Not so much by looking at the content but by looking at the hints I as a designer gave. So if I placed a text layer with 40px font size, #000000 black in Roboto on an artboard that belongs to project X for brand Y, my friendly typographic assistant could suggest to change it to 36px, #1a1a1a, Product Sans, -.01em character spacing and 1.1x line-height because that would be the closest styling that's correct according to the typographic system of the project (and brand). Basically linting for designers. Actually not sure if ML is needed for that, but maybe ML could help to guess the intention of the designer more accurately.
In the creative process, helping to find close alternatives could be helpful. E.g. if I choose a font that is not available or too pricey for a project, quickly get some high quality alternate options with means to compare pricewise. Or finding a font with squishy search terms, like "soft", "warm", "friendly, but not too friendly".
Not so much by looking at the content but by looking at the hints I as a designer gave. So if I placed a text layer with 40px font size, #000000 black in Roboto on an artboard that belongs to project X for brand Y, my friendly typographic assistant could suggest to change it to 36px, #1a1a1a, Product Sans, -.01em character spacing and 1.1x line-height because that would be the closest styling that's correct according to the typographic system of the project (and brand). Basically linting for designers. Actually not sure if ML is needed for that, but maybe ML could help to guess the intention of the designer more accurately.
In the creative process, helping to find close alternatives could be helpful. E.g. if I choose a font that is not available or too pricey for a project, quickly get some high quality alternate options with means to compare pricewise. Or finding a font with squishy search terms, like "soft", "warm", "friendly, but not too friendly".