Measurement and Forecasting Methodology

How demand is forecast

• Horizon

  Every tracked keyword gets a weekly forecast 26 weeks, or 6 months, ahead.

• Seasons and events

  With more than two years of data, yearly seasonality is included. Recurring calendar events are added to the forecast.

• Range

  A forecast is shown as a range with lower and upper bounds, not a single number. A wide range shows the uncertainty on screen.

• Threshold

  Keywords with less than 30 weeks of data get no forecast. The unfinished current week is left out so a partial week does not look like a drop.

• Label

  Forecasts always carry a “Forecast” label and a dashed line, never mixed with measured values.

• Trend verdict

  The last 12 completed weeks are compared with the previous 12: at least 20% up is Rising, at least 20% down is Declining, anything between is Stable. A keyword climbing from a very low base is First signal; without enough data it is Insufficient data.

• Season versus real growth

  The same change is compared with the change in the same 12 weeks last year. If last year moved by a similar amount, the movement gets a “seasonal” note; if it moved the other way or much less, it counts as off-season.

How a gap is diagnosed

When demand rises but your traffic falls, Search Console data is read first: if clicks fall and the same drop happened in the same period last year, it is seasonal; if impressions fall too, demand is shrinking; if clicks are not growing and the average top-10 position slipped at least one place, the problem is ranking. The “Why did it drop?” investigation on the keyword page then reads search interest, Search Console, AI answers, site crawl and Perception Radar data together and scores three scenarios: a market-wide drop, a drop driven by brand perception, or unexplained variation. The scores add up to 100, and one priority action is written for the top scenario. An investigation costs 1 credit.

How the job list is ranked

Each job gets an impact score: the traffic it can bring, multiplied by the likelihood that the fix works, divided by the effort it takes (hours). The Action Center list is ranked by this score; the high, medium and low impact labels come from it. Content Portfolio separately gives each page a keep, revamp, cull or monitor verdict; pages doing well are marked “leave as is”, so you also see where not to spend effort.

impact = traffic impact × likelihood of recovery ÷ effort

The Four Ps framework

Brand visibility in AI platforms isn't just about being present. Where, how, and with what impact matters too. We measure visibility across four dimensions:

How measurement confidence is shown

Not every measurement carries the same confidence. Results you can take to a budget decision are marked apart from results that only show direction:

Data Sources · Transparency

The source types we measure are documented:

Methodology update: 16 October 2026.