You need to decode the string first, then non-ASCII characters are identified correctly as being a word character or not. String literals are automatically decoded by the utf8 pragma:

use strict;
use warnings;
use utf8;

binmode STDOUT, ':encoding(UTF-8)';

$_ = 'été';
my $desc = "test 123 un été à Tanger. élargir elargir ete";

$desc =~ s|\b\Q$_|<a href="url">$_</a>|sg;
print $desc, $/;
__END__
test 123 un <a href="url">été</a> à Tanger. élargir elargir ete
[download]

I've described that in much more detail in this article, and you could also read perluniintro, the Encode documentation, perlunifaq and perlunicode.