How to audit a page step by step with Search Toolbox

Open the page that actually has the problem: production, staging, localhost, a logged-in page or a parameterized URL. Let it finish loading, then record the URL shown by the browser.

Why? Because auditing the wrong variant produces a wonderfully tidy and completely useless diagnosis. Return to this exact URL after every change.

Open Overview and start with HTTP status, robots and canonical. In plain English: does the page respond, allow indexing and name the correct URL as its main version?

If you find noindex, a canonical pointing elsewhere or an HTTP error, stop here. Those are blockers. Optimizing images first would be choosing curtains while the house is on fire.

In Headings, read H1, H2 and H3 elements like a book outline. The H1 states the main topic, H2s divide the main sections and H3s expand them.

Here the tool shows 1 H1, 27 H2s, 49 H3s and 8 H4s. There is no magic target number; the goal is a structure people and search engines can follow without guessing.

In Links, separate internal links from external ones. Then inspect anchor text: the clickable words should explain the destination instead of hiding behind “click here”.

The screenshot inventories 340 links: 330 internal, 10 external and no nofollow links. It proves what the Links panel checks without pretending to prove redirects or image alt text.

JSON-LD is a machine-readable description of the page. In Schema, inspect the detected types and look for duplicates: two competing FAQPage blocks can describe the same content twice.

First pass: six JSON-LD blocks and a duplicated FAQPage. After the fix: five blocks and one FAQPage. Both screenshots show the same panel before and after, so the visual is the evidence.

Hreflang connects language versions of the same content. Verify each language code, URL, HTTP response and return link.

The screenshot checks 59 annotations, an x-default target, their HTTP status and reciprocity. If your page has one language, skip this step: a good audit also knows when not to invent work.

The rendered DOM is simply the final page after JavaScript has run. Do not start here: open Render only when a tag is missing, duplicated or inconsistent with Overview.

The panel compares the JavaScript-rendered DOM, the no-JS output and the raw server source, then prepares a detailed gap analysis. This deeper check explains a difference you have already found; it should not complicate the first minutes of the audit.

Sort findings into three groups: blocks indexing, weakens the SEO signal, improves reading comfort. Fix them in that order. Impact matters more than the raw warning count.

Change one thing, reload the page and repeat the relevant check. The AI agent can help sort findings, but it cannot turn a screenshot from another page into evidence. That is why this step deliberately shows no misleading visual.