Technical SEO Audit

Point this Actor at a page and it tells you, in plain language, what is wrong with it. Not a wall of raw tags: a score out of 100 and a list of findings, each one graded by how much it actually costs you.

Most SEO tools give you a dashboard and leave the interpretation to you. This one does the interpretation. Every finding says what is wrong, why it matters, and what the number is. "The title is 75 characters; past about 60 it gets truncated in results." You can act on that without knowing what a canonical tag is.

What it checks

Whether the page can be indexed at all. This comes first, because everything else is decorative if the answer is no. Three separate switches can keep a page out of a search index and the Actor checks all three: a noindex in the robots meta tag, the same directive arriving as an X-Robots-Tag response header, and a canonical tag pointing at a different URL. The header case is the one people miss, because it is invisible when you view source.

The tags that decide how you appear in results. Title and meta description with their lengths measured against the point where search engines cut them off. Canonical URL, resolved to absolute and compared properly, so a canonical of https://example.com on the page served at https://example.com/ is correctly read as the same address rather than reported as a conflict.

Structure. Heading counts by level, the text of every H1, and whether the outline skips a level. Word count taken from visible prose only, with scripts and styles excluded, plus the text-to-HTML ratio.

Structured data. Every JSON-LD block is parsed and its schema.org types listed. Blocks that fail to parse are reported individually with the line the JSON breaks on. Broken structured data is invisible on the page and silently costs you rich results, so it is worth catching.

Images and links. Images with no alt attribute are counted separately from images with a deliberately empty one, because an empty alt is correct for decoration and flagging it would be wrong. Images with no declared width and height are counted too, since those cause the layout to jump while a page loads. Links are split into internal and external, with nofollow counted and empty anchors flagged.

The full redirect chain. Every hop is recorded with its status code and where it points. A page reached through three redirects wastes crawl budget, and you cannot see that from the final URL alone.

Security headers. Six standard protective headers, each reported present or missing, with a count.

robots.txt, sitemaps and llms.txt. Whether robots.txt exists, how many rules it carries, which sitemaps it declares, and whether those sitemaps actually resolve to real URLs. A sitemap is only counted when the response carries a real sitemap root element, so a site that answers unknown paths with its ordinary page shell is not credited with a sitemap it does not have.

Which crawlers your robots.txt lets in. Search crawlers and AI crawlers are reported separately, because they are separate decisions. A site can be perfectly open to search and closed to model training, and that is usually deliberate rather than a mistake, so it is reported as something to know rather than as a defect. Blocking search crawlers, by contrast, is treated as critical. The allow-or-deny answer comes from the same rule matching a real crawler uses, including group selection and wildcards, rather than from reading the Disallow lines and hoping.

The certificate. Issuer, protocol version, cipher, expiry date, days remaining, and whether the certificate actually covers the hostname you asked for. An invalid or mismatched certificate is critical; one expiring within a month is a warning.

DNS. A, AAAA, MX, NS, TXT and CNAME records, plus whether the domain publishes SPF and DMARC. Missing SPF or DMARC makes your domain easier to spoof, which is worth knowing even when you came here for search rankings. A lookup that never answered is reported apart from one that answered empty, so "no records" never gets confused with "we could not ask".

Scoring

Each page starts at 100 and loses weight for every finding: 15 for a critical, 5 for a warning, 1 for a notice. Results are sorted worst first, so the page that needs attention is the first row you see.

The three severities mean three different things:

• critical stops the page being indexed or served at all
• warning costs you ranking or clicks
• notice is worth knowing but is not a defect

Input

Give it a list of URLs. A bare domain such as example.com is accepted and audited as its homepage. You can also set how many pages to fetch at once, how many times to retry, and optionally which country to request from when a site serves different content by region.

Auditing many pages of one site is much cheaper than auditing many sites. robots.txt, the sitemaps, DNS and the certificate belong to the site rather than the page, so they are looked up once and shared across every page on that origin. In practice fifty pages of one site cost fifty page requests plus one set of site lookups.

Output

One row per page, containing every fact listed above plus:

• score, and counts of critical, warning and notice findings
• findings, each with a severity, a stable code and a readable message
• internal_urls, the internal links found on the page, which you can feed straight back in as the input for a wider audit

A page that could not be fetched is still a row. It carries the error and a critical finding rather than quietly disappearing from the results, because a missing row is the one thing an audit must never produce.

The run summary holds the counts, the average score and the totals by severity.

Notes

Only what the server sends is inspected. Content that a page builds in the browser after loading is not part of the audit, so a fully client-rendered page will honestly report the small shell that was actually served. That is usually the finding rather than a limitation: it is also what a search engine sees on its first pass.

Retries apply only to timeouts and server errors. A 404 is an answer, and it is reported as one.