Search Console
The SEO app can connect to Google Search Console — Google’s own view of your site in search. Once connected, Overwatch joins that data onto the reports it already produces:
- A Search performance section on the SEO report: clicks, impressions, CTR, and average position over the last 28 days, a daily trend, top pages, and top queries.
- Traffic-weighted findings: pages in the SEO report show their search impressions, and broken links that still earn impressions are flagged — a dead page people still find in Google outranks a forgotten one.
- How Google sees your pages on the SEO report: Google’s own verdict on each crawled page from the URL Inspection API — whether Google can fetch it, reads it as a real page, keeps your canonical and indexes it — plus its rich-result errors and when it last visited. See Google’s view below.
Connecting
Section titled “Connecting”Search Console is connected per site, from Settings → Apps → SEO (owner/admin only):
- Connect Google Search Console sends you to Google’s consent screen —
sign in with an account that can read the site’s Search Console property
(the
webmasters.readonlyscope; Overwatch never gets write access). - Back in Overwatch, choose the property. Domain properties
(
sc-domain:example.com) are preferred over URL-prefix properties (https://example.com/) because they cover every protocol and subdomain — a URL-prefix property that doesn’t match the crawled host (e.g.www.versus bare) returns no data. - Data arrives with the next sync (within the hour).
Pick a property that covers the site’s own host. Google answers URL
Inspection only for URLs inside the property, so a staging site connected to
production’s property (say staging.example.dev on sc-domain:example.com)
still gets that property’s search traffic, but never Google’s view of its own
pages. The picker marks matching properties, warns when you choose one that
doesn’t cover the host, and the SEO report says so instead of waiting for
inspections. Overwatch skips inspections for such a connection, so it never
spends the property’s shared daily quota on calls Google would refuse.
The exception is a relaunch: a staging site that names the live site in Replaces is meant to read the live site’s property. Settings and the SEO report then say so instead of warning, and Launch readiness uses that traffic to check the new site before the switch.
The Google refresh token is encrypted at rest and never appears in any API response. Disconnecting revokes it at Google and deletes all stored search data for the site.
How syncing works
Section titled “How syncing works”- Search Analytics (clicks/impressions/position) syncs about once a day per site. Google’s search data lags roughly two days, so “last 28 days” ends two days ago.
- Each sync also stores the searches leading to each page (top 10 per page) and each page’s numbers for the 28 days before, for the per-page view CMS integrations show.
- URL Inspection is quota-limited by Google (~2,000 URLs per property per day), so Overwatch inspects a bounded batch every few minutes, cycling through the latest crawl’s pages oldest-first. A large site’s first full index-coverage picture builds up over a few days.
- If two Overwatch sites connect the same Search Console property (e.g. staging and production), they share that property’s inspection quota.
Search insights and alerts
Section titled “Search insights and alerts”The Search performance section computes a few opinionated findings from the synced data:
- Striking-distance queries — queries ranking between position 5 and 20 with real impressions, where targeted improvements can reach page 1.
- Low-CTR pages — pages ranking well but earning far fewer clicks than their position normally gets; usually a truncated or unappealing search snippet (check the title/description findings).
- Orphan winners — URLs receiving clicks that the crawl never found: removed or unlinked content still in Google’s index.
- URL-variant splits — the same page collecting traffic under http/https or www variants, splitting its ranking signals.
A search-traffic-drop alert (own recipient list under Notifications) fires
when a site’s weekly clicks fall at least 35% against the previous week, with
a minimum baseline so low-traffic noise never pages anyone. It’s
transition-based: one alert per drop, reset by recovery. Generic webhooks
receive a search.dropped event.
Google’s view
Section titled “Google’s view”With Search Console connected, the SEO report opens with How Google sees your pages: a funnel of the gates every crawled page has to pass to be found in search.
- Crawled by us — the latest crawl’s pages that Google has inspected.
- Google can fetch — Google’s last attempt to load the page succeeded.
- Reads as a real page — Google didn’t classify it as a soft 404.
- Keeps your canonical — Google honoured the page’s canonical (or, with none declared, kept the page itself).
- In Google’s index — Google indexes the page.
Each drop-off is a link: it switches the findings to the Google tab and opens the finding that explains it. Below the funnel, Rich results and Crawl freshness summarise structured-data verdicts and how recently Google visited.
Google’s verdicts become findings in the same severity bands as the crawl’s own checks, marked with a G:
| Finding | Severity | When |
|---|---|---|
| Google treats the page as missing | Serious | The page answers 200 OK, but Google reads it as a soft 404 — a dead link our crawler can’t see. |
| Google couldn’t fetch the page | Serious | Google’s last fetch failed: not found, server error, redirect error, access denied/forbidden, other 4xx, blocked by robots.txt, or an invalid URL. Google-side crawl errors aren’t flagged. |
| Google ignored your canonical | Moderate | Google chose a different canonical than the page declares (trailing slashes and www are ignored when comparing). |
| Structured data Google rejects | Moderate | A detected rich result has an ERROR issue, so it can’t show in search. |
| Structured data could be richer | Minor | Only WARNING issues — valid, but missing recommended fields. |
| Google hasn’t visited in 90+ days | Minor | Google last crawled the page 90+ days ago. Only pages meant to rank on their own count: not noindexed, not canonicalised elsewhere. |
| Google hasn’t crawled the page yet | Minor | Google has never fetched the page (unknown to Google, or discovered and queued). Same pages as above. For four weeks after a launch these count as waiting for Google instead. |
The All findings / Our crawl / Google tabs filter the stream by source. Expanding a Google finding shows each page as a “We saw / Google saw” diff; clicking a page opens a side-by-side comparison of every field — response, canonical, index state, each rich-result type, last visit, and where Google found the URL (sitemaps and linking pages). Paths are shown in full with a copy button for the full URL.
Google’s findings don’t change the SEO score and don’t enter the issue lifecycle (no “New” badges, no alerts): the score measures what the crawl saw and stays comparable across runs, while Google’s view moves with every inspection. They are derived when the report is read, from the most recent inspection of each page, so they update as the inspection sweep cycles — no re-crawl needed.
For CMS integrations, the same data is in the /seo-report response under
summary.google: funnel counts, findings (rule id, severity, title,
help, fix, pages, examples) and pageDetails keyed by page URL. It is null
when Search Console isn’t connected.
The URL Inspection results also feed back into the crawl’s own findings: when Google’s chosen canonical is known for a page with a canonical mismatch, the finding says whether Google agreed with the declared canonical or ignored it.
Revocation and errors
Section titled “Revocation and errors”If the Google account’s access is revoked (or the password owner removes the grant), the connection flips to revoked and the SEO app’s settings show a Reconnect button. Sync errors likewise surface there without interrupting the rest of the app.
Note for self-hosters: the connector requires a Google Cloud OAuth client (
GSC_CLIENT_ID/GSC_CLIENT_SECRETsecrets, plus aGSC_TOKEN_KEYfor token encryption) with<worker-origin>/v1/gsc/callbackas an authorized redirect URI. Thewebmasters.readonlyscope is classed as sensitive, so an unverified consent screen runs in Testing mode with allow-listed users until Google verifies the app.