Methodology
Plain-language explanation of how AIOWeather turns market data into conditions, forecasts, and one recommended decision. We do not hide weaknesses. A machine-readable version is available at /fixtures/methodology.json (schema 1.0.0).
Climate, weather, and forecast
Climate is the slow-moving baseline: share over months. Weather is today’s condition. Forecast is what we expect next, each with a probability and an impact. Three different time horizons, kept distinct.
Search referral share
The share of identifiable search-engine referral visits sent to websites. It is a referral measurement, not total query volume, and not clicks inside a search results page.
AI chatbot referral share
Identifiable visits sent from AI assistants to websites. It does not measure every AI interaction or total platform usage. Many AI conversations never produce a referral, and those are outside this metric by design.
Why unlike platforms are separated
Search engines and AI assistants are measured differently and mean different things. We never combine them into one chart or one composite number. Each lives on its own board with its own definition.
Warming and cooling
Direction of share over one, three, and twelve months. Warming means gaining share; cooling means losing it; stable means movement within noise. Direction is reported separately from level.
How probability is calculated
A calibrated likelihood that a forecast condition occurs within its window. Inputs are scheduled events (highest confidence), observed leading signals, and modeled analogs to prior comparable fronts. Probability is calibrated against the graded record over time.
How impact is calculated
The expected marketer consequence if the condition occurs: Low, Moderate, or High. Impact is independent of probability: a low-probability event can be high-impact, and vice versa.
How conditions are assigned
A fixed vocabulary (Stable, Clouds forming, Watch, Advisory, Warning, Front arriving, Aftershock, Clearing, New normal) mapped deterministically from current signals, so the same inputs always produce the same condition.
How marketer decisions are assigned
Each condition maps to one recommended decision, from No action to Act. Calm conditions map to No action by default. The most common recommendation is No action, and that is intentional.
How forecasts are labeled (confidence)
Every forecast carries a visible confidence label, so a confirmed change is never presented like a guess. We distinguish five levels: confirmed change (already happened, verified), observable trend (measurable and underway), reasonable forecast (supported by a scheduled event or a strong analog to prior fronts), early signal (a weak lead indicator that may not develop), and unsupported speculation (which we do not publish as a forecast at all).
How forecasts are graded
After a forecast window closes, it is graded correct, directionally correct, too early, overstated, missed, or inconclusive. Grades are published on the accuracy page and are never deleted.
How stale data is handled
If a source has not refreshed within its expected cadence, the report is marked stale and shown with its last verified time. Stale data is never presented as current, and forecasts that depend on freshness are withheld.
What this product cannot know
Private ranking systems, unannounced releases, and the results of any individual business. AIOWeather reports the market climate, not your business. To evaluate a business, use AIOInsights.
BETA limitations
Source coverage is expanding, forecast rules may be refined, and some regions have partial coverage. Every forecast remains archived and graded. See about the BETA.
Data-source attribution
All active sources, their cadence, coverage, and limitations are listed on the sources page. Future sources are not listed as active.