Questions and answers

Yes. The prompt and the skill file are both free and MIT licensed. There's no account to create here, no email to hand over, no usage cap and no paid tier waiting behind the first audit. You need a Claude account, which has its own free tier.

No, but it helps. A full audit reads every content page and builds a report and code bundle, which takes a lot of messages. On the free tier, ask for a quick audit — homepage plus five key pages — or say "continue" when you hit a limit.

Neither. This is deliberately the simple kind of tool. If you can open Claude and type, you can run it. There's nothing to configure, no Python to install and no terminal involved.

A quick audit is usually done in a few minutes. A full audit depends on the size of the site, since it reads every content page, runs the market check and builds the report, so a larger site can take considerably longer. You'll see progress as it works through its task list.

A full audit reads every content page it can find from your sitemap, navigation and links, skipping legal, login and thank-you pages. A quick audit reads the homepage plus up to five of your most commercially important pages.

A report as both a Word document and a PDF, and a zip of the code: every JSON-LD block, every HTML or config snippet, and a README mapping each file to its fix number. You also get a short summary in the chat with the scores and the top three fixes.

The files need file creation switched on in your Claude settings.

Yes, it works on any publicly reachable site. Running it on two or three competitors and comparing the reports is a genuinely useful exercise, particularly for spotting content gaps.

Yes, and it tracks progress fix by fix. Ask it to re-check the site, and give it the earlier report or code bundle if it's a new conversation. It marks every fix number as done, partly done, not started or regressed, adds anything new, and shows how each score moved.

Yes. The technical checks are language-independent, and the content assessment works in any language Claude handles well. Ask for the report in whatever language suits you.

For anything visible in your page source — titles, headings, markup, links, alt text — it's reading the actual code, so it's as accurate as the fetch. For content judgement it's an informed opinion rather than a measurement, which is true of any audit including a human one.

The scores are a triage heuristic for deciding what to fix first. They are not a Google metric and nobody should present them to a client as one.

The instructions specifically forbid it from reporting anything it hasn't observed, from estimating numbers it can't measure, and from flagging something as missing before checking whether it exists elsewhere on the site. It's also told to say plainly when a page renders with JavaScript and it can't see the content.

That said, verify anything surprising. If it says your homepage has no H1, view the source and confirm before rebuilding a template.

Only for FAQs already visible on a page, copied word for word. It doesn't promise a rich result from it: Google stopped showing FAQ rich results for most sites. The facts page has the dates.

The audit runs inside your own Claude conversation. Nothing is sent to us, because there's no server on our side — this site is a static page that hands you a text file. Your data is governed by Anthropic's terms for your Claude account, the same as any other conversation.

No analytics, no cookies, no fingerprinting on this site. You can verify that by opening your browser's network tab.

Yes. MIT licensed, so you can use it commercially, modify it, and build it into your own process. If you're charging for the report, do the obvious thing and verify the findings before sending it.

Different jobs. Those tools own large proprietary datasets — backlinks, keyword volume, rank tracking — and Screaming Frog crawls far deeper and faster than this ever will. None of that is something a page fetch can replicate.

What this does instead is read your actual pages with judgement, explain the problems without jargon, and write the corrected code. If you already pay for a crawler, this complements it. If you don't, this covers more ground than you'd expect for nothing.

Two ways. Most stop at the report; this writes the fix and tells you where it goes on your platform. And most of the powerful ones require Claude Code with Python and a terminal, which rules out most of the people who need an audit. This runs wherever you already use Claude.

For a small site with obvious problems, run this first and fix what it finds — you'll get most of the available value for nothing. Bring in a person when you need competitive strategy, link building, content at volume, or a technical migration. Those are judgement and labour, not checks.