How the content on checkweather.io is made: who writes it, where the numbers come from, where we use AI, and what you can hold us to.
The short version
We never invent a number. If it cannot be measured or sourced, we leave it out and say so.
Every accuracy figure is graded against recorded weather, not against our opinion. The method is on the methodology page.
Nothing on this site is for sale. No advertising, no affiliate links, no sponsored placements, no paid rankings.
We say what we cannot measure, on the page, not in a footnote.
Ten principles we do not depart from
We never invent a number. Where a model or a city has no data, the page says so instead of filling the gap with an estimate.
Every number has a named source. Forecasts and reanalysis come via Open-Meteo (CC BY 4.0); radar from RainViewer. Both are named on the pages that use them.
We grade against what happened, not against a favourite. ERA5 reanalysis is the benchmark, and the same benchmark applies to every model.
We disclose the bias in our own method. ERA5 is ECMWF-built, which gives the European model a small structural edge. That warning sits on every scoreboard, not in the small print.
We do not grade what we cannot verify. AccuWeather, The Weather Channel and BBC Weather run proprietary blends. We grade the open models they build on, and we say that is what we are doing.
A month is a snapshot, not a verdict. Thirty days shows current form. We do not present it as all-time skill.
Nothing is for sale. No weather service, app or brand can buy a position, a score or a mention. There is no advertising and no affiliate link on this site.
One named publisher stands behind it. The site is published by Nortra Online ApS and every page can be traced to a person who can be written to.
We correct errors quickly. Tell us the page and the number; a specific report gets fixed faster than a general one.
We are not a weather service. We do not issue forecasts or warnings. For official warnings, go to your national service.
How we use AI
Every page is reviewed and approved before it goes live. To make the content as clear and well-founded as possible we use AI in three ways:
Clarity. AI helps tighten language and structure so a guide is easier to read. The editorial choices and the final text are ours.
Research and data. AI helps organise material and find patterns in the accuracy data. Every number is checked against its source before it goes on a page.
Tooling. We use AI to build the scripts that fetch and grade the model data, and the checks that stop a figure that does not match the underlying data.
AI produces no accuracy scores, no forecasts and no quotes here. Every figure comes from the measured pipeline described on the methodology page.
How a page is made
Foundation. We establish what can actually be measured or sourced. If a question cannot be answered with data, we do not publish the page.
Measurement. Archived model forecasts are graded against ERA5 over the last 30 days, per city, for temperature, wind and rain.
Writing. The text is written from the measurements, and every number is traced back to its source.
Check against the data. A figure that does not match the pipeline stops publication, even when the rest of the page is ready.
Update. The accuracy data refreshes on a schedule. The date each page was last updated is printed on it.
Independence and how this site is funded
checkweather.io carries no advertising, no affiliate links and no sponsored content. No weather service, app or brand pays us, and none can buy a score or a placement. There is nothing on this site that changes if a provider is unhappy with its result.
The site is published by Nortra Online ApS (CVR DK43186388), Maglebjergvej 6, 2800 Kongens Lyngby, Denmark and is run at the publisher's own cost. It is not affiliated with any national weather service or consumer weather brand.
What we are not
We are not a weather service. We do not issue forecasts or warnings of our own; we compare the models that other services are built on. For official warnings, always use your national service.
We are also not a review site for weather apps. We grade the open models, because those can be verified. An app built on a proprietary blend cannot be, and we do not pretend otherwise.
Have we got it wrong?
Then we would rather know. Use the contact page and tell us the exact page and the number you think is wrong. Every figure is derived from a named source and can be traced, so a specific report gets fixed quickly.