Everyone generally knows the perfect squares up to at least 12, and then for bigger values, you can use even powers of 2, which I assume people also know. The useful trick is remembering the square root product/dividend property.
Some examples, for a small number, 33, that's less than 36 but more than 25 and it's a lot closer to 36, so I'd guess 5.8 or 5.7 (actual value 5.74..). Halfway between the numbers wouldn't necessarily mean halfway between the known factors. For a bigger number like 1076, you can use 1024, and since it grows more slowly as the factors get bigger, I'd assume something like 32.5 (actual value 32.74..). For a number like 34128, 34128 is ~ 32000, 2 * 16000, 2161000 (almost 1024), so 1.5ish432 so 192ish [1.5*128=128+64] (actual value 184.74).
Heron's method is much better in terms of error though. On the other hand, the method I use is better for larger numbers. I thought about trying Heron's after breaking down the number into smaller factors, but there's something about the division operator with decimals that feels exhausting to even think about.