Discourse is Agent Ready: Here’s How

The audience for forums is changing. Some of your readers aren't actually readers anymore - not in the traditional sense. They're agents reading on someone's behalf, summarizing your content into an answer for a person who might never click through or become an actual member. Whether you run a developer support community, a customer forum, or a fan club, your knowledge is being pulled into AI answers right now.

The good news: if you're on Discourse, you're already in good shape.

From links to citations

Last year we wrote about how community content wins in AI search. The short version: when AI summaries appear on Google, traditional links get clicked about half as often. Discovery is moving from "type keywords, click around" to "state the problem, get a referenced solution."

When clicks are vanishing, what matters most is being the answer the AI cites in the first place.

So what gets cited?

Models trained to detect signals will always prefer clear questions, credible answers, marked solutions, timestamps, multiple perspectives, and visible expertise; in fact, they prefer forums.

Why Discourse fits the agentic web

Three things matter when an AI agent wants to use your community as a source:

1. Can it find the content,
2. Can it parse the content, and...
3. Can it talk to your platform directly?

Discourse has answers for all three.

Every Discourse URL is a JSON endpoint. Add .json to any topic, category, or user URL and you get clean structured data. A topic at /t/some-topic/123 is also available at /t/some-topic/123.json with posts, authorship, timestamps, accepted solutions, and category metadata.

The classic web standards still matter, and we still ship them. From sitemaps, to robots.txt, and from HTML rendering for crawlers that don't run JavaScript, to schema markup to canonical URLs, we both build to and embrace web standards. These are the building blocks for the SEO substrate that AI search is built on; and studies already suggest the vast majority of citations in AI Overviews come from pages already ranking in the top 10 organic results. If Google can't find you, neither can Google’s Gemini.

llms.txt, if you want it. The proposed llms.txt standard gives site owners a way to publish a curated, AI-friendly map of their content. Admins can add an llms.txt file pointing to their best topics, knowledge categories, and canonical solutions, and AI clients that respect it will use that as an entry point. It's optional. Some communities want every model to read everything, and others want to be more deliberate and careful; on Discourse, either approach works.

Finally: the Discourse MCP. Last October we shipped the Discourse MCP CLI, built on Anthropic's open Model Context Protocol. MCP is a way for AI assistants to talk to live data sources directly, with permissions and rate limits, instead of relying on whatever happened to be in their training data. Once configured, Claude or ChatGPT or a local model in LM Studio can search your forum, read topics, and synthesize answers from conversations happening today.

Pair Discourse MCP with Claude and your support team has a research assistant that already knows every solved thread on your platform. Pair it with Gemini and a community manager can ask "what are people stuck on this week?" and get a real answer pulled from real posts. We use it internally, and we’ve already seen how it changes what an enterprise community can be.

The format does most of the work

Setting the technology aside for a moment; the reason Discourse communities punch above their weight in AI citations is our format.

Every solved thread is a Q&A pair: the question in the title, an attempt at a solution, and follow-up replies that test it against real situations. Tags cluster topics into hubs, categories provide structure and trust levels signal expertise. The Solved plugin marks the canonical answer, while edit history keeps accountability visible.

Discourse content is exactly what models are looking for when they search the web - because every forum thread is a tiny landing page for a single problem, with multiple human perspectives and a clear resolution.

What to actually do

If you run a Discourse community and want to be ready for the next year of AI search, the to-do list is short:

Keep your sitemap and robots.txt working. Don't block AI crawlers reflexively. Read our recent piece on the AI knowledge layer before deciding what access you want to grant. Mark solutions on solved threads, use tags consistently and keep titles descriptive - they're often the only thing an AI model considers before deciding whether to cite a topic. Make sure your community guidelines encourage real, attributable answers from real people - it’s what both AI and humans prefer.

If you're running a developer or support community, install the Discourse MCP CLI (npm install -g @discourse/mcp@latest) and hook it up to whatever assistant your team uses. Some of the most interesting things being built with Discourse communities right now start with that one step.

Built for the long haul

We've been writing publicly about data portability, open source, and permanent searchable knowledge for years. None of it was a bet on AI specifically; it was a bet that the web works better when communities own their content and structure it for both consumption and longevity.

That bet is paying off in a shape we didn't entirely predict. Forums are turning into the most-cited part of the open internet, and Discourse communities are showing up in AI answers because the underlying architecture was already correct.

If the agentic future is here, your Discourse community has everything it needs to be ready. If you want to opt out, the options are yours - and we have your back either way.