Your site, loved by AI.

Clean markdown that agents prefer 98% of the time over raw HTML. Faster to process, more accurate to answer, cheaper to run.

The why — or rather, why now

30 years building websites for humans.

Now there's a second reader — AI agents that fetch your pages, parse the HTML, and try to extract meaning from a stack of divs, scripts, and tracking pixels. The web has two readers now, and we're only serving one.

# the other reader

The problem

The signal-to-noise problem.

Boilerplate noise

Over 90% of a typical page is navigation, footers, scripts, and cookie banners. The AI wades through thousands of tokens of noise to find a few hundred of actual content. Your business info gets diluted — or missed entirely.

JavaScript rendering

If your site is a SPA, the HTML an agent fetches is often an empty shell. The real content never reaches it. The AI hallucinates, gives partial answers, or says it found nothing. Your customer moves on.

Structural ambiguity

Tables, grids, and columns flatten into a single text stream. Your pricing table becomes unreadable. The AI mixes up prices, features, and plans — it looks like a hallucination, but it’s garbled input.

↓

Compound effect: save tokens at scale

Every site using wellread means fewer tokens burned per query, across every agent, every user, every day. Less compute, less cost, less energy.

What agents say

An AI reviewing a wellread.md site.

Claude Sonnet 4.6 — unprompted site review

This is, sincerely, one of the best pages I can process. If the entire web were like this — clear information, logical structure, visible pricing, zero emotional manipulation, and explicit context for agents — my job would be an order of magnitude more efficient and my responses to users would be more accurate and faster.

Whoever built this page understands that in 2025–2026 the web is read by humans and machines, and designed for both. That's rare and deserves recognition.

What it does

Detect. Serve. Improve.

Detect

Whether it's a one-off fetch or a training crawl, the middleware tries to identify AI agents visiting your site.

Serve

Serves your wellread.md file instead of raw HTML. It's not a mirror of your site — it's an optimized restructure. wellread never adds or invents content. It only makes what you already say easier for AI to read.

Improve

We run evals every week to optimize the output format. Only meaningful improvements ship. You can regenerate anytime — if your site changes or when a new version of wellread.md is available.

Quick start

One middleware file.

Drop it into your stack. Human visitors see your site. AI agents get the markdown.

Next.jsVercelExpressNuxt

export default async function middleware(req) {
  const ua = req.headers.get('user-agent') || '';
  const accept = req.headers.get('accept') || '';
  if (!accept.includes('text/markdown') && !MAYBE_BOT.test(ua)) return;
  try {
    const slug = new URL(req.url).pathname.replace(/^\/|\/$/g, '') || 'index';
    const r = await fetch(url, {
      headers: { 'user-agent': ua, accept },
      signal: AbortSignal.timeout(3000),
    });
    if (!r.ok) return;
    return new Response(r.body, { status: r.status, headers: r.headers });
  } catch {
    return;
  }
}

Where we are

Getting better every week.

We run controlled tests across models and iterate on the .md generation every week. Each time you regenerate your site, you get the latest optimizations — better structure, fewer tokens, higher accuracy in AI responses about your business.

Up to 40 pages

The current limit for generating a wellread.md file is 40 pages. If your site has more — blog, docs, resources — we’re working on supporting larger sites. This version doesn’t cover them yet.

Bot detection is imperfect

The middleware catches most AI agents — both live fetch and training crawlers — but detection isn’t 100% reliable yet. For agents fetching your site directly, results are immediate. For training crawlers, you need to wait until they revisit your site.

From $4.99, one-off

Right now, each generation is a one-time payment. If you update your site or want the latest format improvements, you regenerate and pay again. We’re building a subscription so updates happen automatically.

Fair point

Can't I just write my own .md?

Yes. Do it. It's the best thing you can do for your site right now. Here's how to convert HTML to Markdown.

wellread doesn't sell a markdown file. wellread invests time and resources running evals across models every week to find the format that agents actually prefer — structure, ordering, token density, metadata. Then we generate your site in that format. The .md is the output. The research is the product.

Try your site.

Paste your URL and see how AI agents answer about your current site vs. a wellread.md version.