← Claude Code Hub
Oct 29, 2026 · 5 min read ✦ Tip #244

Claude Code prompt suggestions cost usage: how to turn them off

Prompt suggestions in Claude Code, the gray text after each reply, come from an extra request to your model that counts toward your plan. How they work, when they skip, and how to turn them off in the terminal, Desktop and the cloud.

Claude Code prompt suggestions: after every reply, one extra request to the same model leaves your next prompt as gray text in the empty input; turn them off in /config in the terminal, in Desktop Settings, or with CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION=false

TL;DR If you never use the gray text after each reply, turn it off: in the terminal, /config and set Prompt suggestions to false; in Desktop, Settings > Claude Code > Prompt suggestions. If you do use it, press Tab to accept and Enter to send: pressing Enter before accepting sends nothing.

Claude Code's prompt suggestions aren't local autocomplete. Every time Claude finishes a reply, Claude Code sends a short background request to your session's model to guess your next prompt, then shows it as gray text in the empty input.

Per the docs, that request counts toward your plan's usage limits or your API bill, alongside everything else that eats your usage limit. It reuses the conversation's prompt cache, so it's mostly cache reads and a few output tokens: small per reply, but it's one per reply, whether you use it or not.

Two things look similar and aren't the same. The gray example you see when a session opens comes from the project's git history, not from your conversation. And the Desktop card with a Start button is a different feature that opens a new session, covered in the tip on suggested tasks in Desktop.

What it looks like

Captured on Claude Code v2.1.312, in a scratch folder with a calc.py whose add() subtracts. The prompt asked what was wrong and said tests would come next. When the reply finished, the input looked like this:

● add() in calc.py subtracts instead of adding: it returns a-b where it should return a+b.
  So add(2, 3) gives -1 instead of 5.
  I haven't edited anything. Next I'll write the tests, then make the fix.

✻ Crunched for 3s · done 9:38 AM
────────────────────────────────────────────────────────────────────────
❯ write the tests
────────────────────────────────────────────────────────────────────────
  ⏸ manual mode on · ? for shortcuts

Nobody typed write the tests: the terminal draws it dimmed (the ESC[2m sequence). What each key does, tested in that same session:

Key What happens
Enter before accepting Nothing. No message is sent
Tab or Right arrow Places the suggestion as normal, editable text
Enter after Tab Sends the prompt
Start typing Dismisses it

The Desktop Code tab works the same way, with one more condition: the suggestion only shows while the prompt box is empty and has no attached files.

How to turn them off

1. In the terminal, from /config

Type /config, filter by prompt, and set Prompt suggestions to false. The toggle writes the promptSuggestionEnabled setting.

2. In your settings file

The same thing by hand, in ~/.claude/settings.json or a project's settings:

{
  "promptSuggestionEnabled": false
}

Started with --settings '{"promptSuggestionEnabled":false}' and the same test conversation, the input stayed empty after the reply.

3. With the environment variable, which beats both

export CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION=false

The variable takes precedence over the setting. The reverse test confirms it: with promptSuggestionEnabled: false in settings and CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION=true in the environment, the write the tests suggestion came back. If you turned them off in /config and still see them, look for the variable in your .zshrc or .bashrc.

4. In Claude Code Desktop

Go to Settings > Claude Code (or Settings > Code, whichever your app shows) and turn off Prompt suggestions. Open sessions stop showing them the next time they start or resume, not right away.

5. In cloud sessions

Add CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION=false to the cloud environment's environment variables, in .env format. Before v2.1.290, new cloud sessions ignored it: if it didn't work for you earlier, it does now. Editing those variables is covered in the tip on Desktop cloud environments.

When they skip on their own

Even when they're on, Claude Code doesn't generate a suggestion in these cases, per the docs and the changelog:

  • Cold cache. If the prompt cache has expired, a suggestion would cost the whole conversation uncached, so it isn't requested. What cools the cache is in the tip on what broke your cache.
  • Close to your usage limit. When your account is running low, they pause. With CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION=true they keep going until you hit the limit (since v2.1.238).
  • Replies billed to credits. Since v2.1.302, no suggestion follows a reply billed to usage credits.
  • Plan mode, a reply that ended in an error, and agent team teammates. In a team, only the lead's session shows them.
  • Twenty unused in a row. Since v2.1.283 the terminal shows them less often and says Showing fewer prompt suggestions · use one to bring them back. Using one brings them back.
  • claude -p. Print mode generates none unless you pass --prompt-suggestions with --output-format stream-json --verbose. A short test query produced no prompt_suggestion message: very short conversations are skipped too.

One more case can throw you: in a session that doesn't fetch feature flags (a third-party provider, a Claude apps gateway, or the first session after an upgrade), they're off by default and the toggle doesn't appear in /config.

Official docs: Prompt suggestions

Tested on Claude Code v2.1.312 on October 29, 2026.

Found this useful? See more of wmedia.es in your Google results.

Add wmedia.es as a preferred source on Google
FREE GUIDE

Approve less, but better.

Ten minutes in, you're hitting yes without reading. That's not review, that's prayer. A 10-page guide with the five levers in order of risk.

  • 1. Rules the first match wins
  • 2. Mode a dial, not a switch
  • 3. Auto mode a classifier approves for you
  • 4. Sandbox a fence the OS puts up
  • 5. Bypass only inside a container
Are you a professional Web developer?

It arrives by email · You join the Gravitas newsletter · Unsubscribe anytime

Claude Code permissions cheat sheet: approve less, keep control
FREE GUIDE

Approve less, but better.

Are you a professional Web developer? · Unsubscribe anytime
Workshop for teams

Multiply your team's output without sacrificing quality: a 6 to 8 hour AI First workshop, online, on the Claude platform.

See the workshop

Tired of approving permissions without reading?