Docs

What the audit checks

An audit opens your site in a real browser, walks the main flows on a desktop and a phone, reads up to 15 pages the way a search engine does and measures the most important ones on a phone. This page lists every check it runs, when each is reported and why it matters.

How an audit works

  • Works. An AI agent walks the main flows (home page, pricing, sign up, contact) on a desktop and then on a phone, like a first-time visitor. It reads and clicks, and in safe mode it never sends a form or pays.
  • Pages. The pages to read are found from robots.txt, the sitemap and the links the agent saw, one of each kind (one blog post, not fifty), up to 15.
  • SEO. Each page is read twice: as the HTML the server sends, the way crawlers that run no JavaScript see it, and after its scripts ran, the way Google renders it.
  • Links. Every link on the pages the agent walked is opened once, new-tab links included: up to 150 to the site itself (its subdomains count as the site) and up to 60 to other sites. A link to the site is broken when it returns 404, 410 or a server error. A link to another site only counts as broken when it returns 404 or 410 or its domain no longer exists; one that refuses automated requests is shown as such, never as broken. A link to the site the agent did not open can be tested as a flow.
  • Test deeper. While the phone pass runs, the audit reads a few more pages without clicking and proposes flows worth testing next, each starting on a page it found and saying what it proves. A free report runs one of them on the same report; a project runs them all.
  • Speed. The start page and the next two most important pages are measured on a phone, and the start page on a desktop. A page that looks slow is measured a second time before it is reported.

Fixed something? Move its card to Fixed on your issue board, or click I fixed it, check again in a free report to get there. Only that check runs again, without the AI: speed measures the pages it was about again, SEO reads the pages again, and a Works issue repeats the steps that led to it. Gone, and the card moves to Verified fixed; still there, and it goes back to Fixing with the new evidence.

Checked before you see it. A finding the audit is less sure of goes to a person on our team first, and appears in your report only once confirmed. The checks below say when that is always the case.

Works

Does every page, link and form work, on a desktop and on a phone.

Blocked by the Content Security Policy

Reported when: The browser refuses to load a script, frame or request because the page’s own policy does not allow its host.

Why it matters: The Content Security Policy tells the browser which hosts the page may load from, and this host is not on the list.

Based on developer.mozilla.org · csp

Page returns a server error

Reported when: A page the agent opened answered with a 5xx status.

Why it matters: The server fails while building this page, so every visitor who opens it gets an error instead of the page.

Based on developer.mozilla.org · http_error_page_5xx

Page not found

Reported when: A page the agent opened answered 4xx, usually 404.

Why it matters: Visitors who follow this path land on a "not found" page and usually leave.

Based on developers.google.com · http_error_page_404

A request fails on the server

Reported when: A request the page made while the agent used it answered with a 5xx status. A GET is asked for again a few seconds later: when it answers then, a person checks it first.

Why it matters: A request the page makes in the background fails on the server, so whatever it loads or saves does not work.

Based on developer.mozilla.org · http_error_request_5xx

A request is refused

Reported when: A request the page made answered 4xx. 401 and 403 are left out: a visitor who is not signed in gets them on purpose.

Why it matters: The page asks the server for something that does not exist or is not allowed, so part of the page stays empty or broken.

Based on developer.mozilla.org · http_error_request_4xx

A request never arrives

Reported when: A request the page made failed at the network level: DNS, connection or timeout.

Why it matters: A request from the page never got an answer: the server is down, too slow, or blocked by the browser.

Based on developer.chrome.com · http_error_network

Error shown on the page

Reported when: The page shows an error text instead of its content, such as “Application error”, “Something went wrong” or “Page not found”.

Why it matters: The page shows an error message to visitors instead of its content.

Based on developer.chrome.com · error_page

Blank page

Reported when: The page loaded but has under 20 characters of text and almost no elements, or still shows a loading indicator after the network went quiet.

Why it matters: The page shows almost nothing, or a loading screen that never ends, so visitors think the site is broken.

Based on developer.chrome.com · empty_page

Script crash

Reported when: The page threw a JavaScript error it did not handle while it was used. An error from a third-party script is reported as low.

Why it matters: A script on the page throws an error it does not handle, and the feature it belongs to can stop working.

Based on developer.chrome.com · page_error

Error in the console

Reported when: The page logged an error in the browser console while it was used. Errors the audit’s own read-only guard caused are left out.

Why it matters: The page logs an error. It may be harmless, but often it is a feature that quietly fails.

Based on developer.chrome.com · console_error

Button that does nothing

Reported when: The agent clicked a button or link and nothing changed: the same address, the same page content and the same scroll position.

Why it matters: Something looks clickable but nothing happens when visitors click it.

Based on developer.mozilla.org · dead_click

Form that cannot be sent

Reported when: A form that was sent got an error from the server, or nothing happened: no message, no new page. Forms are only sent on runs with write access.

Why it matters: Sending the form fails or does nothing, so the messages or sign-ups it should collect are lost.

Based on developer.mozilla.org · form_error

Not set up for phones

Reported when: On a phone the page is laid out wider than 600 px, because it has no viewport meta tag: a shrunken desktop page.

Why it matters: Without a viewport tag, phones show the desktop page shrunk to fit, with tiny text.

Based on developer.mozilla.org · mobile_viewport

Wider than the phone

Reported when: On a phone the page is wider than the screen and scrolls sideways. Not judged while a loading screen is up.

Why it matters: Part of the page is wider than a phone screen, so the page scrolls sideways and content is cut off.

Based on web.dev · mobile_overflow

Speed

How fast the most important pages load on a mid-range phone over slow 4G, with an empty cache.

Main content slow to appear (LCP)

Reported when: Needs work over 2.5 s, poor over 4 s, measured on a mid-range phone over slow 4G.

Why it matters: The main content (usually the hero image or headline) appears late, and visitors leave while they wait.

Based on web.dev · speed_lcp

Screen stays blank (FCP)

Reported when: Needs work over 1.8 s, poor over 3 s, measured on a mid-range phone over slow 4G.

Why it matters: The screen stays blank for a while before anything appears, and visitors think the page is not loading.

Based on web.dev · speed_fcp

Page jumps while loading (CLS)

Reported when: Needs work over 0.1, poor over 0.25, on a phone or a desktop.

Why it matters: Content jumps around while the page loads, so visitors lose their place or tap the wrong thing.

Based on web.dev · speed_cls

Page ignores taps while loading (TBT)

Reported when: Needs work over 200 ms, poor over 600 ms, measured on a mid-range phone over slow 4G. A person checks it before you see it.

Why it matters: Scripts keep the phone busy while the page loads, so taps and scrolls respond late.

Based on web.dev · speed_tbt

Server slow to answer (TTFB)

Reported when: Needs work over 0.8 s, poor over 1.8 s, measured on a desktop connection.

Why it matters: The server takes long to start answering, and every other part of the page waits for it.

Based on web.dev · speed_ttfb

Page shows nothing for 30 s

Reported when: Nothing appeared on a phone within 30 seconds.

Why it matters: Nothing showed on a phone over a slow connection for the whole time we waited, so many visitors would leave first.

Based on web.dev · speed_load

Heavy page

Reported when: A page transfers more than 3 MB on a phone.

Why it matters: The page downloads several megabytes, which is slow and costly on phone data.

Based on developer.chrome.com · speed_page_weight

Main image loads lazily

Reported when: The largest element is an image with loading="lazy", so the browser waits before fetching it.

Why it matters: The main image is lazy-loaded, so the browser waits until layout is done before it even starts downloading it.

Based on web.dev · speed_lcp_lazy

Main image found late

Reported when: The largest image is not in the HTML the server sends and has no preload, so the browser finds it only after scripts or styles run.

Why it matters: The main image is not in the HTML (it comes from CSS or a script), so the browser finds it late.

Based on web.dev · speed_lcp_discovery

Main image too heavy

Reported when: The largest image weighs more than 200 KB.

Why it matters: The main image is a large file, so it takes long to download on a phone.

Based on web.dev · speed_lcp_heavy

Files that hold back the first paint

Reported when: A blocking script, a stylesheet from another host or more than 20 KB of stylesheets in the page head, while the first paint is slow. A person checks it before you see it.

Why it matters: Stylesheets and scripts in the <head> make the browser wait before showing anything.

Based on developer.chrome.com · speed_render_blocking

Images larger than shown

Reported when: Images have more than twice the pixels they are shown at, counting the phone’s screen density, wasting over 100 KB.

Why it matters: Images are much larger than the space they are shown in, so visitors download pixels nobody sees.

Based on web.dev · speed_image_size

Large images in old formats

Reported when: JPEG, PNG or GIF images over 100 KB each, together over 200 KB, that WebP or AVIF would make much smaller.

Why it matters: Images are JPEG or PNG, while WebP or AVIF are usually much smaller at the same quality.

Based on web.dev · speed_image_format

Images without a size

Reported when: Images have no width and height, so the page moves when they load.

Why it matters: Images have no width and height, so the page jumps when they load.

Based on web.dev · speed_image_dimensions

Files sent without compression

Reported when: Scripts, stylesheets or pages of the site over 20 KB are sent without gzip or Brotli.

Why it matters: Text files (HTML, CSS, JavaScript) are sent without compression, often three to five times larger than needed.

Based on developer.chrome.com · speed_compression

Files browsers do not keep

Reported when: Three or more of the site’s own scripts, stylesheets, fonts or images may be kept for less than a day. Service workers are left out.

Why it matters: Files that rarely change are not cached, so returning visitors download them again on every visit.

Based on developer.chrome.com · speed_caching

Text hidden while fonts load

Reported when: An @font-face rule has no font-display: swap or optional, so text is invisible until the font arrives.

Why it matters: Text stays invisible while web fonts download.

Based on developer.chrome.com · speed_font_display

A lot of JavaScript

Reported when: A page loads more than 500 KB of compressed JavaScript.

Why it matters: The page downloads a lot of JavaScript, which a phone has to download, parse and run before the page responds.

Based on web.dev · speed_js_weight

SEO

Can search engines read, understand and show every page.

No title

Reported when: The page has no <title>.

Why it matters: Without a title, Google makes one up for search results and the browser tab shows the address.

Based on developers.google.com · seo_title_missing

Title too long

Reported when: The title is longer than 60 characters, about what Google shows.

Why it matters: Google cuts titles longer than its results line, often losing the part that matters.

Based on developers.google.com · seo_title_long

Title too short

Reported when: The title is shorter than 10 characters.

Why it matters: A one-word title tells searchers nothing about the page.

Based on developers.google.com · seo_title_short

No meta description

Reported when: The page has no meta description.

Why it matters: Without a meta description, Google picks any text from the page for the snippet under the title.

Based on developers.google.com · seo_description_missing

Description cut off

Reported when: The meta description is longer than 160 characters.

Why it matters: Google cuts long descriptions, so the end of the sentence is lost.

Based on developers.google.com · seo_description_long

No main heading

Reported when: The page has no <h1> after its scripts ran.

Why it matters: The main heading tells search engines and screen readers what the page is about.

Based on developer.mozilla.org · seo_h1_missing

Several main headings

Reported when: The page has more than one different <h1>. The same heading twice, one for phones and one for desktops, is fine.

Why it matters: Several different main headings make it unclear what the page is about.

Based on developer.mozilla.org · seo_h1_multiple

Heading levels skipped

Reported when: A heading jumps more than one level down, for example from <h2> to <h4>.

Why it matters: A heading level is skipped (an H2 followed by an H4), so the outline of the page is broken for search engines and screen readers.

Based on developer.mozilla.org · seo_heading_skip

No canonical address

Reported when: The page has no <link rel="canonical">.

Why it matters: Without a canonical link, the same page under different addresses (with tracking parameters, a trailing slash) can be indexed as duplicates.

Based on developers.google.com · seo_canonical_missing

Canonical points elsewhere

Reported when: The page names another address as canonical, so Google indexes that one instead.

Why it matters: The canonical link names another page, so Google shows that one in search results instead of this one.

Based on developers.google.com · seo_canonical_elsewhere

Canonical is broken

Reported when: The canonical address answers 4xx or 5xx.

Why it matters: The canonical link points to a page that returns an error, so Google may drop this page.

Based on developers.google.com · seo_canonical_broken

Hidden from search engines

Reported when: A noindex robots meta tag or X-Robots-Tag header. Critical on the home page.

Why it matters: A noindex tag or header tells search engines not to show the page.

Based on developers.google.com · seo_noindex

No language

Reported when: The <html> element has no lang attribute.

Why it matters: Without a language on the page, search engines guess it and screen readers may read it with the wrong voice.

Based on developer.mozilla.org · seo_lang_missing

Not set up for phones

Reported when: The page has no viewport meta tag, which Google’s mobile-first indexing expects.

Why it matters: Without a viewport tag, phones show the desktop page shrunk, and Google ranks the phone version.

Based on developer.mozilla.org · seo_viewport_missing

Incomplete link preview

Reported when: og:title or og:image is missing, so shared links show no proper card.

Why it matters: Without Open Graph tags, a shared link shows as plain text or a random image on LinkedIn, Facebook, Slack and WhatsApp.

Based on ogp.me · seo_og_missing

Broken structured data

Reported when: A JSON-LD block does not parse.

Why it matters: A JSON-LD block is not valid JSON, so Google ignores it and the rich results it was meant for.

Based on developers.google.com · seo_structured_data_invalid

Images without alt text

Reported when: Images have no alt attribute. Tracking pixels and hidden images are left out.

Why it matters: Images without alt text say nothing to image search and screen readers.

Based on developers.google.com · seo_image_alt

Empty without JavaScript

Reported when: The HTML the server sends has less than a fifth of the text the page shows after its scripts run. A person checks it before you see it: a site that prerenders for crawlers can send Google the full page.

Why it matters: The HTML your server sends is nearly empty; the text appears only after scripts run, which search engines do later and less reliably, and AI assistants and link previews often not at all.

Based on developers.google.com · seo_js_only_content

Title set by JavaScript

Reported when: The title or description is only added by scripts, not in the HTML the server sends.

Why it matters: The title or description is only set by a script, so link previews and crawlers that run no scripts see none.

Based on developers.google.com · seo_js_only_head

Page stayed blank

Reported when: The page showed almost no text even after its scripts ran, so its content was not checked. Costs no points.

Why it matters: The page stayed blank in a fresh browser, even after its scripts ran. If a script fails to load or the app starts slowly for Google too, it indexes an empty page.

Based on developers.google.com · seo_not_rendered

Pages share a title

Reported when: Two or more pages that are their own canonical have the same title.

Why it matters: Several pages share one title, so search engines cannot tell them apart and may show only one.

Based on developers.google.com · seo_duplicate_title

Pages share a description

Reported when: Two or more pages have the same meta description. A person checks it before you see it.

Why it matters: Several pages share one description, so their snippets in search results look the same.

Based on developers.google.com · seo_duplicate_description

Sitemap lists missing pages

Reported when: Addresses in the sitemap answer 4xx or 5xx.

Why it matters: The sitemap lists pages that do not exist, so Google wastes visits on errors and trusts the sitemap less.

Based on developers.google.com · seo_sitemap_broken

Sitemap lists redirects

Reported when: Addresses in the sitemap redirect somewhere else.

Why it matters: The sitemap lists addresses that redirect, instead of the final address of each page.

Based on developers.google.com · seo_sitemap_redirect

Pages missing from the sitemap

Reported when: Pages the site links to are not in its sitemap.

Why it matters: Pages the site links to are missing from the sitemap, so Google may find them later or not at all.

Based on developers.google.com · seo_sitemap_missing_pages

No sitemap

Reported when: robots.txt names no sitemap and /sitemap.xml and /sitemap_index.xml do not exist.

Why it matters: Without a sitemap, Google only finds pages by following links, and finds new ones later.

Based on developers.google.com · seo_sitemap_missing

Sitemap is not valid

Reported when: The sitemap has no <urlset> or <sitemapindex>, for example an HTML page served at /sitemap.xml.

Why it matters: The sitemap cannot be read as a sitemap, so Google ignores it.

Based on developers.google.com · seo_sitemap_invalid

robots.txt blocks everything

Reported when: robots.txt disallows the whole site for all crawlers.

Why it matters: robots.txt tells every crawler to stay away from the whole site, so pages drop out of search results.

Based on developers.google.com · seo_robots_blocks_all

robots.txt fails

Reported when: robots.txt answers 5xx, which tells Google to crawl nothing.

Why it matters: robots.txt answers with a server error, and Google stops crawling the site while it does.

Based on developers.google.com · seo_robots_error

No robots.txt

Reported when: The site has no robots.txt.

Why it matters: There is no robots.txt, so there is no line pointing crawlers to the sitemap.

Based on developers.google.com · seo_robots_missing

No structured data

Reported when: No page read has JSON-LD structured data.

Why it matters: Structured data tells Google what the site is (an organization, a product, an article) and is what rich results are made of.

Based on developers.google.com · seo_structured_data_missing

Opens without https

Reported when: The http:// address answers without redirecting to https://.

Why it matters: The site also answers over plain http without redirecting, so there are two copies of every page and visitors on one see "Not secure".

Based on developers.google.com · seo_https_redirect

www and bare domain are two copies

Reported when: The other spelling of the host (with or without www) answers without redirecting to the one the site uses.

Why it matters: Both www and the bare domain answer, so the site exists twice and links people share are split between them.

Based on developers.google.com · seo_host_redirect

Lab speed and real visitors

The audit measures speed in a lab: one visit on a simulated mid-range phone over slow 4G, with an empty cache. That is the same setting Lighthouse and PageSpeed Insights use, and it shows what a first-time visitor on a slow connection gets. Real visitors can get a faster site, on a fast phone, on Wi-Fi or with the site in their cache, or a slower one, from far away or on a busy network.

  • Main content (LCP) and first paint (FCP) agree with Lighthouse's rating, or are milder, on every site we compared.
  • Blocking time (TBT) is measured on a real, slowed-down load, while Lighthouse simulates one. The two can differ a lot, so blocking time weighs half in the speed score and its finding is checked by a person first.
  • Real-visitor data from the Chrome UX Report, for sites with enough Chrome traffic, is coming to the report next to the lab numbers.

The Scores guide explains how the numbers become the speed score.

Stacks we recognise

From the outside the audit looks for signs a platform leaves: a header, a file it loads, a meta tag, a cookie or a global variable. Each one in the report names the sign it was found by. A fix in the report is written for the stack it recognised, and for plain web advice when it recognised none.

Platforms

WordPress · Shopify · Webflow · Wix · Squarespace · Framer · Ghost

Page builders

Elementor · Divi

Plugins

WooCommerce · WP Rocket · LiteSpeed Cache · W3 Total Cache · Yoast SEO · Rank Math

Frameworks

Next.js · Nuxt · Gatsby · Astro · SvelteKit · Remix / React Router · Ruby on Rails · Laravel · React · Vue

Hosting

Vercel · Netlify · Cloudflare

What the audit cannot see from outside

  • Behind sign-in. A free audit sees what a visitor sees. Pages behind a login are audited in a project with a test account; see Login options.
  • Sending forms and paying. In safe mode forms are filled in but not sent, and nothing is paid. Staging with write access, and test cards, are in Safety.
  • Your server and database. Errors are seen as the browser gets them: a 500, a failing request, an error on the page. Why the server failed is in your logs.
  • Every page. One page of each kind, up to 15, not the whole site. A problem on one blog post usually is on all of them.
  • Search rankings and traffic. The SEO checks say what stops Google reading or showing a page, not where it ranks.
  • A stack that leaves no sign. Custom code, or a build that inlines everything, cannot be named: the report says so rather than guess, and the fixes are generic.