Page audits

A page audit checks one page of your website the way a careful expert would, and shows you every problem it found, worst first, with a screenshot and a fix for each. Your own AI agent does the checking on your computer, using UX Robot’s checks; UX Robot measures the page’s speed and puts it all in one report.

• Accessibility: every WCAG 2.2 level A and AA success criterion that can be tested in a browser.
• Usability: UX Robot’s checks of how well the page helps visitors find, understand, decide, act and recover from mistakes.
• Search: UX Robot’s search checks from Google’s own guidance, kept up to date every week: whether Google can reach and index the page, how it appears in results and whether it answers the searches that bring people to it. See Search checks.
• Speed: Chrome’s real-visitor data for the page or website, and a Lighthouse test of the page.
• Your real traffic (optional): with GA4 connected, the page’s views, sessions, engagement and key events, or your primary goal, over the last 28 days.
• Your searches (optional): with Search Console connected, the searches that bring people to the page, with their clicks, impressions, click rate and average position, over the same 28 days.
• Your data in your agent’s brief: with GA4 or Search Console connected, UX Robot gives your agent the page’s traffic and the searches that bring people to it. Your agent checks whether the page answers those searches and weighs each problem by how many people meet it. It never changes whether a check passes, and your agent never opens GA4 or Search Console itself.
• Microsoft Clarity (optional): your agent reads the page’s click, scroll and attention heatmaps and its errors and dead and rage clicks.

Page audits are part of every plan. Free accounts get five a month; Pro and Enterprise accounts as many as they need.

1. Set-up connects your agent when you create your account. Otherwise, open Settings → Desktop connections, add UX Robot’s address to your agent’s app as the page shows, and sign in to UX Robot when it asks. See Coding agents for each agent.
2. Use an agent that can control a browser and run commands on your computer, such as Codex with browser access or Claude Code with Claude in Chrome, in its desktop app, an editor such as VS Code, or the terminal (see where your agent can run). We recommend GPT-5.6 and higher or Claude Opus 5.5 and higher: we have run full audits with these models, and newer versions do just as well. Older or other models can work too, but they are more likely to stop partway or miss a check; UX Robot does not accept unfinished work, so you may need to ask them to carry on. Each audit names the model your agent used.
3. Optionally connect GA4 and Search Console under Data sources, so each audit shows the page’s real traffic and the searches that bring people to it.
4. If you tick Clarity, sign in to Microsoft Clarity in the browser your agent controls.

1. Open your organisation’s Page audits.
2. Enter the full address of the page and choose the device. It is usually a page on your website, such as your pricing or sign-up page, but it can be on any website, such as a competitor’s or a client’s. For a page on another website your audit has every check and screenshot, but not your GA4, Search Console or Clarity, which only cover your own website and its subdomains, so Clarity is left out even when ticked.
3. Tick We use Microsoft Clarity if you want your agent to read it. It starts ticked when you told set-up that you use Clarity.
4. Tick This page needs me signed in for a page behind a sign-in, such as an account or dashboard page, and sign in yourself in the browser your agent controls before it starts. See Audit a page behind a sign-in.
5. Select Start page audit. A Your turn window opens: select Copy the instruction and paste it into a new chat in your agent. The window also says which models work best and that the desktop app is the easiest place to run your agent. The audit cannot start without your agent, so the window stays until your agent claims its first task, then closes by itself. If your agent is not connected yet, the window starts with Connect your agent. You can leave the page; the audit waits for you.
6. Your agent claims each task, checks the page, uploads its screenshots and submits the results. If Clarity is signed out, it stops and the page opens a Your turn window asking you to sign in; it closes once your agent carries on. The window also opens when your agent needs you for anything else, such as allowing a browser permission, with its request in its own words, and when your AI app must reload UX Robot’s tools, with the steps for your app. Wherever you are in UX Robot, Your agent is waiting for you appears at the top of the page with a link back to the audit.

While your agent works, the audit page shows its progress: a bar that moves with the checks, how long it has been working and when it last saved, a card for each step with what it has done so far (such as 23 of 55 WCAG 2.2 criteria checked and 2 issues found), and your agent’s latest update in its own words. If it needs you, for example to sign in to Microsoft Clarity, the page says so.

When your agent finishes, the page shows the report straight away and says so, and UX Robot emails you and the organisation owner with how many problems it found, the one to fix first and a link to the audit. You do not need to keep the page open. You can also ask your agent to check UX Robot for waiting work, and it picks the audit up with list-waiting-ux-work.

If no agent has picked up your audit a day after you started it, for example because you skipped connecting one during set-up, UX Robot emails you once with how to connect your agent, the instruction to paste into it, and a link to invite your developer to run it instead. Running it then uses no extra page audit.

A page audit can check what a signed-in customer sees, such as an account, dashboard or settings page, on every plan.

• You sign in, your agent does not: sign in yourself in the browser your agent controls. Your agent checks the page in that session as a signed-in user. It never types a password, signs in or out, changes a setting or submits a form.
• If you are signed out: your agent saves its place and the page opens a Your turn window, Sign in to the website being tested, with Open the page. Sign in, leave the page open and tell your agent; it carries on from where it stopped.
• Your account details stay out of the screenshots: your agent scrolls names, email addresses, orders and balances out of view, or records the problem in words when a screenshot would show them.
• What UX Robot cannot measure: UX Robot cannot sign in, so it does not measure the page’s speed or read it as Google does. Instead, your agent checks the page’s title, description, canonical, headings, images, links and structured data in your signed-in browser, and every search check is in the fix list. GA4, Search Console and Clarity still work, because they come from your own accounts.
• Its own issue history: a page checked signed in keeps its issues separate from the same address checked signed out, and the report and the list of audits say Checked signed in. Audit this page again keeps the box ticked. On Enterprise, a deploy’s spot check of these issues asks you to sign in in your agent’s browser before it checks them.

• What to do next: fix the worst problems first, hand the audit to your developer by inviting them, audit the page again once the fixes are live, and audit the next page your customers need.
• Since the last audit: on Pro and Enterprise, when you audit a page again, which earlier issues are verified fixed, still there or came back, and which are new, each linked to its issue. A speed issue Lighthouse stops flagging reads Looks fixed, confirming until the next audit agrees, unless your team has marked it shipped.
• Written by your agent: a short summary of what the page does well and the most important problem to fix first, with the model your agent reported.
• This page in GA4: views, sessions, the share of engaged sessions and key events, or your primary goal’s completions, for the audit’s period.
• Searches that bring people here: the page’s clicks and impressions from Google and its ten most clicked searches, from your own Search Console, for the audit’s period. Search Console files a page under the address Google indexed, so the page is matched with or without www and a trailing slash. Google leaves out rare searches to protect searchers’ privacy, so the searches add up to less than the totals.
• Speed: Chrome’s real-visitor measures and the Lighthouse results, when Google has them. A page checked signed in says instead that UX Robot could not measure them.
• Fix list: every issue, worst first, each with its severity, screenshot, what is wrong and exactly what to change. Each issue says whether it is Measured, proved from the page’s code or a measurement anyone can repeat, such as a contrast ratio, or the Agent’s judgement, an expert review of how visitors would use the page that another reviewer could see differently. Lighthouse’s speed problems and UX Robot’s own search checks are always measured. Under How to see it, each issue your agent found lists the steps to see it yourself: open the page at the size it was tested, then any clicks or keys your agent needed, such as opening the menu. Speed problems are explained in plain words, with the time or download Lighthouse estimates the fix saves and the files it names. Caching problems are never marked high, because a longer cache only speeds up a returning visitor's next visit.
• What we checked: every check and criterion with its result: no issue found, issue found, not applicable, or not safe to test, with the reason. A check that the page’s code or a measurement settles says No issue found when it finds nothing; one that rests on judgement, such as whether a heading describes its section, says Agent’s review: no issue, because it is your agent’s review, not a pass. Neither is a WCAG conformance claim, as the report says beside the table. Each area says how many of its checks found an issue, so you know which to open.
• What your agent saw in Clarity: its notes on the heatmaps and insights, as a sample of real visits rather than research with visitors.

• Free accounts get five page audits a month, counted across the account from the first of each month (UTC). Auditing a page again counts as another audit. Pro and Enterprise accounts have no limit; Pro is USD $249 a month.
• A page audit costs nothing on any plan and never uses an analysis unit.
• One page per audit. Your agent never signs in to your website itself: for a page behind a sign-in, you sign in. It does not follow a journey beyond the page.
• It is a browser check of the success criteria, not a conformance audit, and the write-up is only as good as the model your agent runs on.

• Your agent acts with your account’s permissions through your desktop connection, and only for organisations you can edit.
• It uses safe, reversible actions only. It never signs in to your website or out of it, submits a form, buys anything or contacts anyone, and never asks for or types a password. On a page behind a sign-in it uses only the session you signed in to, and keeps your account details out of its screenshots.
• Screenshots of your page are stored in your workspace like any journey’s evidence. Clarity is never screenshotted; your agent records its figures in words.
• GA4 and Search Console are read with the read-only access you granted when you connected them, and each page’s figures are read at most once a day.