Replace words with emoji, or keep both, the difference between writing a rebus and decorating a caption.
The two modes produce genuinely different things. Replacing a word with its emoji creates a rebus, where the reader has to decode each symbol, fun as a puzzle, and unreadable at length. Adding the emoji after the word keeps the sentence intact and decorates it, which is what most social copy actually wants: the text still reads normally and the symbols add tone. Around 55 words are mapped, covering feelings, food, weather, animals, travel and common objects. Words outside that set pass through untouched, which is why a paragraph of abstract prose comes back nearly unchanged.
Screen readers announce the full name of every emoji, so a sentence with an emoji after each word is read as an exhausting stream of descriptions. Emoji used for decoration mid-sentence is worse still, because it interrupts the sentence being read aloud. If your text needs to be accessible, put emoji at the end of a message rather than throughout, and never rely on one to carry meaning that is not also in the words. The replace mode in particular produces text that is effectively unreadable to anyone using assistive technology.
Only around 55 common words are mapped. Anything outside that set passes through unchanged, so abstract prose comes back nearly as written.
Append. It keeps the sentence readable and adds tone. Replace mode produces a rebus that is fun briefly and hard to read at any length.
Not very. Screen readers announce each emoji by name, so text with one after every word becomes an exhausting stream. Keep them at the end rather than throughout.