SEO audit

Can AI search reach
your client's site?

A crawl of the client's own website that starts with AI search: which AI crawlers robots.txt and the firewall let in, which cited pages are broken, what a crawler reads in the HTML. Then the classic SEO checks, every issue ranked by what the pages it touches are worth.

Pages are included in the plan: nothing is charged per page and there is no overage. It reads the HTML your server sends and does not run JavaScript.

Start 14-day trial Read the documentation

AI search first, then the rest

Most audit tools start with titles and broken links. This one starts with whether AI search can reach and read the site.

AI crawler access, tested

robots.txt is matched page by page against the AI search crawlers and AI fetchers. Then the audit asks for the home page once with the published user agent of each of 8 AI crawlers, which catches a firewall or CDN that refuses a crawler robots.txt allows. A refusal is reported as a refusal, with the layer when it can be named. Connect your server logs and it also shows where verified AI crawlers were refused in the last 14 days.

Pages that AI answers cite

A page that AI answers cited in the last 30 days and that now returns an error, redirects, is set to noindex or is closed by robots.txt is listed first. This needs AI visibility tracking for the client.

What a crawler reads in the HTML

Pages with almost no text in the HTML, an empty app shell, nosnippet directives, hidden instructions aimed at AI systems, missing structured data on key pages, a missing or broken llms.txt, and whether pages offer a Markdown version.

Ranked by what the pages are worth

Every page gets a value tier from its Google clicks (Search Console, 28 days), the AI answers that cite it (30 days) and its tracked keyword rankings. Issues sort by priority, then clicks, then pages. The figures describe the affected pages; they never promise what a fix would recover.

Two scores, never averaged

Site Health and AI Search Health are each the pages without errors out of the pages evaluated, shown with both numbers and the crawl's date. A page we could not measure is not evaluated, and notices never move a score.

Every check shows its basis

Each of the 189 checks says whether it rests on a standard, a platform's documentation, an emerging convention or a heuristic, and links its source. Correct configuration is never a finding, and a check that could not run says so instead of passing.

And the classic checks

The classic SEO audit is in there too

Status codes and redirect chains, canonical tags, titles and meta descriptions, headings, internal links, sitemaps, structured data, HTTPS and certificates, duplicate content and page weight.

Orphan pages found from sitemaps, Search Console, GA4 and AI crawler logs, not from the crawl alone.
Google's own verdict on the most valuable pages (URL Inspection), when Search Console is connected.
Hreflang is judged only on sites with more than one language: a one-language site is never flagged for lacking it.

From finding to fix

The audit sits in the workspace the agency already runs, so a finding does not end in a PDF.

Crawl against crawl

New, fixed and still open issues between two crawls, shown only when both had the same scope. Fixed means checked again and passing.

Signals, not another dashboard

After a crawl the audit can raise Signals with both dates and the pages evaluated: new errors on key pages, a drop in Site Health, an AI crawler blocked, a cited page that broke. Automations can act on them.

Tasks and drafted fixes

Turn an issue into a task in the Approvals queue with an owner and a due date. Draft the fix with Crunch writes the tag, JSON-LD block, robots.txt lines or redirect rule for one issue on one page, from what the crawl stored, only when you click.

In the client report

Site Health, AI Search Health, the top issues and AI crawler access are a report source for the builder, on a share link and in the PDF, under the agency's brand.

From your assistant

12 MCP tools read the audit, compare crawls, change settings, mute issues and create tasks. Starting a crawl is behind the same opt-in as actions that spend AI.

In four languages

The explanation and the fix advice of every issue are written natively in English, German, Spanish and French.

Polite by design

The crawler identifies itself as CrunchJunkieSiteAudit, reads robots.txt before any page and obeys it, with no override. It sends at most one request a second, follows a Crawl-delay of up to 10 seconds, and stops when the site refuses more than 20 % of recent requests or answers robots.txt with a server error.

How to allow or block our crawler

It crawls only the website that belongs to the client, after someone confirms they may audit it. Never a competitor's.
It keeps what an audit needs (status codes, titles, links), not the text of the pages. Crawls run from Frankfurt on our EU infrastructure.
A page the firewall turns away is reported as refused, never as broken.

Pages are included in your plan

Each workspace has a page allowance per billing period, shared by all its clients. A page counts when the crawler requests it; pages robots.txt closes, images, scripts and probes do not count. Nothing is charged per page and there is no overage: a crawl that does not fit the allowance left does not start.

Plan	Pages a month	Pages per crawl
Solo	5,000	1,000
Starter	25,000	5,000
Pro	100,000	5,000
Agency	250,000	5,000
Scale	1,000,000	5,000
Trial	1,000	500

Pages per crawl is the most one crawl can request. Scale is priced individually; its row is the default allowance.

What it doesn't do

One crawl requests at most 5,000 pages, and a client can be audited 5 times a day (UTC). It reads the HTML your server sends and does not run JavaScript, so text that only appears after scripts run counts as missing. It audits the client's own website only, never a competitor's. It does not estimate what a fix would recover in rankings or traffic. Accessibility checks and AMP validation are not included.

FAQ

Questions before the first crawl

No. The GEO audit scores a small sample of pages for AI-search readiness in seconds and is free. The SEO audit is a full crawl of the client's website with every issue ranked by what its pages are worth.

It sends at most one request a second, follows a Crawl-delay, slows down when a server answers 429 or 503 and never crawls one site twice at the same time. A crawl you did not start yourself exists only if someone switched a schedule on; schedules are off by default.

Only the website of a client in your workspace, and the first time someone confirms they are authorised to audit it for its owner. There is no public crawler and no crawling of competitors.

The page is reported as refused, not as broken, and a crawl stops when more than 20 % of recent requests are refused. To let it through, allow requests whose user agent contains CrunchJunkieSiteAudit. Our addresses change, so allow by user agent, not by IP address. The crawler page has the details.

No. It reads the HTML the server sends. A page whose text only appears after scripts run shows up with little text, and the audit says so; it does not execute your scripts.

It depends on the plan; the table above lists pages a month and pages per crawl. A trial gets 1,000 pages in all, up to 500 per crawl.

Every issue is explained, and its fix described, in English, German, Spanish and French. A report uses the report's own language.

What an audit needs: status codes, titles, headings, links and counts, not the text of the pages. Page-level data is kept for the last two crawls and older crawls keep their summaries. It is deleted with the client. We process it for you as a processor under our data processing agreement.

Run it next to the AI visibility figures for the same client: AI Visibility

See what AI search sees on your clients' sites

Run the first audit during the trial. Pages are included; the first run asks you to confirm that you may audit the site.