A toggle is simply a switch that turns a feature on or off.

A setting is a choice you make about how a product behaves. A toggle is the simplest kind of setting: a switch with two positions, on and off. Claude offers toggles for things like looking up live information on the internet, remembering useful facts about you, and searching your earlier conversations. It also offers text boxes and options where you describe your preferences or pick a writing style.

These choices matter for three big reasons. First, they affect quality, because an assistant with the right context and tools gives better answers. Second, they affect privacy, because some features store or use information about you. Third, they affect trust, because you should always know what the assistant can see and do. A person who understands the settings can enjoy the benefits while staying comfortable.

Before learning each switch, it helps to see the whole neighborhood.

In the Claude apps on the web, desktop, and mobile, you usually reach settings through your profile or account menu. Inside you will find groups of options. Names and arrangement can differ between versions, but the main groups are easy to recognize by their purpose.

flowchart TD
    S["Claude Settings"] --> P["Profile and preferences"]
    S --> C["Capabilities and features"]
    S --> M["Memory controls"]
    S --> N["Connectors"]
    S --> D["Privacy and data"]
    S --> A["Account, plan, and admin"]
    C --> W["Web search"]
    C --> H["Search past chats"]
    C --> G["Generate memory from chats"]
        

Web search lets Claude look things up instead of relying only on what it learned in training.

Every AI model learns from information up to a certain date, called its knowledge cutoff. Anything newer is simply not in its built-in memory. Web search solves this problem. When it is switched on, Claude can search the internet, read what it finds, and use those pages to answer you, ideally showing the sources so that you can check them.

You can usually control web search in two places: a switch you can flip inside a conversation, and a related option in the settings area. Having it on means Claude can decide to look something up when your question needs current facts. Having it off means Claude answers from its built-in knowledge and anything you provide directly.

When to turn it on

When you might leave it off

For writing help, brainstorming, explaining a stable concept, or working only on a document you pasted in, search is not needed. Leaving it off can also keep a conversation focused on your own material, and some organizations restrict it for policy reasons.

People often mix them up, but they work in different ways.

Claude has two related features that both help it continue your work across conversations. They are controlled by separate toggles, and understanding the difference will save you confusion.

Memory is like a short profile that Claude carries with it. Chat search is like a filing cabinet that Claude can open when you say, “Find what we talked about before.” The memory summary is meant to capture durable patterns, not every detail of every chat. The search feature is for retrieving the details of a specific past conversation.

You are not stuck with whatever Claude has saved. The controls are in your hands.

A key design idea is transparency. You can open the memory area and see what Claude has remembered, usually organized into topics. You can read an entry, edit it if it is wrong or out of date, and delete individual entries you do not want kept. You can also ask Claude in a conversation to adjust what it remembers or to ignore a certain topic.

Some information is deliberately left out, no matter what you ask.

Memory is designed to hold everyday working context, such as your role, the projects you are on, and how you like to communicate. It is not meant to become a vault for your most sensitive data. Anthropic’s help pages explain that certain kinds of information are never saved to memory even if you ask, including government identification numbers, financial account numbers, criminal history, and immigration status. When this happens, Claude tells you it cannot save that detail.

There is also a separate choice about sensitive topics more generally. Some versions of the app offer an option to include sensitive topics in memory, which is off unless you choose otherwise. If it is off, Claude avoids saving things in those sensitive areas. If you would rather not keep re-explaining a sensitive topic, you can choose to include it, understanding the trade-off.

Sometimes you want a fresh start for just one conversation.

Turning memory off in your settings is a big step. Often you only want privacy for a single chat, such as a sensitive brainstorm or a confidential idea. Claude provides tools for this narrower need.

Incognito chat

An incognito chat is a temporary conversation. In supported apps you start one with a ghost icon. It does not appear in your chat history, does not add to memory, and is not pulled up when Claude searches your past chats. Your existing memory and history stay untouched. For work accounts, remember that an organization’s data retention and export rules can still apply, so incognito is not the same as invisible to your employer.

Memory off for one chat

Some versions let you turn memory off for a single conversation from a menu near the message box. In that chat, Claude does not use what it remembers about you, and nothing from the chat is saved to memory. Unlike incognito, the chat can remain in your history. This choice is usually set before you send your first message and then locked for that conversation.

Projects keep related work together, and they can keep memory separate too.

A project is a workspace where you group chats and files around one goal, such as a report, a product launch, or a course. Anthropic describes memory in projects as having its own space, so the context inside one project stays focused and does not leak into your other work. When you work in a project, Claude can use that project’s memory, and depending on your settings it may also use the memory from your chats outside of projects.

Each project has its own memory setting on the project’s page. If you want a project’s memory to stay completely separate, there is an option there to stop using your account-level memory. Which setting you choose decides whether the project and your general memory can see each other, so it is worth checking when privacy matters.

flowchart LR
    A["Account memory
from regular chats"] --- B["Project A memory"]
    A --- C["Project B memory"]
    B -. "kept apart from" .- C
    D["Project set to separate"] --> E["Uses only its own memory"]
        

Three tools shape how Claude talks to you, and each has a different reach.

Anthropic describes several ways to personalize your experience. They are easy to confuse, so here is a clean comparison. The main difference is how wide the setting reaches and what it is meant to do.

Writing good preferences

Be direct and specific. Say how you want answers to look, which terms you use, and what you do not want. Concrete behaviors work better than vague wishes. For example, describing that you prefer short answers with the key point first is more useful than saying you want good responses.

Connectors let Claude reach beyond the chat window, so permission matters.

A connector links Claude to another app or data source that you choose, such as your documents, calendar, or team tools. When one is enabled, Claude can read from it, and sometimes act in it, to answer questions about your real, current information instead of guessing. This is powerful because it makes answers specific to your situation.

Because connectors reach into your own data, they deserve careful thought. Turn on only what you need, understand what each connection can read and change, and review these settings from time to time. For actions with side effects, such as sending a message or changing a record, a careful assistant asks for your clear approval first.

Depending on your plan, device, and the current version, more options may appear.

Claude keeps growing, and so does its list of controls. You may come across options for choosing between different models, adjusting how much deep reasoning is used for harder problems, using voice, running longer research tasks, or giving Claude a place to create and keep documents. Not every user sees every option, and availability can depend on the plan and the app.

At work, some switches are controlled by an administrator, not by you.

On Team and Enterprise plans, organization owners can have a say in what members can use. For example, owners can manage network access under their admin capabilities settings, and for Enterprise organizations, owners can switch the memory feature off for the whole organization. When an organization-level switch is off, individual members may not be able to turn that feature on for themselves.

Help pages also note that Team plans do not have organization-level memory controls, so individual members manage their own memory settings, while some Enterprise controls sit with owners. Details can change, so ask your administrator or check the current help center if a feature seems missing at work.

Every switch buys you something and asks for something in return.

The goal is not to switch everything on or everything off. The goal is to switch on what clearly helps you, keep control of what is stored, and revisit your choices as your work changes.

Real examples make it easier to picture a sensible combination.

flowchart TD
    A["What do you need?"] --> B{"Current facts
or news?"}
    B -- "Yes" --> W["Turn on web search"]
    B -- "No" --> C{"Want Claude to
know your context?"}
    W --> C
    C -- "Yes" --> M["Turn on memory
and set preferences"]
    C -- "No" --> I["Keep memory off or
use incognito"]
    M --> R["Review settings regularly"]
    I --> R
        

Good settings habits protect you without making Claude less useful.

Privacy is not one switch. It is a set of habits that combine the toggles with your own judgment. Start by deciding what you are comfortable having remembered. Use incognito or memory-off chats for sensitive topics. Delete conversations you no longer want, remembering that deleting a conversation and deleting a memory entry are separate actions, so check both places if you want something fully gone.

Be aware that if you delete a conversation, a memory entry that was created from it may remain until you remove it from the memory area. The help center notes that individual memories can be deleted at any time, so clean up in both places when needed. Also review how your plan treats your data, including any retention rules and options for how chats may be used, in the privacy information provided by Anthropic.

These errors are easy to make and just as easy to fix.

A short, regular check keeps everything working in your favor.

A few common beliefs about Claude’s settings are only partly true.

Quick answers to the questions beginners ask most.

Start with your profile preferences, the web search option, and the memory controls. These three change the experience the most, and you can adjust each of them later.

Memory is context Claude builds from your chats over time. Project instructions are rules you write yourself for one project, and they apply only to chats inside it.

Yes. The memory area lists what is stored, usually grouped by topic, and lets you edit or delete entries.

It keeps what is already stored but stops Claude from using it or adding new entries until you resume. It is good for a temporary change of focus.

Use an incognito chat where it is available, or turn memory off for a single chat if your version offers that option. On work accounts, remember that organization policies may still apply.

Not necessarily. Memory entries and conversations are managed separately, so after deleting a chat, also check the memory area and remove any entry you do not want to keep.

The help center has described deleting the chat as the reliable way to make sure it is not included, since the search feature as a whole is controlled by a single toggle. Check the current help center in case this has changed.

It may depend on your plan, your app version, or your organization’s rules, and menu names can change. Look in the official help center for the current location.

It is best not to. Even if a detail is not stored in memory, you should keep secrets out of chats as a general habit.

No. Turn on the features that clearly help your work and keep the rest off. Review your choices regularly as your needs and comfort level change.

Anthropic’s official help center, because features and labels are updated there first.

Claude’s settings and feature toggles let you decide how much the assistant knows, how current its information is, and how it speaks to you. Web search brings fresh facts, memory carries useful context forward, chat search retrieves details from earlier work, and preferences, project instructions, and styles shape the tone and format. Incognito chats, per-chat memory choices, project separation, and the controls to view, edit, pause, and reset give you real privacy options. At work, administrators may set the outer limits. The best setup is a deliberate one, reviewed from time to time.