You asked AI for 100 characters. What does the destination count?

Define whether spaces, bytes or visible character clusters count, then check the final text with the receiving form’s counter.

A character limit needs a counting rule. Unicode distinguishes code points from grapheme clusters, which approximate user-perceived characters. A text can have different valid counts depending on which unit is used.

One visible item can contain several units

We tested short strings locally in Node. The emoji 🙂 uses two JavaScript UTF-16 code units, one Unicode code point and one grapheme cluster. Composed 가 has one code point; its decomposed form has two, while both count as one grapheme cluster in our test.

Specify spaces and the target

The invented text “A B” has three code points including the space, or two when that space is excluded. Ask AI to draft below your limit and state the intended counting rule, but do not accept its declared count as the final check.

Use the counter that decides acceptance

Our recommendation is to paste the finished draft into the destination’s preview or counter before submitting. If the destination uses a byte limit, a visible-character count answers a different question. Check again after editing, since the submitted string is what matters.

Avoid cutting through a visible character

When a script shortens text, use a method that respects grapheme boundaries if the goal is visible characters. Even then, read the result for meaning. We did not test a destination form or claim one counting rule applies to every service.

Published October14, 2026. UTF-16, code-point, grapheme and byte counts of invented strings were measured locally using the installed Node/ICU runtime. No AI answer or receiving application was tested, and runtime Unicode versions can differ.

Sources: Unicode · Text segmentation.