Smart quotes

The straight marks are a hardware compromise. A typewriter had a fixed number of keys, so one vertical tick stood in for the opening quote, the closing quote, the apostrophe, and the prime, and ASCII inherited the compromise wholesale. Real typography has four separate marks for the single and double pairs: U+2018 and U+2019 for single, U+201C and U+201D for double, each drawn to lean into the words it encloses. The apostrophe is not a fifth character, it is the right single quote U+2019, which is why a curly apostrophe and a closing single quote are the same glyph.

Primes are the other half of the confusion. Feet and inches, minutes and seconds, and coordinates take the prime U+2032 and double prime U+2033, which are slanted marks and not quotes at all. A straight tick is nobody’s correct answer there either. It is only correct where the character is code: source, JSON, CSS, shell, HTML attributes, regular expressions, and anything a machine will parse, where a curly quote is a syntax error and an editor that inserts one has broken the document.

Autocorrect is what most writing runs on, and it fails predictably at elisions. Anything that opens with an apostrophe standing in for dropped characters gets the wrong mark: ’twas, rock ’n’ roll, the ’90s, and ’til all want the closing form, and every smart-quote engine reaches for the opening one because it sees a space before it. The other regular failure is the leading apostrophe in dialect and in abbreviated years. Fixing them by hand is the work; there is no rule that saves you from reading.

The marks are also local. German uses low-9 opening quotes and a raised closing pair, French uses guillemets with spaces inside them, and Japanese uses corner brackets, so a quotation is one of the details a translated interface has to re-typeset rather than reuse. CSS knows this: the quotes property and the q element take a locale-appropriate pair and the :lang() selector switches them, which beats hard-coding a glyph into a template. Store the correct characters in your content and normalise to straight marks only at the boundary where a search index or a legacy system demands them.