Is Your Website Ready for AI Agents? How to Test It Free and What to Fix First

Bots now send the majority of HTML requests on Cloudflare's network. Vercel's free Is Agentic scanner shows how your site reads to AI agents, how big brands score, and which fixes…
OutrankAI SEO platform
Updated
Reading time
5 min read
Is Your Website Ready for AI Agents?
In this article
  1. What "ready for AI agents" means
  2. How to test your site with Is Agentic
  3. How well-known sites score
  4. Which checks matter for your kind of site
  5. What to fix first
  6. Agents still need something worth reading
  7. FAQ

Your website is ready for AI agents when software can find it, read it and act on it without a human in the loop: the content is in the raw HTML, missing pages return real 404s, the structure is clean, and (if you sell software) your API is documented in a way a machine can parse. The quickest way to check is Is Agentic, a free scanner Vercel launched with Ora in August 2026. Enter a domain and it returns a score out of 100, the evidence behind every check, and a prompt you can paste into a coding agent to fix the gaps.

Here is what it checks, what well-known sites score, and which fixes are worth your time.

Key takeaways

  • Is Agentic is free with no signup. Essential checks are worth 80 points, Recommended checks 20, plus up to 5 bonus points. Checks that don't apply to your site are left out, not failed.
  • Big brands don't ace it: Vercel 96, Zapier 90, Stripe 71, Figma 65, Ahrefs 63 and Linear 57 in reports dated September to October 2026. Our own site, outrank.so, scores 69.
  • The most common gap in that group: 5 of the 7 homepages don't return a Markdown version when an agent asks for one.
  • Agents are still a small slice of AI traffic: 1.7% of AI-driven traffic by December 2025, after growing 7,851% in a year, according to HUMAN Security.
  • For a content or service site, fix the essentials (text in the raw HTML, honest 404s, clean headings, organization details) and ignore API checks you don't need.

What "ready for AI agents" means

An AI agent is software that uses the web on someone's behalf: it searches, reads pages, compares options, fills in forms and, increasingly, buys. That's one step beyond the crawlers that only read pages to train a model or answer a question.

The traffic is real but still early. Automated traffic passed human traffic for HTML page requests on Cloudflare's network in 2026, at roughly 57% by mid-year according to Cloudflare Radar data reported by Search Engine Land. Most of that is crawlers, not agents. HUMAN Security's 2026 benchmark report found that by December 2025, training crawlers made up 74% of AI-driven traffic, scrapers 24% and AI agents and agentic browsers 1.7%. That last slice grew 7,851% year over year, and 77% of it landed on product and search pages.

So agents are small, fast-growing and showing up exactly where people buy. That argues for cheap fixes now, not a rebuild.

How to test your site with Is Agentic

  1. 1

    Scan your domain

    Go to is-agentic.com and enter your domain. Developers can run npx is-agentic yourdomain.com in a terminal instead (add --json for machine-readable output). It's free, with no account or API key.
  2. 2

    Read the score and the three tiers

    Essential checks carry 80 points (server-rendered content, correct HTTP behavior, clear document structure). Recommended checks carry 20 and cover extras such as structured data, agent instructions and, when the scan finds them, your API, OAuth flow or MCP server. Bonus signals add up to 5 points and never lower your score.
  3. 3

    Open every failed or partial check

    Each one shows the evidence (what the scanner actually received) and a recommendation, often with a curl command to verify the fix yourself.
  4. 4

    Hand the fixes to a coding agent

    Click Prompt to improve to get a brief you can paste into Claude Code, Codex or Cursor.
  5. 5

    Rescan after you ship

    A report reflects one URL at one point in time, so rescan to confirm the score moved.
Is Agentic report for feather.so: score 74 out of 100, labelled "Ready with a few material gaps", with Essential at 7 of 9 checks passed, Recommended at 7 of 20 and 9 bonus signals, plus a Prompt to improve button
An Is Agentic report: the score, the three tiers and the Prompt to improve button

Is Agentic describes its report as a prioritized technical review, not a guarantee that an agent can complete a task, and security still needs its own audit. Treat it as a to-do list.

How well-known sites score

We pulled the latest stored report for a handful of well-known homepages from Is Agentic's public API. Nobody gets 100, including Vercel, which built the tool.

Is Agentic scores for well-known sites (October 2026)

Latest stored homepage report for each site. Scores change on every rescan.
vercel.com968 of 10Thin raw HTML: readable text is 1.0% of the markup2026-10-19
zapier.com907 of 10No Markdown body on 404 pages; API scopes not machine-readable2026-10-17
stripe.com715 of 10No Markdown version of the homepage; no OpenAPI spec at a standard path2026-10-19
outrank.so694 of 8No Markdown version of the homepage; headings skip from H2 to H42026-10-07
figma.com654 of 9Only 1,695 characters of text in the raw HTML; no Markdown version2026-10-15
ahrefs.com633 of 9No Markdown version of the homepage; no JSON error responses2026-09-22
linear.app574 of 10Missing pages return 200 instead of 404 (soft 404s)2026-10-19

Click a column header to sort.

Source: Is Agentic public report API, scans dated September 22 to October 19, 2026

Two patterns stand out. Five of the seven fail Markdown content negotiation, a newer check most sites haven't heard of yet. And several get partial credit on "content without JavaScript" because the readable text is a tiny share of the markup (Figma's homepage serves 1,695 characters of text in its raw HTML).

Which checks matter for your kind of site

Checks that don't apply are excluded rather than failed, so a blog isn't punished for having no API. Spend your time on the rows that fit you.

Your siteChecks worth fixingChecks you can ignore
Blog, publisher, local business, agencyContent without JavaScript, agent-friendly 404s, heading structure, metadata, Organization schema, About and Contact pagesOpenAPI, OAuth scopes, MCP server, rate-limit headers
SaaS with an APIAll of the above, plus an OpenAPI spec at a standard path, scoped OAuth, JSON error responses, an MCP server and a "when to use us" note for agentsVery little
Online storeThe content checks above, plus whatever Ora's payments checks flag on your checkoutAPI checks, unless you expose one

What to fix first

  1. Put your content in the raw HTML. The "content without JavaScript" check wants at least 500 characters of meaningful homepage text in the HTML, one clear H1 and headings in order. Most AI crawlers don't run JavaScript at all, which we cover in how AI reads your website.
  2. Return real 404s. Linear's missing URLs return a 200 status with the app shell, so an agent probing for a page concludes it exists. Missing pages should return 404.
  3. Fix your heading order. Our own homepage loses points for jumping from H2 to H4. It's a five-minute fix that also helps screen readers and search engines.
  4. Say who you are in structured data. Add Organization schema with a contact point and make sure About and Contact pages exist. Our report flags both.
  5. Offer Markdown on request. When a request arrives with Accept: text/markdown, return a Markdown version of the page with a Vary: Accept header, and keep serving HTML to browsers.
  6. If you have an API, document it. Publish an OpenAPI spec at /openapi.json, return JSON errors, declare OAuth scopes, and add an MCP server if agents are a channel you want.

Stop there. Chasing the last few points rarely pays when even Vercel sits at 96.

Agents still need something worth reading

Readiness gets an agent to your page. What it does next depends on whether your pages answer the buyer's question better than a competitor's. When we tracked what ChatGPT reads and cites across 48 questions, it opened about 36 pages per answer and cited about 4. Agents and assistants also phrase the same question many different ways, which is why one prompt test tells you little (see how one word can change AI recommendations). And if your robots.txt blocks the bots behind AI search, no score will help: check what Google-Extended and the other AI tokens control.

The content side is where Outrank fits. It finds the questions your buyers ask, writes an answer-first article for each one and publishes it to your CMS every day, so there's always a clear page for an assistant or agent to land on. To go further on getting recommended, read how to get ChatGPT to recommend your product.

FAQ

What does it mean for a website to be ready for AI agents?

It means AI software can find your pages, read them without running JavaScript, understand their structure, and complete tasks such as signing up, calling your API or buying, without a human stepping in. In practice: content in the raw HTML, correct status codes, clean headings, structured data and, for software companies, a documented API.

What is Is Agentic?

Is Agentic (is-agentic.com) is a free tool from Vercel, with scans run by Ora, that scores a public website out of 100 for agent readiness. It groups checks into Essential (80 points), Recommended (20 points) and Bonus (up to 5), shows the evidence for each, and generates a prompt to fix the issues.

Is is-agentic.com free?

Yes. The website, the read-only report API, the npx is-agentic command-line tool and its MCP server are free and need no account or API key.

What is a good Is Agentic score?

There's no official pass mark. In the reports we checked in October 2026, well-known sites ranged from 57 (Linear) to 96 (Vercel), so anything above 80 puts you ahead of most of them. Fix every failed Essential check before you worry about the total.

Do AI agents run JavaScript?

Agents that drive a full browser can, but the crawlers and fetchers behind most AI answers don't. Vercel found that OpenAI, Anthropic and Perplexity crawlers download JavaScript files without executing them, so put key content in the HTML to cover both.

Does agent readiness help SEO?

Indirectly. Most Essential checks overlap with technical SEO basics such as crawlable HTML, correct status codes and clear structure. There's no evidence the score itself affects Google rankings or AI citations, so treat it as a technical to-do list, not a ranking factor.

Share this article

About the author

Outrank

Outrank researches, writes and publishes SEO articles like this one, then builds the backlinks that help them rank. Yes, Outrank writes its own blog posts.

Show up in AI answers, not just on Google

Sites with a Domain Rating of 30+ get 29× more AI citations than sites under 10. Outrank builds the content and the authority behind both, on autopilot.

Free tool: see what ChatGPT, Perplexity, Gemini and Claude cite