Claude Settings Worth Changing: The Recommended Setup (2026)

Claude settings worth changing: with Help improve our AI models on, de-identified chats can be kept up to 5 years for training; off, the standard 30

Short answer: the Claude settings worth changing are your Instructions (tell Claude how you like answers), memory (review what it keeps), the “Help improve our AI models” privacy switch (decide whether your chats train Claude), Projects for work you repeat, and the model and effort picker. Most people never open these, and they make the biggest difference. Below is each one, where to find it, and what we recommend.

Two menu layouts. Since September 2026, Anthropic has been rolling out a “new Claude experience” that merges Chat and Cowork and moves some settings. In the new layout, settings live under Settings (Your Claude, Privacy, Capabilities, Appearance, Wellbeing) and a separate Customize menu. If your Claude looks older, the same switches are under Settings: Account, Memory, Capabilities, General, and Time and focus. We give both paths.

1. Instructions: tell Claude how you want answers

Where: Settings, then Your Claude, then Instructions (older layout: Settings, then Account, “Instructions for Claude”). Plans: all, including Free.

Whatever you write here applies to every chat. Keep it short and practical. For example:

I work in [YOUR ROLE] at a [TYPE OF COMPANY].
Give me the answer first, then the detail.
Use plain language and short paragraphs. No jargon unless I use it first.
If you're not sure about a fact, say so instead of guessing.
When I ask for writing, match a friendly, direct tone.

Don’t put passwords, account numbers or anything confidential in your Instructions.

2. Memory: check what Claude remembers

Memory is on by default on Free, Pro, Max and Ultra. On Team and Enterprise, your company’s owner decides.

  • The main switch is “Generate memory from chats” under Settings, then Your Claude (older: Settings, then Memory). Turning it off gives you two choices: Pause memory keeps what it has, Reset memory permanently deletes it.
  • See and edit what it knows: click Manage next to Your memory. Memories are grouped by topic, and you can edit or delete each one.
  • “Include sensitive topics in memory” is off by default. We’d leave it off.
  • “Search and reference chats” lets Claude look back through your past conversations. It’s on paid plans only.
  • One chat without memory: before your first message, open the “+” menu and switch Memory off for that chat.

Incognito chats (the ghost icon on a new chat) aren’t saved to your history or memory and are never used for training. Anthropic still keeps them for 30 days. They’re available on every plan, but not inside a project.

Moving from ChatGPT or Gemini? Claude has a memory import (“Start import”) on the web and desktop apps. Anthropic labels it experimental, so check what it brought in afterwards.

3. Privacy: decide whether your chats train Claude

Where: Settings, then Privacy, then “Help improve our AI models” (same place on web and mobile). This applies to Free, Pro, Max and Ultra accounts.

SettingWhat Anthropic does
OnNew and resumed chats can be used to train future models, and de-identified data is kept for up to 5 years.
OffNot used for training. Chats are kept for 30 days after you delete them, the standard period.

Some data is kept longer either way: chats flagged by Anthropic’s safety systems (up to 2 years) and thumbs up or down feedback you send (up to 5 years). Team, Enterprise and API accounts aren’t used for training by default. On mobile there’s also an “Allow us to use your voice data” switch, off by default.

Our take: if you paste work documents, client details or anything personal into Claude, turn “Help improve our AI models” off. It doesn’t change what Claude can do for you.

4. Projects: set up once, reuse every week

A project keeps instructions and files together, so you don’t re-explain the same context in every chat. Go to claude.ai/projects, click New Project, then add project instructions and upload files to its knowledge.

  • Free plans can create up to 5 projects. Paid plans get more room for files, because Claude searches project knowledge instead of reading it all at once.
  • Each project file can be up to 30MB.
  • Each project has its own memory, plus a “Use account memory” switch if you want to keep work and personal separate.

Good first projects: your weekly report, a job search (resume plus target roles), or a class you’re taking.

5. Model and effort: match the job

The model picker sits next to the send button on the web and desktop, and at the top of the screen on mobile. Free plans get Sonnet and Haiku; Pro and higher add Opus. Fable is on Max and Ultra, and on Pro it uses usage credits.

  • Effort runs from Low to Max. Lower effort answers faster and uses less of your limit. Use it for quick questions and save higher effort for hard problems.
  • Thinking makes Claude reason before answering. On the newest models it’s always on.

Keep running into “limit reached”? Paid plans show your session and weekly usage under Settings, then Usage. Our guide to Claude’s usage limits covers what the numbers mean and how to stretch them.

6. Files, artifacts and code

Where: Settings, then Capabilities, “Code execution and file creation” (new layout: “Create and edit files”). It’s on by default and lets Claude make documents, spreadsheets and slides and run code. Artifacts need it on. There’s also an “Allow network egress” switch underneath, which lets that code reach the internet; turn it off if you don’t need it.

7. Web search, Research and connectors

  • Web search: in the “+” menu on every plan. In the new Claude experience there’s no switch: Claude searches when it needs to.
  • Research (longer, multi-source reports): paid plans. In the “+” menu, or type /deep-research in the new layout.
  • Connectors let Claude read Gmail, Google Calendar, Google Drive and other apps. Manage them under Customize, then Connectors. Connect only the accounts you actually want Claude to see.

8. Comfort settings most people miss

  • Appearance: light or dark mode, and a Dyslexic Friendly chat font (Settings, then Appearance; older: General).
  • Wellbeing: break reminders and quiet hours (Settings, then Wellbeing; older: Time and focus). In beta, set on the web or desktop, and they apply on all your devices.
  • Language: the Language option in your profile menu changes Claude’s interface language.

Where did Styles go?

Anthropic has retired the old Styles menu (Normal, Concise, Explanatory, Formal). Custom styles you made were turned into skills: you’ll find them under Customize, then Skills, switched off until you turn them on. For the same effect without skills, put your tone preferences in Instructions.

SettingOur recommendation
InstructionsWrite 3 to 5 lines about your role and how you like answers
MemoryLeave on, review it monthly; keep “sensitive topics” off
Help improve our AI modelsOff if you paste work or personal details
ProjectsOne per recurring task
EffortLow or Medium for quick questions, higher for hard ones
ConnectorsOnly the apps you need
Sensitive one-off questionsUse an incognito chat

New to Claude? Start with how to use Claude, then compare Claude Free vs Pro before you pay.

How we checked

We read Anthropic’s own help and privacy pages on November 21, 2026, including memory and chat search, the model improvement setting, how long Anthropic stores data, model, effort and thinking, and Claude’s pricing page for which plan gets what. Anthropic is moving settings around as the new Claude experience rolls out, so a menu name on your screen may differ slightly.

Related guides

Frequently asked questions

What are the best Claude settings to change first?

Write a few lines of Instructions about how you want answers, review your memory, decide on the "Help improve our AI models" privacy setting, set up a Project for work you repeat, and use lower effort for quick questions.

How do I stop Claude from training on my chats?

Go to Settings, then Privacy, and turn off "Help improve our AI models." Your chats then aren't used for training and are kept for the standard 30 days after deletion instead of up to 5 years.

How do I turn off Claude's memory?

Open Settings, then Your Claude (older layout: Settings, then Memory) and turn off "Generate memory from chats." Choose Pause memory to keep what it has or Reset memory to delete it permanently. For a single chat, switch Memory off in the "+" menu before your first message.

Where did Claude's styles go?

Anthropic retired the Styles menu. Custom styles became skills under Customize, then Skills, switched off until you turn them on. You can also put tone preferences in your Instructions.

Does Claude save incognito chats?

Incognito chats aren't saved to your history or memory and aren't used for training, but Anthropic keeps them for 30 days. They're available on all plans but not inside projects.

Which Claude models can I use on the free plan?

Free plans get Sonnet and Haiku. Pro and higher plans add Opus, and Fable is available on Max and Ultra, or on Pro using usage credits, according to Claude's pricing page.

Scroll to Top