Three different averages
"Average temperature" is ambiguous, and the three meanings differ by a lot.
Normal high is the average of the daily maximum temperatures for that month — roughly what the middle of a typical afternoon does.
Normal low is the average of the daily minimums, which is early morning rather than evening.
Mean sits between them. It is the figure most often quoted as the average, and on its own it is the least useful of the three, because it describes a moment that mostly does not occur.
The gap between high and low is the one people overlook and the one that changes what you pack. Denver in October has an 18 °C afternoon and a 3 °C night. Los Angeles in October has 28 °C and 14 °C. Similar spread, very different clothes — and both are unlike London's 15 °C and 8 °C, which barely varies at all.
Where the data comes from
Daily maximum and minimum temperatures for 1991 to 2020, from the ERA5 reanalysis retrieved through the Open-Meteo archive, averaged by calendar month.
That window is the current standard for climate normals: thirty years is long enough to smooth out individual unusual years and short enough to describe the present climate rather than a historical one. Using the standard window makes these figures comparable with published normals from national weather services.
Each place is a single named point, with its elevation shown in the caption. That matters more than it sounds: a point 200 metres higher up the same valley is a measurably different climate.
Why there is no address search
The tool offers a list of named points rather than a search box, and that is a deliberate limit.
Interpolating between observation points to an arbitrary address is possible, but the result carries an air of precision the underlying data does not support — and you would still need to adjust for your own elevation, shelter, paving and cold-air drainage, which can shift a local figure by more than the distance between two towns.
Naming the point is more honest. Pick the nearest comparable one and apply your own knowledge of how your spot differs, which is the same work you would do with a more precise-looking answer. Why the differences are so large is in hyperlocal weather.
What an average cannot tell you
A normal is the middle of thirty years, so half of all years came in above it and half below, and the scatter is wide — in a continental climate, fifteen or twenty degrees between the warmest and coldest instance of the same calendar date is ordinary.
Which sets the honest boundary on this tool. It is good for choosing a month, judging what to pack, and knowing whether an outdoor plan is plausible. It cannot tell you what a particular week will do, and no historical dataset can, because that is a forecasting question.
The practical division: beyond ten days out, use normals. Inside ten days, use the forecast. For the twelve months written up individually, with what each one is actually decided by, start at weather in January; for choosing a month to travel in, best time of year to visit.
A note on coverage
The list covers 131 points: the largest US metros, every US state capital, and a spread of world cities. It is deliberately a cross-section rather than an attempt at completeness.
The state capitals are there for a specific reason. A search like "Maine average temperature by month" wants a state answer, and a single point cannot honestly represent a whole state — so the list names the point instead, and Maine appears as Augusta. That way the figure says what it actually describes.