The general problem with the regex is that . matches any character (because you've used /s), and that .+ is by default greedy, that is, it will match as many times as possible. Have a look at perlretut (pay attention to mentions of "greed").

One way to solve this is to change the dot . to an expression that says "match anything except this". For single characters, one could use a negated character class, as in [^a-z]* which will match any characters except lowercase ASCII letters. For longer strings, one way to do it is with a negative lookahead: (?:(?!not|these|strings).)* will give you the "hard stop" you want.

You've only provided one example string, but when developing regular expressions, the more test cases the better. I showed one way to write tests for regexes here: Re: How to ask better questions using Test::More and sample data. For now I've guessed a few test cases, but you should fill that out. This also works on your long string:

use warnings;
use strict;
use Test::More;

my $regex = qr{
        \b INSERT \b
        (?:  # the next thing may not be one of these:
            (?! \b SELECT \b | \( | \; ) .
        )+   # match one or more times
        \b doc \b  # target string
    }imsx;

  like "INSERT doc (",          $regex;
  like "INSERT INTO doc bar (", $regex;
unlike "INSERT foo ( doc",      $regex;
unlike "INSERT foo SELECT doc", $regex;
unlike "INSERT foo ; doc",      $regex;

done_testing;
[download]

Update before posting: Eily already posted something very similar here, I'll post this anyway. Update: Adding \b is a good point. Update 2: Minor edits for clarity.