Educational organization
F
Not ready for AI agents
Capped at F: your page has no text an assistant can read before scripts run. Your checks earned 85.
See the goal behind itSpeed: Pass (Google's lab run of this page)Core Web Vitals, Google
Your home page holds 18 words before any script runs. An assistant that runs no script reads only what your server sends.
What to do Make the home page send its text from the server, so it shows up with scripts turned off.
Your web team · a few daysSee the check
Your home page offers no tools to AI assistants and No WebMCP tools detected on page load.
What to do Expose the main action on your site, such as booking or contacting you, as a WebMCP tool.
Your web team · about a daySee the check
Your site has no ARD catalog at /.well-known/ard.json, so an agent cannot discover an MCP server from it.
What to do Run an MCP server for your main actions and list it in the site's ARD catalog.
Your web team · a few daysSee the check
These 9 fixes add up to about three weeks of your web team's time.
An email with what is wrong and a link to this report, so whoever runs the site has every fix and its line for the web team.
Send to your agencyYour own AI reads the report and tells you how to fix each issue and the effort each takes. It gets this, and nothing else:
Read this report on our website: clutch.io/grade/apessoa-com. Tell me how to fix each issue and the level of effort each fix takes.
Clutch rebuilds apessoa.com from its address, and the new site doesn't have these issues. Free to preview, $39 a month once it goes live.
See your site rebuiltCan AI agents and search engines reach and list your pages?
25% of your grade · 10 of 10 goals pass
Cloudflare's scan reached your site. An agent that is refused at the door reads nothing. Claude's web fetch tool gets only what a site answers to its request, so a firewall that challenges or blocks automated requests leaves it without your page.
Your robots.txt allows all 20 of the AI crawlers Cloudflare checks. A crawler is a robot that reads sites for a search engine or assistant, and one that is blocked cannot recommend you. OpenAI, Anthropic and Perplexity each say their crawlers follow robots.txt.
Overview of OpenAI crawlers, OpenAIReport an issue with this check
Page isn’t blocked from indexing Google leaves out a page marked noindex, and a page its crawler cannot read cannot be listed or used as a supporting link in its AI features.
AI features and your website, GoogleReport an issue with this check
robots.txt is valid Google says a robots.txt file tells crawlers which addresses they may request. A line a crawler cannot read is skipped, so a rule you meant to set may never apply.
Robots.txt Introduction and Guide, GoogleReport an issue with this check
sitemap.xml exists with valid structure. Google and Bing use an XML sitemap to find your pages, and neither promises to index every page in it. A page must be indexed to appear in Google's AI answers.
Build and submit a sitemap, GoogleReport an issue with this check
All 2 audits in this goal pass. Google says the vast majority of the new pages it finds each day come through links, and it follows only links that are real <a> elements with an address.
SEO Link Best Practices for Google, GoogleReport an issue with this check
Document has a <title> element Google builds the title link in search results from the page's title, and may change one that is missing, vague, boilerplate or repeated. A title keeps your wording in front of searchers.
Influencing Title Links in Google Search, GoogleReport an issue with this check
A page that does not exist answers 404. Google treats content behind a 404 or 410 as gone, and flags empty pages that answer 200 (success) as soft 404s.
How Google Search handles HTTP status codes, GoogleReport an issue with this check
All 2 audits in this goal pass. Google uses one canonical address per page, and uses hreflang to send searchers to the version of the page in their language.
How to Specify a Canonical with rel="canonical" and Other Methods, GoogleReport an issue with this check
All 2 audits in this goal pass. Each extra redirect is another trip across the network that can delay the page by hundreds of milliseconds.
Avoid multiple page redirects, ChromeReport an issue with this check
25% of your grade · 7 of 11 goals pass
Your home page holds 18 words before any script runs. An assistant that runs no script reads only what your server sends. Claude's web fetch tool does not support pages that fill in their text with JavaScript, and Google reads such a page only after its first HTML, once it reaches the rendering queue. A page whose first response is empty waits or looks empty.
While it fails, your letter is an F.
What to do Make the home page send its text from the server, so it shows up with scripts turned off.
For your web team: Render the page's content on the server or at build time, and use scripts to add to it. Fetch the page without a browser to check the text is in it.
Your web team · a few days · Critical
Understand JavaScript SEO basics, Google Search CentralReport an issue with this check
Your site does not publish an ARD catalog at /.well-known/ard.json. The Agentic Resource Discovery (ARD) draft is a format that helps assistants discover what a site offers, such as tools and agents. Lighthouse audits the catalog, and the grade counts a standard that is likely to be adopted.
What to do Publish an ARD catalog once your site has an MCP server or agent to list.
For your web team: Serve a JSON file at /.well-known/ard.json with an entries list that follows the ARD schema: each entry carries a urn:air: identifier, a displayName, a type and a url. An MCP server's entry has the type application/mcp-server-card+json and points at its server card, at /mcp/server-card. Name the catalog from every page with <link rel="ard" href="/.well-known/ard.json">. Lighthouse reads that path and warns about a catalog it finds only at the older /.well-known/ai-catalog.json.
Your web team · about a day · Important
Lighthouse agentic browsing scoring, ChromeReport an issue with this check
Site does not support Markdown for Agents. Cloudflare calls Markdown the common language of agents: its explicit structure gives better results and wastes fewer tokens, and an agent asks for it with an Accept header.
What to do Serve each page as Markdown to agents that ask for it.
For your web team: Answer a request whose Accept header names text/markdown with the page's content as Markdown and Content-Type: text/markdown. Cloudflare can convert pages at its edge, or the server can render the Markdown itself.
Your web team · about a day · Important
Markdown for Agents, CloudflareReport an issue with this check
No Content Signals found in robots.txt. Content Signals follow AI Preferences, a proposal at the IETF for a way for a publisher to say how automated systems may use its content.
What to do Decide whether your content may be used for search, AI answers and AI training, and ask the people who run your site to say so in robots.txt.
For your web team: Add a Content-Signal line to robots.txt under the user agents it covers, such as Content-Signal: search=yes, ai-input=yes, ai-train=no.
Your web team · about an hour · Important
Content Signals, Content SignalsReport an issue with this check
The home page's structured data names ProfilePage, Person, PostalAddress, Organization, Occupation, Place, EducationalOrganization, EducationalOccupationalCredential, Language, InteractionCounter, and CreativeWork. Structured data spares an assistant from guessing what a page means. Google says it gives explicit clues about a page; Bing says it may support clearer grounding and promises nothing.
Intro to structured data, GoogleReport an issue with this check
Heading elements appear in a sequentially-descending order Headings give a page a clear structure.
Optimizing your website for generative AI features on Google Search, GoogleReport an issue with this check
Image elements have [alt] attributes Alt text is a short written description of an image. Google calls it the most important attribute for adding metadata about an image, and it also uses computer vision on the picture.
Image SEO Best Practices, GoogleReport an issue with this check
All 2 audits in this goal pass. WCAG, the web accessibility standard most laws point to, requires the structure a page shows on screen to be in its markup as well, and the axe rules check lists and tables for it.
Rule list: <ul> and <ol> must only directly contain <li>, <script> or <template> elements, Deque axeReport an issue with this check
llms.txt follows recommendations llmstxt.org proposes the file so assistants find what they need and follow its links. Lighthouse audits it as an emerging convention, and the grade counts a standard that is likely to be adopted.
Document has a meta description Google can show a page's meta description as its snippet and asks for a different one on every page. It is less likely to show a list of keywords, so a specific summary has the best chance of being used.
How to Write Meta Descriptions, GoogleReport an issue with this check
All 2 audits in this goal pass. WCAG requires the language of each page to be set in its code, so software reads it with the right rules and voice.
Rule html-has-lang: <html> element must have a lang attribute, Deque axeReport an issue with this check
Can they take the next step for a customer?
25% of your grade · 6 of 8 goals pass
Your home page offers no tools to AI assistants and No WebMCP tools detected on page load. WebMCP lets an assistant book or reserve through a site's own code instead of guessing its way around the screen.
While it fails, your letter stays at B or lower.
What to do Expose the main action on your site, such as booking or contacting you, as a WebMCP tool.
For your web team: Call document.modelContext.registerTool({ name, description, inputSchema, execute }) as the home page loads, or add toolname and tooldescription attributes to the form. A name is 1 to 128 letters, digits, _, - or .
Your web team · about a day · Important
WebMCP for developers, ChromeReport an issue with this check
Your site has no ARD catalog at /.well-known/ard.json, so an agent cannot discover an MCP server from it. ARD is how an agent finds what a site offers from its domain alone, and the MCP project accepted server cards so that a client can discover a server before it connects.
While it fails, your letter stays at B or lower.
What to do Run an MCP server for your main actions and list it in the site's ARD catalog.
For your web team: Serve the MCP server card at /mcp/server-card, as the server card extension places it, and list it in /.well-known/ard.json as an entry with a urn:air: identifier, a displayName, the type application/mcp-server-card+json and the card's URL.
Your web team · a few days · Important
SEP-2127: MCP Server Cards, Model Context ProtocolReport an issue with this check
Links have a discernible name
Rule button-name: Buttons must have discernible text, Deque axeReport an issue with this check
Accessibility tree is well-formed (All audits passed) Agents review the accessibility tree to find what they can click or fill in, and missing labels can block them.
Accessibility for agents, ChromeReport an issue with this check
[aria-hidden="true"] is not present on the document <body> Wrong ARIA tells an assistant a control is something it is not. WAI-ARIA defines the roles and attributes that are valid, and the axe rules check them.
Rule aria-roles: ARIA roles used must conform to valid values, Deque axeReport an issue with this check
Document has a main landmark. WCAG, the web accessibility standard most laws point to, requires every function of a page to work from a keyboard.
Accessibility for agents, ChromeReport an issue with this check
All 3 audits in this goal pass. WCAG requires text to contrast with its background by at least 4.5 to 1, and the axe rule color-contrast checks for it.
Rule color-contrast: Elements must meet minimum color contrast ratio thresholds, Deque axeReport an issue with this check
All 2 audits in this goal pass.
Accessible tap targets, GoogleReport an issue with this check
No Link headers found on target page, API Catalog not found, Agent Skills index not found, and A2A Agent Card not found. Each is a published way for a site to tell an agent what it can call or learn there. Few sites that sell to the public need them yet, so they earn points and cost none.
What to do Ask whoever runs your site which of these your business has a use for, such as an API catalog if you publish an API.
For your web team: Publish the files that fit: an API catalog at /.well-known/api-catalog, an Agent Skills index, an A2A agent card at /.well-known/agent-card.json, and Link headers on the home page that point to them.
Your web team · about a day · Important
Is it fast for the person who clicks through?
Speed comes from real visitors of your page over the last 28 days when Google has their numbers, and from a test on a phone when it does not.
25% of your grade · 8 of 10 goals pass
Improve image delivery (Est savings of 3,501 KiB). Serving images larger than the version rendered on screen results in wasted bytes and slows page load.
What to do Resize and compress your images before they go on the page.
For your web team: Serve several sizes with srcset and sizes, use a modern format such as WebP or AVIF, and keep each image's width and height in the proportion it is shown.
Your web team · about a day · Important
Avoid enormous network payloads (Total size was 4,316 KiB) and Reduce unused JavaScript (Est savings of 92 KiB). Large network payloads are highly correlated with long load times and cost visitors money on mobile data.
What to do Remove code and files the page does not use.
For your web team: Minify scripts and styles, remove unused and duplicate code, and stop shipping legacy polyfills. PageSpeed lists the files by size.
Your web team · a few days · Important
Avoid enormous network payloads, ChromeReport an issue with this check
All 3 audits in this goal pass. Lab test on a phone: too few real visitors to measure it. LCP is how long the main picture or text takes to appear, and visitors leave slow pages. web.dev says 2.5 seconds or less is good, one of Google's three Core Web Vitals.
Total Blocking Time (10 ms) INP is the wait between a tap and the page responding, and under 200 milliseconds is good. Total Blocking Time, the lab test's stand-in, has the same bar.
Total Blocking Time (TBT), GoogleReport an issue with this check
Cumulative Layout Shift (0) CLS measures how much the page jumps while it loads, which makes people tap the wrong thing. web.dev names it one of three Core Web Vitals, and 0.1 or less is good.
Uses HTTPS A secure connection helps prevent intruders tampering with traffic between your site and its visitors.
All 3 audits in this goal pass. A CDN (a network of servers near your visitors) reduces the network latency that slows a server's first response.
Reduce server response times (TTFB), ChromeReport an issue with this check
Use efficient cache lifetimes Fetching resources over the network is both slow and expensive.
Prevent unnecessary network requests with the HTTP Cache, GoogleReport an issue with this check
All 7 audits in this goal pass. An error in the console means a problem on your page that you need to resolve.
Browser errors were logged to the console, ChromeReport an issue with this check
All 3 audits in this goal pass. Users are mistrustful of or confused by pages that ask for their location as soon as they load.
Requests the geolocation permission on page load, ChromeReport an issue with this check