Regex for phone numbers, and why strict ones lose customers

The preset is a reasonable sketch of a North American number: an optional country code, three digits with optional brackets, then three, then four, with a hyphen or a full stop permitted between the groups. It handles +1-800-555-0199 and 555.867.5309 exactly as you would expect.

It also demonstrates the whole problem, because it does not match (555) 123-4567. The separator class is a hyphen or a full stop and contains no space, so once the closing bracket is followed by a space the attempt fails, and there is no shorter run of digits in the string long enough to match instead. That form appears in the preset's own sample text, which is the clearest possible illustration of how easily a phone pattern misses a format its author had in mind.

Widening the class fixes that case and invites the next one. Real input contains non breaking spaces, en spaces from a word processor, slashes, extension suffixes, and country codes written half a dozen ways. Every widening also loosens the check, and a pattern permissive enough for real users stops rejecting much of anything.

The approach that survives contact with users is to normalise first. Strip every non-digit, then decide based on the digit count and any leading country code, which reduces a formatting problem to an arithmetic one. If the number has to be genuinely valid rather than merely plausible, that is a job for a phone number library with per country rules, not for a pattern.