Do you know the range of values for the possible variable, or can you estimate it? If so, it reduces the range of the brute-force. That's probably your best approach.
It's worth recognising that with a well-designed hash algorithm you shouldn't, ideally, be able to draw any conclusions from string > hash conversion about the output of string + second_string > hash. To use the examples you've provided, the MD5 output of "fresh apples is tasty!" is 53aff0275a241760e654b3fe0e2184c4
- the output for "fresh oranges is tasty!" is e7c124fef3816ba5a6d6917e20bd5a8b
. Very different results. So, if what you're looking for is a way to split the hash output and focus just on a substring, or something, you're not going to find it. Your best bet is reducing the brute-force's range.