawler consumption does not necessarily correlate with increased human referral traffic. - Implementing content negotiation and alt endpoints can be done in most CMSs with minimal engineering effort. - Early adopters of GEO/AEO practices may enjoy temporary visibility advantages that diminish as practices become widespread. ### 📊 Methodology - Study designs, data sources, and analyses to test the hypotheses. - Log Analysis: Compare crawler behavior (frequency, endpoints used, user agents) before and after introducing `.md` or equivalent endpoints. - Citation Tracking: Use prompt‑based queries across popular AI systems to track mention frequency and attribution patterns for a set of experimental sites. - Traffic Impact: Measure changes in human referral traffic from AI‑associated domains or features over time, controlling for seasonality. - A/B Experiments: Randomly enable alt endpoints on a subset of similar pages or domains and compare crawler adoption and answer‑engine citations. - Implementation Survey: Collect data from CMS maintainers on time/complexity to add content negotiation and alternate links. ### ❓ Questions - Does making content more AI‑friendly ever produce meaningful, measurable benefits for small and medium publishers? - What technical standards (if any) will emerge for AI‑optimized representations beyond ad‑hoc Markdown/JSON feeds? - How can creators negotiate fair attribution and compensation when AI uses their structured content at scale? - Should CMSs enable “third audience” features by default, and how should they expose configuration and opt‑out options? ### 🔮 Future Research - Longitudinal studies on the relationship between AI crawler adoption of alt formats and publisher revenue/traffic trends. - Comparative analysis of different structured formats (Markdown vs JSON vs custom schemas) for GEO/AEO effectiveness. - Governance models for AI crawlers that balance open access with rate‑limiting, robots policies, and creator rights. - UX research on how end users perceive answers that do or don’t visibly credit original sources. ### 💡 Implications - Practice: Developers can trivially expose AI‑friendly endpoints, but should pair that with monitoring and clear policies. - Policy: Regulators and industry bodies may need to set norms on AI crawling, attribution, and compensation. - Security: Increased bot traffic to structured endpoints could create DDoS or scraping risks if unmanaged. - Education: Content strategists and engineers need training on AI audiences, not just human UX and classic SEO. - Funding models: New mechanisms may be required to reward creators whose content heavily trains or powers AI systems. ### 🧠 Learning Styles Assessment - Reading level estimate: Roughly upper high‑school to early college (Flesch‑Kincaid around 11–13) due to technical terms and abstract concepts. #### 👀 Visual learners - Present: Conceptual contrast between three audiences (humans, search engines, AI agents). - Improve: Diagrams of request flows (HTML vs Markdown), architecture of auto‑discovery, before/after content examples. #### 🔊 Aural learners - Present: None mentioned. - Improve: Short audio summary or podcast segment explaining the experiment and findings. #### 🗣️ Verbal learners - Present: Clear narrative, analogies to SEO and RSS auto‑discovery, defined new terms (GEO, AEO, “third audience”). - Improve: Glossary of terms and more explicit comparisons to traditional SEO practices. #### 🧩 Kinesthetic learners - Present: Implied steps (configure content negotiation, add `.md` URLs, add `<link rel="alternate">`). - Improve: Step‑by‑step tutorial or lab with code snippets and exercises. #### 🧠 Logical learners - Present: Cause‑and‑effect reasoning (change → crawler behavior), structured argument about incentives. - Improve: Tables comparing formats, decision trees on when to adopt AI‑friendly endpoints. #### 🧑‍🤝‍🧑 Social learners - Present: Framed as a community‑relevant experiment with shared concerns about the web’s future. - Improve: Discussion prompts, forum threads, and Q&A sessions about GEO/AEO ethics and tactics. #### 🧘‍♀️ Solitary learners - Present: Reflective sections on the ethics and personal discomfort with the value exchange. - Improve: Explicit reflection questions about whether to optimize for AI audiences. #### 🌿 Naturalistic learners - Present: None explicitly. - Improve: Analogies between web ecosystems and natural ecosystems (e.g., resource extraction, symbiosis, collapse). ### 📊 Accessibility Assessment #### Screen reader semantics - Present: Standard HTML pages likely have basic structure; Markdown representations are clean and linear. - Improve: Ensure headings, landmarks, and alt text remain robust even as AI‑oriented changes roll out. #### Keyboard navigation - Present: Core HTML presumably supports basic tabbing. - Improve: Confirm focus states, skip links, and no keyboard traps independent of AI‑focused optimizations. #### Contrast and readability - Present: Not discussed. - Improve: Check contrast ratios and line height; ensure readable typography while experimenting with new audiences. #### Text resizing and reflow - Present: Likely responsive layout. - Improve: Test at 200% zoom and with user preferences like reduced motion; ensure alt endpoints don’t bypass human‑friendly layouts. #### Media and captions - Present: Mainly text‑based content. - Improve: Add captions/transcripts for any future audio/video explanations of the experiment. #### Forms and interactions - Present: Minimal interactive elements in this article. - Improve: Any added controls for AI settings (opt‑outs, toggles) should have clear labels and error handling. #### Standards reference - Map gaps to WCAG 2.2 AA criteria. - Focus: 1.4.x contrast, 1.4.10 reflow, 2.1.x keyboard, 2.4.x navigation, and 3.3.x input assistance. ### 🌍 DEIB Assessment #### Cultural breadth - Present: Topic is globally relevant but centered on a single personal site. - Improve: Incorporate examples from diverse regions and languages facing different AI/web power dynamics. #### Inclusive language - Present: Neutral, non‑derogatory tone; minimal jargon, reasonably explained. - Improve: Add plain‑language summaries and define all acronyms on first use. #### Neurodiversity support - Present: Clear structure and short sections. - Improve: More consistent subheadings, summaries, and visual cues for key concepts. #### Representation - Present: Single‑author, first‑person narrative. - Improve: Include perspectives from smaller publishers, marginalized creators, and non‑English communities. #### Psychological safety - Present: Respectful framing; acknowledges uncertainty without alarmism. - Improve: Provide guidance on constructive responses (e.g., experimentation, advocacy) rather than just raising concerns. ### ✅ Logical Validity #### Strong reasoning - The article uses causal reasoning: enabling Markdown and auto‑discovery leads to immediate crawler behavior change, supported by logs and timing. - The analogy to RSS auto‑discovery is apt and helps explain why the pattern works for bots. - The author clearly distinguishes observed facts (hundreds of requests) from speculation (potential visibility benefits). #### Fallacies - No major explicit fallacies; some extrapolation about “hollowing out of the web” is speculative but framed as concern, not certainty. - To strengthen, the piece could reference data or prior studies on traffic displacement by AI answer engines. #### Tone of voice - Reflective, experimental, and cautiously skeptical. - Uses first‑person narrative to humanize a technical change and admit moral ambiguity. ### 🛡️ DARVO Analysis - Deny: - The author does not deny harms; he explicitly acknowledges potential negative impacts on the web and creators. - Attack: - There is no attack on critics or other stakeholders; AI companies are mentioned but not vilified. - Reverse victim/offender: - No attempt to cast himself or publishers as victims to excuse harmful behavior. Instead, he positions himself as a cautious experimenter aware of systemic risks. - Overall: - The piece is exploratory rather than defensive; DARVO patterns are not present. ## 🎯 Conclusion - Restate the bottom line. - A small technical tweak—serving Markdown via content negotiation and announcing it through `rel="alternate"`—was enough for AI crawlers to rapidly adopt a cleaner representation of Dries’s blog, underscoring that AI agents are now a distinct “third audience” for the web. The change is easy to implement but raises unresolved questions about how much it benefits creators versus AI platforms and what it means for the future of the open web. - Note practical next steps for readers. - Audit how your own site appears to AI crawlers, experiment with structured formats and discovery mechanisms, and track both technical outcomes (crawler behavior) and business impacts (traffic, attribution) while staying engaged in broader conversations about fair AI use. ## 🧱 Refactor Appendix (append after 🎯 Conclusion) ### 📚 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] [The Third Audience](https://dri.es/the-third-audience) by [[Dries Buytaert]] via [[dri.es]]. Published 2026-01-14. > “AI agents are now the third audience for websites.” Two lines: context and use. Cite data if relevant. [view highlight](https://dri.es/the-third-audience#H1) #xref #articleseed ID:rw_[MISSING] ### [H2] 5–10-word present-tense claim > [!quote] [The Third Audience](https://dri.es/the-third-audience) > “Markdown auto-discovery made AI crawlers switch formats quickly.” Two lines: context and use. [view highlight](https://dri.es/the-third-audience#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] AI crawlers adopt cleaner formats when discovered - When sites expose structured alternates via `rel="alternate"` and negotiation. - Pointer to article experiment logs. - Links: [[AI-Crawlers]] [[Content-Negotiation]] #### [Z2] Third web audience emerges as AI agents - When content is used in answer engines and LLMs. - Pointer to “The Third Audience” article. - Links: [[SEO-Evolution]] [[Generative-Engines]] --- ### ✏️ 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] Prototype Markdown auto-discovery for my main site — context/https://dri.es/the-third-audience #fleeting #inbox - [F2] Measure AI crawler behavior before/after alt endpoints — context/https://dri.es/the-third-audience #fleeting #inbox - [F3] Draft policy stance on AI access to structured feeds — context/https://dri.es/the-third-audience #fleeting #inbox --- <!-- deep-read:sections end -->