Which page to fix this week, and whether it worked.

Google’s side beside yours

For every page, the report shows Google’s figures (searches, times shown, clicks, average position) and StepMetrics’s (the visits Google sent, how many stayed, how many took the next step you count as a conversion). A page Google shows 4,000 times that converts nobody is a different problem from a page shown 300 times that converts a fifth of its visitors, and neither tool on its own can tell them apart.

Connecting is read-only and takes one click on the Search page. Each site is matched to its property by domain, 90 days are copied, and the copy is refreshed every six hours. Days are Pacific Time on both sides, because that is how Google counts. Google’s clicks and your visits from Google are kept apart rather than forced to agree: a click from a browser that blocks trackers is a click with no visit.

Five suggestions

Four compare a page with the same site’s own average, one compares a page with itself in the window before. Each needs minimum evidence before it is shown (200 times shown or 20 visits by default) and comes with the figures behind it and a first thing to try.

• Just off the first page. A search that shows your page at an average position of 8 to 20. Searches for your own name are left out.
• Seen but rarely clicked. A first-page page whose click rate is under half the site’s own first-page rate. The title and description are the place to start.
• Searchers leave. A page whose share of engaged Google visits is under half the site’s.
• Searchers never convert. A page that converted none of its Google visits where the site’s rate predicts three or more.
• Losing clicks. A page with 40% fewer clicks than the window before, naming the three searches that fell most and whether position or the number of times shown moved. Withheld until both windows are complete, because a window Google has not finished reporting would read as a fall that is only missing days.

They say where to look, not why. A change made after one is not proven to have caused what follows.

Searches for your name, kept apart

Tell StepMetrics the words that mean someone searched for you by name, and the report splits into branded, the rest, and what Google reports in the totals but lists under no search. Ranking low for your own name is a different problem from a page to tune, so a branded search is never a “just off the first page” suggestion.

Whether the change worked

Mark a finding as changed when you ship the fix, with a note. StepMetrics then compares Google’s clicks, times shown, click rate and position for the 14 and 28 complete days before and after, leaving the deployment day out. The review is observational and says so: a difference is shown as a difference, a Google update that lands inside the window is named as an overlap, and a window Google has not finished reporting shows “awaiting data” rather than a number. The mark is your own report of a change, not a verified fix, and the original finding stays on record.

Google updates on the chart

Google’s confirmed core and spam updates sit on the Search and Overview charts as markers, so a move that coincides with one is read as a coincidence first. Your own deploys, content and campaigns go on the same chart as annotations, at workspace or site scope.

Technical checks

An audit reads up to a hundred of your pages the way a crawler does, with no JavaScript and no cookies, and reports what would keep Google from showing them: HTTP errors, redirect chains, a robots rule blocking Googlebot, a noindex that reached the sitemap, missing or duplicate titles and descriptions, a missing H1 or canonical, broken JSON-LD, pages with few internal links. It asks Google’s URL Inspection about up to twenty pages, reads any sitemap errors Google reports, and adds Core Web Vitals from your real visitors. Run one when you like, or weekly.

Findings carry severity, the URL, evidence and a first action. There is no score. Each report says which findings are new since the last one and which are gone, and an alert goes out by email, Slack or a signed webhook when a new one appears or when the Search Console connection breaks.

Your agent

Over MCP, your coding agent reads the same report, suggestions, audits and reviews as the dashboard, from the same query functions. With an edit key it can run an audit, turn the weekly schedule on, mark a finding as changed, set brand terms and add an annotation. A read key can only look. Because the agent that reads the finding is the one that edits your repository, a suggestion can become a commit and a marked change in one conversation. What your agent can read and set up.

Limits

• Nothing here promises a ranking. Google decides those. This shows you where to look and what happened after.
• Device and country breakdowns of search data are not synced yet.
• Backlinks and competitor data are not part of the product.
• StepMetrics never writes to Search Console, submits a sitemap or requests indexing, and never edits your site.
• Google omits searches it considers too rare to report, so the query list sums to less than the totals.

Connect Search Console and read the report · How bot filtering works