On-page SEO checker

The tool requests the address once, with a plain crawler's user agent, follows the redirects if there are any, and reads the HTML that comes back. It does not run scripts. That is deliberate: it is the same view Googlebot gets on its first pass and the only view the answering AI crawlers ever get, so what the tool cannot see, they cannot see either.

• Title tag: present, and its length against the sixty or so characters a results page shows.
• Meta description: present, and its length against the hundred and fifty or so a snippet shows.
• One h1: the page states one subject, not none and not three.
• Heading order: h2 under h1, h3 under h2, no level skipped, more than one heading on a long page.
• Canonical: declared, and whether it points at this address or somewhere else.
• Indexable: no noindex in the meta robots tag or in the X-Robots-Tag header.
• Visible text: how many words the HTML holds, and what share of the file they are.
• Readable without JavaScript: whether the text is in the HTML or arrives later by script.
• Image alt text: how many images, how many with no alt attribute, and which ones.
• Links: internal, external and nofollow, counted.
• Open Graph: title, description and image for shares.
• Structured data: the JSON-LD types found, or the block that does not parse.

Below the table the tool prints the facts it read: the title and description in full with their counts, the canonical, the robots directives, the html lang attribute, the heading outline in order, the first five images without alt, and the hreflang versions declared.

Most of what decides whether a page ranks is on the page: a title that says what it is, a description that earns the click, headings that carry the outline, text a crawler can read, links that hand standing on. A crawler does not know what you meant; it knows what the HTML says. When the title says one thing and the h1 another, when the canonical points at the staging copy, when a template shipped with noindex on, the page loses quietly, and nothing in analytics says why.

The AI crawlers raise the stakes on one check in particular. OAI-SearchBot, PerplexityBot and ClaudeBot fetch the HTML and read it as it is; a page whose text is rendered by JavaScript after load is empty to them, however good it looks in a browser. The readable-without-JavaScript check is the one to look at first if you want to be cited.

Each check ends in one of three words. Pass means nothing to do. Look means the page works but something is off the usual measure, a title of 25 characters, a canonical pointing elsewhere, a page with two headings. Fix means a crawler is losing something: no title, noindex on a page that should rank, an empty HTML shell, most images without alt. The verdict at the top names the three to fix first, failures before warnings.

A worked example. yourproduct.com/pricing comes back with a 71-character title that repeats the site name twice, a description of 40 characters, two h1 tags because the plan cards each carry one, a canonical pointing at yourproduct.com/pricing?ref=nav, and 180 words of visible text. Nothing fails; five things say look. The order to take them: the canonical (a parameter address is claiming the page), then the two h1 tags (the template), then the title and description (five minutes in the head), then the text, which for a pricing page is a judgement rather than a rule.

1. 1

   Fix the checks that fail, in the order given

   A missing title or a stray noindex is a five-minute change with a large effect. An empty shell is a rendering decision, the one change that may take a sprint, and the one that matters most for AI answers.

2. 2

   Take the warnings that come from the template

   Two h1 tags, a skipped heading level, images without alt in a component: each is one edit that fixes every page built from the template. Check a second page of the same kind to confirm.

3. 3

   Rewrite the title and description last

   They are the easiest to change and the easiest to change badly. Say what the page is, for whom, in the words the searcher used; keep the site name once, at the end.

4. 4

   Run the check again after the deploy

   A cache in front of the site can serve the old HTML for a while. When the tool reads the new tags, the crawlers will too, on their next visit.

A fixed page reads like this: one title of about 55 characters that names the page and the product, one description of about 140 that says what the reader will find, one h1 that agrees with the title, h2 sections in order, a canonical pointing at itself, no robots directive, the text in the HTML, every image described, a dozen internal links, Open Graph set, and one JSON-LD block that parses.

• One address per check, thirty checks an hour per visitor. The page must be public: the tool refuses private addresses and follows at most five redirects.
• The HTML is read as sent, up to two megabytes. Text added by JavaScript after load is not read, which is the point of the readable-without-JavaScript check and the reason the word count can differ from what you see in a browser.
• Length checks use character counts. Google truncates by pixel width, so a title of 58 characters in wide letters can still be cut; the title checker measures pixels if you want that precision.
• The tool reads what the page says about itself. Whether the title matches what people search, whether the text answers the query, whether the page has links from elsewhere: those are questions for the free check, which reads the site, its searches and its rivals together.