Forecasting & experiments

Two reports in the Advanced group look beyond what already happened. Forecast projects your organic traffic forward from its own history, and Experimentation measures whether a change you made to a set of pages caused a difference. Used together, they turn "we think this will help" into a target before the work and a measured result after it.

The two reports

• SEO Traffic Forecast (menu: Forecast): loads up to 16 months of daily Search Console clicks, or your own CSV of sessions, removes seasonality, fits five trendlines and picks the most reliable one, then projects 3, 6 or 12 months ahead with a 95% band. A growth scenario and a conversion-rate × order-value model turn the projection into revenue.
• Experimentation: compares a treatment group of pages you changed with a control group you left alone, each against its own 28 days before launch, and runs Welch's t-test on the two groups' daily values. You get the uplift, a confidence interval, a p-value and a running total of the extra clicks.

A typical cycle

1. Forecast the section you plan to work on, to know what "no change" looks like for the next quarter.
2. Make the change on a set of pages, keep a comparable set untouched, and record the date with an annotation.
3. After at least two weeks, open Experimentation for the result, then rerun the forecast with the new data.

Limits that apply to both

Both reports read Search Console clicks, which are not sessions and carry Google's privacy and row limits (how Search Console counts data). Neither can see outside events: a core update, a competitor's launch or a seasonal event that only happens some years will show up as noise or as a false effect. A forecast is a baseline to plan against, and an experiment is strong evidence rather than proof; for a statistical decomposition of the same history, see Time-Series Decomposition.