Help
Help
Audits
An audit scans a site the way Explore does, then checks every page it reaches: for accessibility issues, for what search engines read on it, for the layout it uses, and for buttons, form fields, and checkboxes whose style breaks the site’s own pattern.
What an audit checks
- Accessibility
- axe-core’s checks against WCAG 2.0 to 2.2, levels A and AA, plus its best practices: missing labels and alt text, low contrast, and more. On by default; switch off Check accessibility to skip it.
- Search
- What search engines read on each page: its title, meta description, canonical link, whether it may be indexed, its H1s, social preview tags, and structured data. On by default; switch off Check search basics to skip it.
- Layouts
- Every page’s frame (header, sidebar, footer, how wide the content runs) and the sections it stacks, grouped into the layouts the site uses and their variations, with the places a layout almost matches the rest.
- Buttons
- Every button’s font, colors, corners, padding, and size, compared across the site.
- Form fields
- Text inputs, selects, and text areas: font, height, borders, fill, and placeholder color.
- Checkboxes and radios
- Checkboxes, radio buttons, and switches, each compared only with its own kind.
- Page loads
- Whether each page loaded cleanly (no error page, none of the site’s files missing, no script errors), and how long it took to show its main content.
“Breaks the pattern” means two things. A rare trait is a style few elements use while most share another, like the one pill-shaped button on a site of square ones. A near miss is a value just off a common one, like a green two shades away from the brand green. Differences too small to see, like 1px, are listed separately as minor.
Start an audit
On Audits, choose New audit, paste the start page, and pick how many pages to scan.

It scans 12 pages unless you change it, up to 24, following links the way Explore does. Audits always browse as a desktop.
Start with the verdict
The report opens with its verdict: one sentence naming what every check found, worst first, then what the element checks found in more detail. Under it, Needs attention lists each check that found something, with its number, and Passed lists the rest. Each one jumps to its section, and the sections that found something come first.
Beside the verdict, Watch the reel opens the audit as a short video on a page of its own: each top issue on its page, outlined where it is, with who it affects. It’s a quick way to show whoever makes the fixes what’s wrong, and the reel’s page has a link to send it on its own (see Watch a run as a reel). When the audit has run before, a few lines there say what moved since, with a link down to every change.

Accessibility issues
Each failed rule is one row, numbered, and grouped by how badly it affects people:
- Blocks some people from using the page.
- Makes the page very hard to use for some people.
- Makes the page harder to use for some people.
- A nuisance for some people.
Open a row to see the page with that rule’s elements outlined, what to fix, and the failing markup on every page it fails on.

- Copy fix prompt
- Copies a prompt with the rule, the pages, and the failing markup, ready to paste into a coding assistant. “Copy a prompt for all issues” copies every rule at once.
- How to fix
- Deque University’s page for the rule, with examples.
- Needs a manual check
- Rules the check couldn’t decide on, like contrast over an image. Look at these yourself.
What search engines see
The Search section reads what each page says about itself, the way a search engine does, and holds it to a fixed list of rules. Each failed rule is one row, grouped by how much it matters:
- Can keep the page out of search results, or send searchers to the wrong page.
- Makes the page show up worse in results than it could.
- Worth a look when tidying up.
Open a row to read why it matters, what to do, and every page it fails on with what the page says there: the title two pages share, the address a canonical link points at, the robots tag that hides a page.

- Hidden from search engines
- A page the site links to that carries a noindex instruction, so search engines leave it out. Right for a staging site; a mistake on a page meant to be found.
- Titles and descriptions
- A page with no title or description, two pages that share one, a title over 60 characters, or a description outside 50 to 160. “Every page’s title and description” at the end of the section lists them all in one table.
- Canonical links
- A page with none, one whose canonical points at another page, or one that names two.
- The site’s pattern
- When most titles end the same way (“ | Northwind”), the ones that don’t are flagged, like a button that breaks the site’s style.
- Also checked
- More than one H1, missing or incomplete social preview (Open Graph) tags, structured data that doesn’t parse, no viewport tag, and pages with under 100 words of content.
- As a search result
- Open any page in the Pages grid to see its title, address, and description laid out the way a results page shows them, with its own issues and what the check read underneath.
- Copy fix prompt
- Like the accessibility rows: a prompt with the rule, why it matters, and every page with what it says, ready for a coding assistant. “Copy a prompt for all issues” covers the whole section.
Layouts
Layouts says how many layouts the site uses and how many variations of them. Each layout gets a card with a wireframe of its usual frame, the pages that use it, and its variations, each with what sets it apart (“a sticky header, content 1216px wide”). Pages are read at desktop width.
- Landing page
- A hero with the page’s headline, then sections stacked under it.
- Sidebar and content
- A navigation column beside the content, as docs and apps use.
- Grid of cards
- A short introduction over cards laid out side by side, as pricing and overview pages use.
- Article
- One narrow column of text, made for reading.
- Form page
- The site’s header over one main form, such as signing up, contact, or settings.
- Stacked sections
- Sections stacked down the page, without a hero.
- Focused page
- No site header: the page is one task, such as signing in.
Above the cards, the section lists where layouts drift: a content width within 64px of the site’s usual one, a header a few pixels taller or shorter than the rest, a header that scrolls away where most stay put, or a header or footer missing where most pages have one. Differences of 8px or less are minor. Drift isn’t counted in the verdict, so Layouts comes after the checks that found something. Under the cards, Sections the pages use counts each kind of section (a split hero, a 3-column grid, a table) and where it appears.
Styles that break the pattern
Each kind of element gets its own section, with one card per element, or set of elements, that’s off the same way.

- The problem
- In words at the top, like “Fully rounded corners: 1 of 10 buttons this size. Most use 6px.”
- Compared with
- The site’s usual style beside the one that breaks it, each with its CSS selector so a developer can find it in the code.
- The traits table
- Only the traits that differ, with the odd value highlighted.
- Find it on the site
- Opens the live page scrolled to the element’s label, or jumps to the page’s screenshot in the report.
Page loads
After the element sections, Page loads lists only the pages that stood out as they loaded. A page that loaded cleanly, at about the speed of the rest, isn’t listed.

- Errors
- The page was an error page (like Page not found), the page loaded but says it doesn’t exist (the row names the link that led there, which is likely broken), one of the site’s own images, scripts, styles, or fonts didn’t load, or a script failed. Another site’s ad or tracker doesn’t count.
- Slow
- The page took at least 1.5 times as long as a typical page on the site to show its main content, and over 2.5 seconds.
- Jumpy
- Its layout shifted as it loaded, so text or buttons moved under where someone was about to read or tap.
- Load times
- Open a page to see its own times under the screenshot: main content, first paint, the server’s answer, and layout shift.
Every page scanned
The report ends with every page it scanned. Failing elements are outlined in their impact’s color and numbered like the rules above. Select a page to open it, with its issues listed underneath.

When pages go unchecked
An audit stops scanning after about seven and a half minutes. A page it reached but couldn’t check is marked Not checked, and the report says why: the time limit, a page with more than 25,000 elements, or 3 pages in a row that failed to check. Scan fewer pages to give each one more time.
Before checking a page, an audit declines its cookie banner (never accepting it) and closes newsletter, promotion, and other popups. The report lists the pages where it declined a banner, and any page a popup still covered, since what was behind it may be missing from the results.
Some sites serve one page at two addresses, such as / and /home. An audit scans that page once and lists the other address as left out, so its issues aren’t counted twice.






