← Blog
Changelog4 min read

Site audits now work on slow sites and show what to fix

OpenRush site audits now finish on small and slow websites, report fewer false alarms, and attach proof to every problem so an AI agent can fix it.

OpenRush Team
Marketing intelligence for AI agents

OpenRush site audits are deeper and more reliable, and every finding now says what to fix, at the same speed. They now finish on small and slow sites, report fewer false alarms, attach proof to every problem, and can re-check only the pages you fixed.

Audits now finish on small and slow sites

Before, a slow server could leave an audit with one or two of the 20 pages it tried to check. Audits now get nearly all of them. On sites that had failed before, the number of pages checked went from 0 to 1 up to 14 to 20.

An audit with two pages can't say much about a site. With most of the sample in hand, the results describe the site instead of one slow template.

Fewer false alarms

Two kinds of pages were being flagged wrongly, and both are fixed:

  • Translated utility pages: a shop's cart page in Dutch, French, German, Spanish, Italian, or Portuguese is no longer reported as a page that can't rank.
  • Non-Latin alphabets: pages written in non-Latin scripts are no longer skipped by mistake.

Pages meant to stay out of search, such as noindex pages, canonicalized duplicates, and non-HTML files, are still skipped. They are counted in excluded_summary so you can see what was left out.

Every problem comes with proof

Each issue now includes the evidence behind it:

  • The actual title that is too long
  • Which images lack alt text
  • The full redirect path
  • The broken link and the page it lives on

That evidence is what lets an AI agent fix the site straight from the report. It doesn't have to open the page and hunt for the problem, and you don't have to translate a finding into a file path.

New checks for problems people care about

The audit now catches:

  • Broken links inside your site
  • Pages that bounce through several redirects
  • Pages that respond slowly, and oversized pages
  • Important pages that nothing links to
  • Sitemaps that list pages Google is told to ignore

Slow responses are measured from the server on raw HTML. This is not a Core Web Vitals test.

Real fixes first, nice-to-haves last

Missing alt text and missing social-share tags used to crowd the top of the list. They are now grouped as polish and ranked last, so a broken link or a redirect chain isn't pushed off the first screen. The audit also labels each issue by page type (core, content, utility, or ephemeral), so a problem on your pricing page outranks the same problem on a cart page.

Check your fixes right away

Give audit_site the exact pages you changed and it audits only those. That makes a fix-then-verify loop short enough to run inside one conversation with Claude Code or another agent: the agent audits, fixes what the evidence points to, and re-audits the changed pages to confirm each problem is gone.

Fix the pages that bring traffic first

If you've connected Google Search Console, the audit suggests ranking its findings by real Google traffic as an optional next step. A missing title on a page that earns clicks then outranks the same issue on a page nobody visits. Connecting Search Console isn't required to run an audit.

Same speed

Most audits still finish in a few seconds, with no waiting or background jobs. The audit reads raw HTML and samples your site instead of crawling all of it, which is why it stays fast.

What the audit doesn't cover

An audit is a sample of up to 20 pages, one per URL template, so it shows patterns rather than every URL. It reads raw HTML only, with no JavaScript rendering. When some pages look client-rendered, the result includes a render_warning and those pages are less complete. Check coverage, discovery_source, and fetch_failures in the result before you trust a thin one.

Try it with your agent

Copy this prompt into your agent:

Use OpenRush to audit my site, list the most important problems with proof for each, fix them, then re-audit only the pages I changed to confirm they're resolved.

audit_site is part of the SEO MCP toolset. To connect Claude, ChatGPT, Codex, Cursor, or another compatible agent, follow the MCP setup guide. Tool prices are on the credits page.

Frequently asked questions

What does the OpenRush site audit check?
It checks on-page tags, redirects, slow responses, oversized pages, broken internal links, important pages nothing links to, and sitemap entries that conflict with a noindex or canonical. Each issue comes with the evidence needed to fix it.
How many pages does an audit cover?
A sample of up to 20 pages, 10 by default. The audit picks one page per URL template and puts money and content pages first, so a large site is represented without crawling every URL. You can also give it an exact list of pages.
How do I re-check the pages I fixed?
Pass the exact URLs to audit_site in the pages field. It audits only those pages, so a fix-then-verify loop with Claude Code or another agent takes seconds.
Does the audit run JavaScript or measure Core Web Vitals?
No. It reads each page's raw HTML. A render_warning tells you when some pages look client-rendered, and those results are less complete. Slow responses are measured from the server, not from Core Web Vitals.
Can an AI agent fix the problems from the audit?
Yes. Every problem includes proof, such as the title that is too long or the broken link and where it lives, so an agent can edit the right file without guessing. Then it re-audits the pages it changed.
Share

Give your agent live search data.

One-click setup with Claude or any MCP-compatible agent.