Synthetic control
A modeled stand-in for what would have happened anyway, so your lift is measured against a fair baseline.
To know whether your change worked, you need to know what would have happened if you had done nothing. You cannot rerun the month, so we build a stand-in: a weighted blend of similar, consented peer sites whose traffic tracked yours before the change. That blend is the synthetic control.
Because the control lives through the same weeks you do, it absorbs the things you did not cause - seasonality, a Google algorithm update, a category-wide slump. Whatever is left after subtracting the control is the part your change plausibly explains, reported as an estimate with a confidence interval.
This is the honest ceiling of what first-party data can tell you. We say "estimated causal lift with stated assumptions", never "proves causation". The assumptions are stated on every card, and when the peer pool is too small the estimate is withheld.
Worked example
You update 40 pages. Your clicks rise 18%. But the matched peer control also rose 11% that month (an algorithm update lifted the whole category). The synthetic-control estimate of your lift is the difference, about 7%, shown with a band - not the raw 18% that would over-credit your work.
Honest limits
Unreliable when the pool of similar peers is small: below 30 eligible donor sites we do not make a causal claim at all and fall back to a labeled before/after. It also assumes no other big change hit you alone at the same moment; overlapping changes are flagged, not silently absorbed.