SEO Change Impact Checker: GSC Before/After Test

Did your SEO change work, or is it noise? Give it a Google Search Console CSV and the change date. It returns REAL WIN, REAL LOSS, NOISE or TOO EARLY, with the numbers behind it.

A plain before/after in Search Console fools people all the time. A core update, a holiday, a weekend-heavy window or the last 2-3 days of incomplete GSC data can fake a win or hide one. This Actor runs the test an analyst would:

• Matched windows of whole weeks before and after the change, a recrawl buffer (default 3 days) and day-of-week adjustment
• Control group diff-in-diff: changed pages vs. all other (or chosen) pages, so sitewide swings and seasonality cancel out
• Noise band from your own history: the same test re-run on 100+ placebo change dates where nothing happened (weekly-block bootstrap if history is short)
• GSC data-lag guard: drops the last 3 days and tells you the exact date to re-run when it's TOO EARLY
• Drivers: impressions, average position and CTR before vs after
• Warnings for overlapping changes, sitewide shifts, low volume and partial windows

Use it for title/meta rewrites, content refreshes, internal-linking pushes, schema roll-outs, page-speed work, migrations or "is this traffic drop real?".

Input

• csvUrl (public CSV or Google Sheets link), csvText (pasted), or gscRows (Search Analytics API rows with dimensions [date, page])
  • Date + page CSV (Date, Page, Clicks, Impressions[, Position]) enables the control group. Get it from the GSC API, Looker Studio or BigQuery bulk export.
  • The Performance > Dates export works for a sitewide or UI-filtered before/after (placebo band only).
  • The Pages tab has no dates, so it can't be used.
• changeDates: one or more YYYY-MM-DD
• treatmentPatterns: regex for changed pages, e.g. /guides/ (empty = sitewide)
• optional controlPatterns, noControl, windowDays (28), bufferDays (3), minEffect (0.05), asOf

{ "csvUrl": "https://example.com/gsc-date-page.csv", "changeDates": ["2026-08-17"], "treatmentPatterns": ["/guides/"] }

Run with no data = free SYNTHETIC demo on a fake site (example-synthetic.test) with a planted +20% win, so you can see the output first. Demo runs are never billed.

Output (one dataset item per change date)

verdict, netEffect, treatmentLift, controlLift, noiseBandLow/High, noiseMethod, impressionsEffect, clicks/day, CTR and position before/after, windows, summary (plain English), warnings. A Markdown report is saved as REPORT in the key-value store.

Example (real run on the SYNTHETIC demo data): REAL WIN, net effect +20.0%, noise band -6.8% to +8.5% (placebo test on 80 fake change dates), avg position 9.0 -> 7.45. In the same data, a page group with no real change shows -9% in a naive before/after, but with the control group it's NOISE, because the whole site fell 12% that week.

Pricing

Pay per event: $0.01 per change date analysed. TOO EARLY results, failed inputs and the synthetic demo are free. Apify platform usage is included.

FAQ

• Is this CausalImpact? No. It's diff-in-diff with a placebo/bootstrap noise band: simpler, transparent and explained in the report.
• Do you store my data? Only in your own run's dataset and key-value store. No Google login is needed; you provide the export.
• Why NOISE? The effect is inside the normal swing of your own data or below minEffect. Wait 2-4 weeks or test on more pages.
• Small sites? Under ~100 clicks in the pre window the band gets wide and results are mostly NOISE. That's the honest answer.

Limitations

Evidence, not proof: other edits on the same pages or updates that hit page types differently can still explain a result. Needs daily data. Clicks are the primary metric.

Also by Cygnix Labs: Redirect Checker, Internal Link Audit, Content Decay Audit, Keyword Cannibalization Finder. Agent skill version: agensi.io (SEO Change Impact Checker).