There are a couple ways to estimate the magnitude of an earthquake based on other data.

The easiest, especially from a historical perspective, is just to estimate the degree of destruction. It's possible, even with historical buildings, to estimate their strengths and how much damage they would sustain from different amounts of ground shaking. From that you can work backwards to determine the amount of ground shaking in a given area based on historical records of the amount of damage sustained to buildings of various constructions. However, that's just one point along the way. You also need to do this for lots of different nearby areas so that you can figure out where the likely epicenter was, which is sort of a matter of triangulation. For example, the amount of local destruction will be different if you are 50, 100, or 300 km away from a magnitude 8 quake, as it would be for similar distances from a magnitude 7 quake, and sometimes there's some ambiguity involved there, which is why it helps to have more geographically spread out data. This also applies to tsunamis as well.

Additionally, sometimes there are attempts to map historical earthquakes to specific faults or other sources. To some extent there's a degree of modeling different hypothetical earthquakes and what the resulting damage patterns would look like, then picking which of the modeled quakes most reflects the evidence from the historical records.

And, of course, sometimes archaeological and other research data ties into this as well (both in terms of buildings as well as other info such as sediment deposits or dead trees that could indicate changes in local elevation due to the quake).

Also, there is a strong (though non-linear and sometimes complicated) relationship between the duration of ground shaking and the absolute magnitude of an earthquake. So even if you are far away from an earthquake and you don't experience buildings falling down but do experience shaking (enough to, say, cause stuff to fall off tables and bookcases or what-have-you) the duration of that shaking will often be longer for stronger quakes. Sometimes some of that data exists in the historical record and can be used to inform estimates of earthquake strength.

There's another factor which can help narrow down estimates as well. Earthquakes are like very loud sounds, like a thunderclap, that travels through rock. Air has no sheer strength so there are some modes of sound transmission that don't work in air but that do work in rock, this is why earthquakes have different types of "waves", the most relevant are P-waves and S-waves. P-waves are compressional (like sound waves in air), S-waves are transverse (side to side) waves, caused by sheer forces. These waves travel at different speeds and also dissipate in different amounts through different materials. S-waves are much slower but also much stronger. If you have historical reports about the precise timing and details of shaking in a location you can use this to figure out which events described correspond to which waves arriving (P or S, as well as other types of waves I haven't mentioned) which can help estimate the distance as well as depth of the earthquake (making it easier to narrow down the location and thus the intensity of the quake).

Here's a nice little article going through a lot of these details for the 1755 Lisbon quake: https://www.volcanocafe.org/the-lisbon-earthquake/