25. Page monitoring

Download change evidence

Under Detected changes, choose Export retained history as CSV to download up to 500 retained changes across all history pages, newest first. When viewing one change from an alert or direct history link, Export this change as CSV downloads only that change. Authorized project readers can export; editing permission is not required.

Each changed field has its own row with change ID, page URL, before-check and detection times in UTC, significance, field name, before/after values and an optional linked annotation ID. Times describe observations; the exact edit time between checks is unknown. Comparisons use the same rules as the screen, so missing HTML on errors/redirects is not treated as removed metadata. Booleans are Yes/No and an absent value is Not present. Empty history downloads column headers only. Multiline headings, commas and Unicode are preserved; spreadsheet formula-looking text is escaped using the existing CSV protection.

Exports read saved evidence without fetching pages, calling providers or changing watches/annotations. History outside the newest 500 retained changes is not available in this export.

Open a project → Improve → Page monitoring. Watch up to 50 important landing pages per project using exact HTTPS URLs on the project domain (including its www variant). Project editors manage watches; authorized readers can review saved evidence. The existing Site audit licence feature covers these checks.

From Pages → page details, choose Monitor this page for weekly SEO changes to prefill its URL, then explicitly select Watch weekly.

Weekly checks

Choose Watch weekly. Cron queues new pages at the next hourly dispatch, then checks each watch every seven days. The dispatcher queues at most 50 due pages per hour across the installation. One bounded background job checks one page, respecting robots.txt, public-address protections and the existing crawler's response-size limits. Reads and watchlist changes do not fetch the website during the browser request.

The first complete check establishes a baseline and does not create a change alert. Subsequent checks compare the title, meta description, first 20 H1 texts and total H1 count, resolved canonical, noindex (HTML and HTTP header directives), and HTTP status. Text is bounded to the crawler's existing limits. JavaScript-generated content is not rendered; redirects are recorded without following them. Metadata on non-HTML responses is unavailable rather than evidence that a title or canonical was removed.

Robots blocks, network failures and truncated responses retain the last complete evidence and show a check message. An HTTP error response such as 404 or 503 is observed status evidence and can be compared. A stopped watch is no longer checked; its detected change history remains. Re-adding it establishes a new baseline and invalidates old queued work. If the project's domain changes, old watches show a message and must be replaced.

Before and after

Expand Latest saved evidence on a watched page. Detected changes shows the values before and after, with the two check timestamps in UTC. History retains the newest 500 changes per project, displayed 25 at a time. The actual website edit time between observations is unknown; an observed change is not proof of an SEO problem or traffic impact.

Canonical, noindex and HTTP status changes count as significant. They feed the existing deduplicated alert system, subject to your personal alert preferences and project access; configured team webhooks use their existing choices. Title, description and H1 changes remain visible in history without sending routine text-change alerts. Under Notifications, choose Important page SEO settings changed to enable or mute this alert type.

Record a change and compare impact

Review the detected change and enter the actual edit date if known. The form defaults to the detection date in UTC; comparisons using that date are approximate context. Choose Record annotation and compare impact to save an exact-page annotation containing the observation interval and a bounded before/after summary, then open the existing impact comparison.

Repeated clicks reuse the linked annotation rather than creating duplicates. If it was explicitly removed, the action can recreate it. The impact report uses the existing equal 14/28/90-day windows, data coverage checks and Google/GA4 evidence. It may be unavailable until enough time and imported data exist. Association does not prove causation.

No page is edited or published by this feature. It adds no paid provider calls. Existing read-only AI annotation and impact tools can read the annotation after you save it; there is no new monitoring write tool.