The 10 Best AI Agent Readiness Scanners in 2026, Tested on One URL

Verified 29 August 2026. Every score below was captured from the tool's own public endpoint on that date, and every disagreement was adjudicated by hand against the live site.

A category of tool has appeared over the past eighteen months that grades how well a website can be read and used by AI agents. There are now at least ten of them. Nobody had run the same URL through all of them and published what came back, so we did.

The result: five of the ten produced a score for https://stripe.com. Those scores were 65, 62, 41, 38 and "Level 1". Two of the ten returned an answer that was factually wrong about the site, and one of the two wrong answers was ours.

Quick summary: which scanner to use, and for what

Start with the free scan, then read your robots.txt yourself. No scanner in this group agreed with any other on all three of the disagreements we found. The score is a starting point for a conversation with your own site, not a verdict on it.

The single most useful finding: a scanner that reports a missing file has usually checked one path. Stripe publishes an Agent Skills index. Two scanners found it, two reported a 404, and the file is real.

How we tested

The criteria came first, before any tool was run, and they were chosen so that a tool could win on them without us winning:

1. Can it be run without an account? A scanner behind an email form is a lead magnet with a scanner attached.
2. Does it show its evidence? A number with no fetch log cannot be checked or argued with.
3. Does it publish its rubric or weights? A hidden rubric is an opinion presented as a measurement.
4. Does it tell you how to fix what it found? Detection without remediation moves the work, it does not do it.
5. Was it right? Where two tools disagreed, we fetched the underlying artifact by hand and settled it.

The test subject was https://stripe.com, picked because it is large, well-engineered, publicly examined, and belongs to nobody in this comparison. Every tool was pointed at that one URL between 14:00 and 14:20 UTC on 29 August 2026. Where a tool exposed a public endpoint, we called it directly and recorded the raw JSON. Where it did not, we recorded why, and we did not estimate a score in place of one we could not obtain.

We did not create an account, submit an email address, or pay for anything. That biases the results toward tools that let you in without a transaction, which is deliberate: it is the same bias a reader has at the moment they are looking for a scanner.

The ten scanners side by side

The five that produced a score

1. Cloudflare's Agent Readiness check: best for evidence you can audit

The tool at isitagentready.com shipped alongside Cloudflare's Agent Readiness announcement on 17 April 2026. It refuses to give a number. Instead it returns a level, and stripe.com came back at Level 1, "Basic Web Presence", with 4 checks passing, 17 failing and 1 neutral out of 22.

What separates it from everything else in this list is the evidence array. For every check it returns the requests it made, in order, with status codes and response headers, and a one-line finding for each. That is how we learned it tries two different Agent Skills paths, and it is why this tool caught something four others missed.

Best for: an engineer who wants to see what was actually requested before believing a verdict.

2. AgentGrade: best for continuous integration

AgentGrade scored stripe.com at 65%, grade B−, from 17 of 26 applicable checks across 7 of 12 applicable groups. Its catalog is larger than that; groups it judged irrelevant to the site, including Machine Payments and MCP endpoint discovery, were skipped rather than counted as failures. That is the most defensible scoring decision any tool in this group makes, and it is a large part of why its number is the highest.

It publishes a GitHub Action, a CLI and an MCP server, so a readiness check can gate a pull request. There is no pricing page. /pricing and /docs both return 404, and no paid-tier language appears anywhere on the site, so the tool appears to be free.

Best for: a team that wants readiness to fail a build rather than appear in a monthly report.

3. Agent Ready: best for depth on one page

Agent Ready returned 62, labelled "fair", alongside three sub-scores: llms.txt 97, accessibility 84, and a percentile of 65 against a corpus of 1,410 sites. That corpus figure is the only comparative baseline any of these tools volunteered, and it is worth more than the headline number.

Our scan returned 73 individual check results, 54 passing and 17 failing. The free tier allows 25 pages per scan and 10 scans per 30 days once signed in, with keyless CLI and MCP scans at a per-IP rate. Pro is $19/month for 250 pages per scan, 50 scans a month, 90-day history, weekly monitoring and REST API access (pricing page, fetched 29 August 2026). It also accepts x402 machine payments at $0.02 per 25-page scan.

Best for: auditing one important page in depth rather than a whole site cheaply.

4. AIScan: best for fixes written for your CMS

Our own scanner put stripe.com at 41, Level 2 "Readable", under rubric version 2026.08.2. That is the third-lowest number in this comparison and lower than two competitors, which is the honest result and we are publishing it as it came back.

The gap is explained by weighting rather than by disagreement about facts. AIScan gives 8 points to Markdown content negotiation (C1) and 5 to explicit AI bot rules (B2), both of which stripe.com fails, while its discoverability dimension scored a full 100. It also scored two things nobody else reported: 4,782 words of server-rendered text under E3, and a correct hard 404 on a probe path under E1.

It got one thing wrong, described in full below. It also misidentified the platform, reporting stripe.com as a Lovable site at 50% confidence on the strength of a /_next/static/ asset and the string "lovable" appearing somewhere in a JavaScript bundle. That is a false positive in the platform detector and it is being fixed.

Best for: someone who has a score and now needs the exact change to make in WordPress, Shopify or Next.js.

5. IsAgentReady: best for published weights, when it completes

IsAgentReady returned 38, grade D. That number should not be trusted, and the tool is transparent enough to show you why: three of its five category checkers reported "Checker error" after timing out at 15,000ms, and each was scored as zero. AI Search Signals, Agent Protocols and Security & Trust all failed to run. The two categories that did complete scored 69% and 87%.

Its weights are printed on its own about page: AI Content Discovery 30%, AI Search Signals 20%, Content & Semantics 20%, Agent Protocols 15%, Security & Trust 15% (fetched 29 August 2026). Publishing weights is rare and good. It does not publish a check count.

Best for: understanding how a weighted readiness rubric is put together, once you have re-run a failed scan.

The five we could not run, and why that is the finding

Half of this category cannot be used by an agent, which is a strange property for a tool that grades whether your site can be used by an agent.

• Search Engine Land / Semrush's checker is free and carries the most authority of anything here. Fetched without a browser it returns a loading shell listing its own steps, including that it tests 15 ChatGPT citation prompts. The score arrives only after client-side JavaScript runs.
• WordLift's AI Audit returns HTTP 403 with the body {"error":"Access denied. Please use the web application."}. It does publish a WebMCP tool manifest describing a run-audit tool, which is a browser-side interface rather than a callable endpoint.
• SiteSpeakAI's scanner returns HTTP 419, {"message":"CSRF token mismatch."}. It is a Laravel form that needs a live session.
• aireadinesschecker.com requires a URL, an email address and a marketing consent checkbox before it will scan. We do not submit email addresses to run a test.
• The Apify actor needs an Apify account and bills per usage. Its own listing showed 2 total users and 1 monthly active user when we checked, so treat it as an experiment rather than a product.

None of this makes those tools bad at scanning. It does mean their results cannot be reproduced by a reader, compared against another tool, or put in a pipeline, and a category built on measurement should care about that.

Where they disagreed, and who was right

Three checks produced contradictory verdicts on the same site on the same day. We fetched each underlying artifact by hand with curl to settle them.

Disagreement 1: the sitemap. Agent Ready was wrong.

Agent Ready reported S8 sitemap.xml not found, a 404 at the literal path. Cloudflare, AgentGrade and AIScan all passed the same check.

Ground truth: https://stripe.com/sitemap.xml really is a 404. But stripe.com's robots.txt opens with Sitemap: https://stripe.com/sitemap/sitemap.xml, and that URL returns 200 with 805 bytes of valid XML. The declaration is exactly what RFC 9309 provides for. A scanner that probes the conventional path and stops has found a convention, not a fact.

Disagreement 2: the Agent Skills index. AIScan was wrong.

Cloudflare passed this check and reported 8 skills. AIScan reported HTTP 404 and marked it informational.

Ground truth: Cloudflare is right and we are not. https://stripe.com/.well-known/skills/index.json returns a 301 to https://docs.stripe.com/.well-known/skills/index.json, which returns 200, application/json, 6,422 bytes, listing 8 skills including stripe-docs, stripe-apps and connect-recommend. Cloudflare's evidence log shows exactly how it got there: it requested the newer /.well-known/agent-skills/index.json path first, received a 404, logged "v0.2.0 path returned 404 — trying legacy path", and fell back.

AIScan probes only the newer path. Its own remediation text names /.well-known/agent-skills/index.json and nothing else. A real, published, spec-valid skills index was reported as absent, and it took a competitor's scanner to show us. The P3 check is being changed to try both paths and follow redirects.

Disagreement 3: robots.txt and AI crawlers. Neither was wrong.

Cloudflare passed robotsTxtAiRules with the message "No AI-specific bot rules; wildcard rules apply to all crawlers including AI bots." AIScan failed B2 with the evidence "0 known AI agent(s) addressed in robots.txt."

Both read the same file and described it accurately. stripe.com's robots.txt names three user agents: ia_archiver, * and rogerbot. Not one AI crawler appears. Cloudflare scores that as a pass because nothing is blocked. AIScan scores it as a failure because nothing is declared, which matters given how many AI crawler user agents now exist and that Google's robots.txt specification applies only the single most specific matching group, so a bot with its own group never reads your wildcard rules.

This one is not a bug in either tool. It is two different opinions about what "ready" means, expressed as a number.

Why the scores diverge

Once you have adjudicated the disagreements, the 24-point gap between the highest and lowest clean score stops looking like error and starts looking like disclosure.

The practical consequence: a readiness score is not portable between tools, and comparing your score to a competitor's on a different scanner is meaningless. Pick one, record the rubric version, and track the trend.

A dated timeline of this category

Which scanner to run, based on what you already know

The reader who arrives here has usually already run one scan. Branch on what it said.

• You got a score you think is too low. Re-run on a second tool before changing anything. Two of the five wrong-looking results here were the scanner's fault, not the site's. Check whether a failed item is a file the scanner looked for at one path.
• Two tools disagree on a specific file. Fetch it yourself: curl -sIL https://yoursite.com/<path>. Follow redirects with -L. A 301 to a docs subdomain is what defeated our own check.
• Every tool agrees you fail Markdown negotiation (C1) and llms.txt (C2). Those are the two most commonly failed checks in this whole category and the two most easily fixed. Start there.
• Your score is fine but your content is not appearing in AI answers. Readiness is not visibility. Check first whether your pages are server-rendered at all, because a site can rank on Google and be invisible to ChatGPT.
• A scanner reports a file missing that you know you published. Ask it which path it requested. If it will not say, that is your answer about the scanner.

What to do after a bad score

A low score is a work queue, not a diagnosis. In order, cheapest first:

1. Run the free scan and read the check IDs, not the number. npx aiscan-cli yoursite.com, or paste the URL at aiscan.site. It names the specific checks: C1 Markdown content negotiation, C2 llms.txt, C3 and E3 server-rendered HTML, B2 explicit AI bot rules, D2 XML sitemap, P3 Agent Skills index. If you would rather check by hand, curl -sI -H 'Accept: text/markdown' https://yoursite.com/ answers C1 in one line, and curl -s https://yoursite.com/robots.txt answers B2.
2. Fix llms.txt and robots.txt first. They are files, not architecture, and they move three checks between them. The llms.txt generator will build a valid one from your sitemap.
3. On WordPress, use one plugin rather than three. ThinkRank handles robots.txt, robots meta, schema, sitemaps and llms.txt from a single place, which is the answer to the usual problem of three SEO plugins each rewriting the same robots.txt. It migrates settings from Rank Math, Yoast, All in One SEO and SEOPress, so switching costs nothing in re-entered configuration. Rank Math and Yoast both remain stronger on classic on-page SEO workflows and editorial scoring, and if that is the bulk of your work they are reasonable choices; neither generates llms.txt.
4. On Shopify, StoreSEO is the shortest path to llms.txt, agents.md and product schema generated from your real catalog rather than hand-maintained. It is rated 5.0 from 685 reviews on the Shopify App Store as of 27 August 2026. Yoast SEO for Shopify is the main alternative and is stronger on editorial metadata; it does not generate llms.txt.
5. Re-scan and record the rubric version. A score without a rubric version cannot be compared to itself next month.

Nobody in this category publishes a recovery timeline, ourselves included, because no crawler operator publishes re-crawl intervals. What is documented is propagation after a robots.txt change: OpenAI roughly 24 hours, Perplexity up to 24 hours, Amazon roughly 24 hours with robots.txt cached up to 30 days, DuckDuckGo 72 hours.

Where AIScan fits, and where it doesn't

What AIScan cannot see. It cannot tell you whether an AI assistant actually cites you, because that requires prompt sampling rather than crawling. It cannot tell you whether a crawler executed your JavaScript, because no operator documents rendering behaviour. It cannot see anything behind a login. And, as this test established, it cannot currently see an Agent Skills index published at the legacy path. If you are choosing on evidence quality alone, Cloudflare's tool is better than ours; if you are choosing on CI integration, AgentGrade is. Our older roundup of free AI readiness checkers covers the wider set of free tools, including visibility monitors this test excluded.

Your 10-minute checklist

• Run one free scan, no account: npx aiscan-cli yoursite.com
• Run a second scanner on the same URL and compare check by check, not score to score
• For any file reported missing, curl -sIL the path yourself and follow redirects
• Read your robots.txt and count how many AI crawlers it names
• Test curl -sI -H 'Accept: text/markdown' https://yoursite.com/ for C1
• Confirm your sitemap is declared in robots.txt, not just present at the conventional path
• Record today's date and the rubric version alongside the score

Run the same test on your own site

Run the scan on your own site and read the check IDs rather than the number: npx aiscan-cli yoursite.com, or paste the URL at aiscan.site, free and without an account. The checks this article turned on are C1 and C2 under content, B2 under bot access, P3 under capabilities, and D2 and E1 under discoverability. Then run a second scanner and see whether it agrees. Where it does not, the disagreement is usually more informative than either score.

Every check in the rubric is documented at aiscan.site/guides, and if llms.txt is one of your failures, the llms.txt generator will build a valid file from your sitemap in about a minute. The evidence on whether that file is read by anyone is in our llms.txt review.

Testing and analysis by M Asif Rahman. All scans run 29 August 2026 against the tools' own public endpoints; raw results available on request.