Ghost Text: The Quietest Interface on Your Mac
There is an interface on your Mac that you have probably used without ever naming it. It is the faint gray text that appears at the end of what you are typing, offering the next few words, and then disappears the instant you keep typing as if it was never there.
That is ghost text, and its defining quality is that it is easy to miss. Not because it is subtle in the way a well-designed button is subtle. Because it is built to be ignored. The whole design assumes you will most often want nothing to do with it, and it has arranged itself so that ignoring costs you nothing. No dialog to dismiss. No badge to clear. You just keep typing and it is gone.
A decade inside code editors
Ghost text grew up in code editors. Developers met it years ago as inline completion: type the start of a name, and the editor shows the rest in gray, one keypress away. It spread through every serious editor because it fit the job. Programmers type long identifiers and repetitive structures, and the completion is usually obvious from the first few characters.
What is interesting is how long it stayed there. The technology was never the blocker. Reading the text before a cursor and predicting what comes next is a general problem, not a programming one. But the economics worked against leaving the editor. A single key has no subscription attached, and it is very hard to bolt a usage meter onto muscle memory. So the habit stayed quarantined in the one place where developers had built it themselves.
- Appears at the cursor, where you are typing
- One key accepts, any key ignores
- Vanishes with no cleanup
- Short by design: a few words, not a paragraph
- Can write whole paragraphs
- Requires leaving the field, pasting, prompting
- Result arrives in a different place than your draft
- Each use is a context switch
Why it feels like nothing
The reason ghost text is hard to notice is that a good suggestion and a good dismissal feel the same. Both let you keep typing. When the gray words are right, you press Tab and the thought continues as if your fingers had done the work. When they are wrong, you type over them and the moment passes. Either way, the sentence you were writing stays the sentence you were writing.
That is the property that makes it feel like nothing happened. The interaction is the point, more than the model behind it, because the interface disappears into the act of writing. You are not operating a tool. You are typing, and occasionally the words are already there.
Compare that with the assistant model, where the tool is always visible: open a sidebar, paste a draft, write a prompt, wait, copy the result back. The tool announces itself constantly. Ghost text earns its name by never announcing itself at all.
When it earns its keep
Ghost text matters most for the sentences you type on repeat without noticing they are repeats. The follow-up email. The scheduling note. The Slack reply that begins the same three words every time. These are not boilerplate you could put in a snippet expander, because they vary slightly each time. They are near-repeats, and a model that has seen your writing can finish them in a way that sounds like you.
That is the quiet promise: not that it writes for you, but that the words you were going to type anyway arrive a little faster, in your voice, without you leaving the app you are in. WriteAmp does this on-device, so the sentences you type while it learns are not the product. The network ledger shows the three times it touches the internet, none of which carry your writing.
The honest edge cases
Two things are worth knowing so the first week does not disappoint you.
The first is that ghost text is a poor fit for long-form generation. If you want a tool to draft a page from a prompt, this is not the interface, and WriteAmp is deliberately not that tool. It predicts the next few words, because a short, ignorable suggestion at the cursor is a different job than a paragraph in a sidebar.
The second is coverage. Ghost text needs to know where your cursor is, and apps that render text on a canvas hide that information. Google Docs, Notion, and Figma text all need extra help in the form of a screen-context permission. It works, and it is on by default with per-app controls, but it is the difference between the seamless native feeling and a tool doing its best with what the system lets it see. The privacy page explains the permission honestly.
The test
Ghost text is not something you evaluate by reading about it. Type a few sentences in whatever app you live in and watch whether the gray words are right more often than they are annoying. The first day you will notice the suggestions. The third day you will stop noticing them and simply press Tab when they fit, which is the entire product: a reflex, not a workflow.
The homepage has a 30-day trial with nothing gated off. Try it in Mail, in Slack, in your notes app, in the browser. The version of ghost text that lives everywhere is the version that finally feels like nothing at all, and that is exactly the point.
Sources
- Apple Accessibility documentation: how apps expose the focused text field and caret that ghost text relies on
On macOS 26+ Macs, Apple Intelligence mode stays free even after the trial ends — you always keep a working path to suggestions.
Written by Amit Ashwini, who builds WriteAmp and runs its marketing. More: why the Tab key beats the chat box · mini, midi, and max compared · benchmark methodology.