--- title: "When AI Answers Become the First Recommendation: Making WordPress Agent-Ready" description: "When people interact with a web application today, there's often no browser in the loop — just an AI answer. An agent-ready website is the technical foundation for showing up in those answers. How does that work with WordPress? A practical assessment." date: 2026-04-29 modified: 2026-05-09 author: "Eberhard Lauth" url: "https://netzkundig.com/en/blog/wordpress-agent-ready/" featured_image: "https://netzkundig.com/wp-content/uploads/2026/04/agent-ready-wordpress.jpeg" categories: - name: "AI" url: "https://netzkundig.com/en/thema/ai/" - name: "Code" url: "https://netzkundig.com/en/thema/code/" - name: "WordPress" url: "https://netzkundig.com/en/thema/code/wordpress/" language: "en-US" --- # When AI Answers Become the First Recommendation: Making WordPress Agent-Ready Anyone looking for a recommendation today, comparing a product, or clarifying an industry question increasingly starts that process inside an AI application. ChatGPT alone reports around 810 million daily users ([as of March 2026](https://www.superlines.io/articles/ai-search-statistics/)). A Similarweb study for the U.S. shows that 35 percent of respondents use AI tools in the early phase of a purchase decision, while traditional search engines come in at 13.6 percent ([U.S. data, January 2026](https://www.similarweb.com/blog/marketing/geo/gen-ai-stats/)). By the time a user actually clicks through to a website, the AI answer has already made the selection of which brands to mention. Anyone who isn’t in that answer is not agent-ready, which means effectively invisible — even though their website still exists. The effect is visible in the numbers, too. According to Cloudflare Radar ([as of April 2026](https://technologychecker.io/blog/search-engine-market-share)), Google still delivers 87.5 percent of all search referrals to websites; all AI chatbots combined account for less than 0.3 percent. Reading those numbers and concluding that the shift is manageable misses two points. First, the recommendation is decided before the click happens. Second, the quality of these clicks is significantly higher: data from Conductor ([cited here](https://thedigitalbloom.com/learn/gen-ai-website-traffic-share-february-2026/)) shows that a ChatGPT session on a website lasts roughly twice as long as a classic Google click and converts three times better. For every publisher, this raises a single question: how do you make sure your own website is understood by AI systems, cited correctly, and used according to the rules you’ve defined? ## What does _agent-ready_ mean? The term “agent-ready” was popularized by Cloudflare in April 2026 with the public tool [isitagentready.com](https://isitagentready.com). It works similarly to Google Lighthouse for website performance: you enter a URL, the tool checks a series of standards, and returns a score with concrete suggestions for improvement. Four dimensions are evaluated. 1. Discoverability covers the question of whether an AI system can structurally explore the site at all. A cleanly configured robots.txt, a sitemap.xml, and link headers per RFC 8288 belong here. 2. Content describes the form in which the site delivers its texts. This is where Markdown delivery via HTTP content negotiation and the `/llms.txt` file come into play — a kind of table of contents for language models. 3. Bot access control is probably the most consequential dimension for European companies. It governs whether content may be used for traditional indexing, for real-time inclusion in AI answers, or for training models. This is handled through what are called Content Signals. 4. Capabilities goes beyond mere reading. It defines whether — and how — an agent can interact with the site or its services: skill definitions, API catalogs per RFC 9727, and MCP server cards belong in this area. Cloudflare’s inventory is sobering. In a study of 200,000 top domains, 78 percent had a `robots.txt`, but only four percent declared any AI usage preference at all, just under four percent supported Markdown negotiation, and API catalogs were found on fewer than 15 sites in the entire dataset. ([More on this here](https://blog.cloudflare.com/agent-readiness/)) Anyone who wants to test their own site can do so in one or two minutes at [isitagentready.com](https://isitagentready.com). The probability that the result will be sobering currently sits above ninety percent. ## Why investing in agent-readiness pays off now Three arguments speak for making a website agent-ready as soon as possible. 1. Visibility in a changed discovery landscape is the most obvious one. When an AI system names three or five brands in response to an industry question, the first contact falls on that selection. Anyone not appearing in the answer has already lost the contest for attention. Sites that deliver cleanly structured content to AI systems empirically have a higher probability of being cited. Addressing this early pays off over a longer period, because the memory effects in training data and indexes are cumulative. 2. Data sovereignty and legal clarity matter especially for regulated industries, B2B SaaS providers, and knowledge platforms. The old `robots.txt` logic only knows “may crawl” or “may not crawl”. Content Signals allow you to express in a differentiated way that content may appear in search results but not be used for model training, or that real-time use in answers is permitted while model training is not. In an EU context, such a machine-readable statement counts as an explicit reservation in the sense of Article 4 of EU Directive 2019/790. That shifts the burden: a site operator no longer has to prove in a dispute that no consent was given. More on this [here](https://blog.cloudflare.com/content-signals-policy/) and in the — fair warning, unnecessarily complicated language —[full text of EU Directive 2019/790, Article 4](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32019L0790). 3. The currently low adoption of agent readiness: anyone moving early in a field where only a single-digit percentage of the studied sites implement the relevant standards can secure a technical pole position and benefit from it. ## The current state of the WordPress ecosystem WordPress remains the most widely used CMS in the world, with a market share of around 43 percent ([as of April 2026](https://w3techs.com/technologies/details/cm-wordpress)). When it comes to agent-readiness, the plugin ecosystem is still trailing behind. Three approaches are currently visible. Cloudflare offers an edge solution called[ “Markdown for Agents”](https://blog.cloudflare.com/markdown-for-agents/). It’s enabled via a toggle in the Cloudflare dashboard and converts HTML to Markdown on demand. For many sites this is a reasonable minimum standard, since no changes to the origin (the WordPress installation) are required. The conversion, however, works generically and includes the entire HTML — navigation, footer, sidebar, and cookie banner along with everything else. On top of that, Cloudflare sets the Content-Signal header `ai-train=yes, search=yes, ai-input=yes` on every converted response, the maximally open default. Anyone serious about data sovereignty wants to control this granularly and individually. Joost de Valk, known as the founder of Yoast SEO, has released a first plugin called [“Markdown Alternate”](https://github.com/progressplanner/markdown-alternate). It operates inside the WordPress core, has access to the full post object, and produces a noticeably richer frontmatter with author, date, categories, and tags. That’s substantially better than the generic edge solution but only covers the Markdown delivery aspect. Content Signals, `llms.txt`, well-known endpoints, and link headers are still open. Alongside these, several smaller plugins each cover partial aspects. [“Markdown for AI Agents”](https://wordpress.org/plugins/markdown-for-ai-agents/) and [“Serve Markdown”](https://wordpress.org/plugins/serve-md/) on wordpress.org as well as various GitHub projects also focus on the Markdown variant. A [commercial offering with a license model](https://imado.co/plugins/markdown-for-agents) rounds out the picture. They all share one trait: each solves one slice of the agent-readiness matrix, but no single plugin covers the whole picture. That’s an unsatisfying situation. What’s needed is a tool that brings all four dimensions from the isitagentready.com test together in a single, well-considered solution. ## What a complete solution has to do The requirements of the four assessment dimensions translate into a concrete checklist. I also use it as an audit grid when examining the agent-readiness of an existing site. A complete solution should cover the following points: - Markdown delivery via three trigger paths: through the `Accept` header, through a `.md` URL variant, and optionally through a query parameter. - It should cleanly separate the content from the theme surroundings, converting only the main content of a post. - It should produce a WordPress-typical, rich frontmatter with title, description, author, date, categories, tags, featured image, and language. - It should let individual blocks in the editor be excluded from the Markdown variant, so that ad slots, newsletter prompts, or related-posts boxes remain visible to humans but don’t appear in the agent-facing variant. - It should set a Content Signal header on every response and add a consistent policy to robots.txt in parallel. - It should generate a `/llms.txt` as a table of contents for language models and optionally a more detailed `/llms-full.txt`. - It should provide the endpoints under `/.well-known/` that the isitagentready tool checks for: api-catalog, agent-skills, mcp.json, and ai.txt. - It should set link headers per RFC 8288 and inject a `` into every HTML document. - Finally, it should include a diagnostic tool that checks the configurations and makes the result transparent. This list is extensive, but it covers the state of the standards as of mid-2026. Anyone taking it seriously inevitably has to build a dedicated plugin, because none of the existing ones offer the full range. ## Conclusion: a custom plugin is needed That’s the reason I’m currently developing a WordPress plugin that brings every checklist item together in a modular architecture. It consists of modules for all components of AI readiness, each of which can be enabled or disabled individually. Each comes with its own settings and its own logic. A central diagnostics tab — similar to Cloudflare’s logic — examines the site and shows which building blocks are working correctly and which still need attention. The most distinctive differentiator from existing plugins is block-level control of the Markdown output. For every individual content block, the advanced settings let you decide whether it should appear in the Markdown variant. For container blocks like Group, Columns, or Cover, this flag applies to the entire contained sub-tree. Markdown rendering is therefore no longer generic and arbitrary as in Cloudflare’s edge solution. What goes to agents is a deliberately curated selection. A second distinctive differentiator is end-to-end multilingualism. The frontmatter generator detects Polylang and WPML automatically, the cache is kept separate per language version, and the settings interface and all hints are available translated. For European sites with English as a secondary language, this matters in practice — otherwise language-specific quirks show up as small but annoying blemishes in the Markdown variant. The plugin follows a filter-first design. Every output and every decision runs through a WordPress filter, so theme- and industry-specific behavior can be adjusted without touching the plugin core. If you need a particular frontmatter line, run a theme with unusual wrapper structures, or want to exclude certain content from the Markdown variant by your own rules, you can do that through your own filter hooks. ## Status as of late April 2026: test phase The plugin has been running in test mode since mid-April 2026 on several sites, including netzkundig.com itself. Results in the isitagentready test land between eighty and ninety percent depending on configuration. The individual modules have been tested repeatedly over the past weeks against real theme and plugin combinations that, in practice, tend to produce surprising side effects — Yoast SEO with dynamic breadcrumbs, GenerateBlocks query loops, and ACF blocks among them. In the coming weeks I’ll be making the toolkit available to additional selected pilot customers. If you’d like to be among them, [feel free to get in touch](#kontakt). Also in the works is a separate blog post with a detailed architecture description and code examples. ## How I can already help If you’d like to make your own WordPress site agent-ready, I offer four building blocks. An assessment examines the current state using the isitagentready test together with a manual analysis of the site structure, the existing robots.txt, the Cloudflare configuration, and the content. A configuration sets up the plugin and complementary components to fit the site, its theme, and its industry requirements. A **customization** handles edge cases — heavily individualized themes, multilingual setups, or specific data protection requirements. An ongoing engagement over several months ensures that newly emerging standards like Web Bot Auth, MCP server cards, or Agentic Commerce can be adopted directly, without having to re-examine the site each time. My focus is on enterprise sites with substance: regulated industries, B2B SaaS providers, knowledge platforms, editorial sites with a long-term content strategy. For these use cases the effort pays off most clearly, because both visibility and data sovereignty are business-relevant. And because even minimal shifts in traffic can have a large effect that’s worth avoiding. ## Two (simple) next steps For a quick read on your own site, two things are worth doing: The first step is to run your own domain through [isitagentready.com](https://isitagentready.com). The result delivers a clear reference point on your current position in two minutes. The second step is to test the Markdown variant of this very post. [You can view it here](https://netzkundig.com/en/blog/wordpress-agent-ready/?format=markdown) and see what a cleanly curated agent answer looks like. If you’d like to bring your own site up to this standard, or you’re looking for an initial analysis, [all contact details are here](https://netzkundig.com/en/contact/). One closing note: Markdown negotiation, Content Signals, and the other standards are preferences, not technical barriers. They tell an AI agent what is wanted and what isn’t, but they cannot prevent actual usage in the way a firewall would. Anyone keeping that in mind judges the impact of these measures realistically. They’re a lever, not a cure-all, and that’s exactly the role in which they’re worth the effort. They’re probably also never finished, because since the broad availability of generative AI the digital world has been changing yet another tick faster than we were used to before.