It lets Claude look through your earlier conversations when you ask it to.

Normally, Claude can only see the conversation you are in right now. Anything you said in a different chat last week is out of view. When the chat search feature is turned on, Claude gains a way to look back. You can say something like “find what we discussed about my budget plan,” and Claude can search your earlier conversations, open the relevant ones, and use what it finds to continue your work.

This turns a pile of separate chats into a searchable history. You do not have to scroll through titles or copy and paste old answers into a new conversation. Claude does the looking, and then it can summarize what you decided, pull out a specific detail, or carry on from where you stopped.

Most valuable work with Claude does not fit in a single sitting.

Real projects stretch over days and weeks. You might draft an outline on Monday, refine the argument on Wednesday, and polish the language on Friday, each time in a different chat. Without a way to reach back, each new conversation starts from nothing, and you have to explain the background again. Chat search removes that friction.

This feature is controlled by a setting, and availability can depend on your plan.

Chat search is controlled by a toggle in your settings, usually labeled something like “Search and reference chats.” You can switch it on or off whenever you like. Anthropic’s help center states that you remain in control of whether Claude can search past chats and use memory, and that you can disable these features and enable them again when needed.

Availability can differ by plan, and it has changed over time. At the time of writing, the help center describes searching past chats as a feature for paid plans, including Pro, Max, Team, and Enterprise, on the web, desktop, and mobile apps. The exact menu location and wording of the toggle have also moved between versions of the app, appearing under areas such as Capabilities, Features, or Memory. If you cannot find it, check the current help center for your version.

Claude runs a search, looks at the matches, and reads the parts that matter.

When you ask Claude to find something from the past, it does not scan every word of your entire history at once. It performs a search using keywords, much like a search box. The search returns snippets from conversations that seem to match. If a snippet looks promising, Claude can open that conversation at the right spot and read a few messages around it. Then it uses what it learned to answer you.

In the app, this often appears as a visible step, such as a search action with results shown, so you can see that Claude is looking things up. It may also show which earlier chats it used, so that you can open them yourself. This transparency is useful because you can check the source instead of just trusting a summary.

flowchart TD
    A["You ask Claude to find
something from the past"] --> B["Claude searches your chats
using keywords"]
    B --> C["Snippets from matching
conversations appear"]
    C --> D{"Is a match promising?"}
    D -- "Yes" --> E["Claude opens that chat
and reads around the match"]
    D -- "No" --> F["Claude tries different search words
or asks you for more detail"]
    E --> G["Claude answers using what it found"]
    F --> B
        

They are separate features that both help with continuity, and they behave differently.

People often confuse chat search with memory. Memory is a short, lasting summary of useful things about you, such as your role, your projects, and your preferences, which Claude can apply in the background. Chat search is an action Claude performs when you ask it to look something up, and it retrieves details from specific earlier conversations. Each has its own toggle, and you can use either one without the other.

flowchart LR
    M["Memory
Short lasting summary"] --> C["Claude in a new chat"]
    S["Chat search
Lookup on request"] --> C
    C --> A["Answer that fits
your context and history"]
        

Boundaries decide what Claude can find, so it helps to know them.

Chat search works within certain boundaries. According to Anthropic’s help center, if you are working inside a project, searches are limited to conversations within that project. If you are outside any project, searches cover your chats that are not in projects. This keeps work in separate projects from mixing together unless you choose otherwise.

Some conversations are intentionally out of reach. Incognito chats are not saved to your history and are not pulled up when Claude searches. A chat you have deleted is gone from your history, which is also the reliable way, according to the help center, to make sure a particular conversation is never found. Claude also cannot find things you never discussed with it, or details that were only said in your head.

Natural phrases about the past are the cue that tells Claude to look back.

You do not need special commands. When you write as if Claude already shares your history, such as referring to “my project,” “the plan we made,” or “what you suggested,” Claude can recognize that you are pointing to something outside the current chat. With chat search on, it can then look that up before answering. If the reference is too vague to search, a good assistant will ask which thing you mean instead of guessing.

The search matches words that actually appeared in the conversation.

Because the search looks for words, the best terms are content words: the topic, the name of the project, the people, the product, or the unusual phrase that came up. Words that merely describe the act of talking, such as “discussed,” “conversation,” “yesterday,” or “last time,” rarely appear in the old chat and do not help the match. A handful of distinctive words usually works better than a long sentence.

Sometimes you remember when, not what.

You may only remember that you worked on something roughly last week, or you might want an overview of your recent activity. In that case, a time-based request is more suitable than a topic search. Claude can look at your recent conversations in a time window and list or summarize them. You might ask for a recap of what you worked on over the past few days, or for the chats from a certain period.

Time-based lookups are good for summaries, for finding the chat you had right before a break, or for building a weekly review. When you know both the time and a topic, combining them is the strongest approach. A specific time window narrows the field, and topic words pinpoint the chat inside it.

flowchart TD
    Q["What do you remember?"] --> T{"A topic or keyword?"}
    T -- "Yes" --> K["Search by topic words"]
    T -- "No, only roughly when" --> R["Ask for recent chats
in a time window"]
    K --> C["Combine with the time
if you remember it"]
    R --> C
    C --> A["Open the best match
and continue your work"]
        

What Claude finds is evidence. You still decide what it means.

After a search, Claude will typically tell you what it found and may name or link the conversations it used. Treat those as pointers you can check. If a detail matters, such as a figure, a deadline, or a decision, open the original chat and confirm it. Summaries are convenient, but they can lose nuance, and a quick look at the source takes seconds.

Also notice how much Claude found. If it reports finding several relevant chats, it may be summarizing across them. If it found only one small match, its answer may rest on thin evidence. A good response will say plainly what it found and what it did not, rather than filling gaps with guesses.

An old chat contains both your words and Claude’s, and they are not equal.

This is one of the most important habits when using old chats. In a past conversation, Claude may have suggested several options, drafted ideas, or explored hypotheticals. Those are suggestions, not decisions. A decision is something you clearly chose or committed to. If a summary says “you decided to do X,” check that you actually said so, rather than merely reacting positively to Claude’s proposal.

Brainstorms deserve the same care. Ideas explored in a what-if discussion are hypothetical, and they should not be recalled as facts or plans. When you ask Claude to recall past work, you can request that it separate what you stated, what Claude suggested, and what was only a hypothetical.

A miss is normal and usually easy to fix.

Sometimes Claude reports that it cannot find what you described. Before you decide the chat does not exist, try a few simple fixes. Use different words. People often describe a topic differently when remembering it than when they first wrote it. Try a person’s name, a place, a product, or a technical term that would have appeared in the chat. Also check that the feature is switched on and that you are searching from the right place.

A searchable history is convenient, so it deserves some thought about what you keep.

Chat search means your past conversations can be brought back into new ones. That is the point, but it also means you should think about what you are comfortable having reachable. You have several controls. You can turn the feature off in settings. You can delete conversations you no longer want, which is the reliable way to keep a specific chat from being found. You can use incognito chats for conversations that should leave no trace in your history. And you can keep sensitive work inside separate projects.

On work accounts, remember that your organization may have its own data retention and access rules, and these can apply to chat search and incognito chats. If you are unsure, ask your administrator and read Anthropic’s privacy information for your plan.

A little tidiness now saves a lot of hunting later.

Familiar situations show how chat search saves time.

Chat search is helpful, but it works best with realistic expectations.

A sensible approach is to switch it on if you do ongoing work, use specific words, check sources for important details, and use incognito or deletion for what should not be kept.

These habits cause most disappointing searches.

Follow these steps when you need something from the past.

A few popular beliefs about searching past chats are only partly true.

Quick answers to the questions beginners ask most.

It lets Claude look through your earlier conversations when you ask, find relevant ones, and use them to answer or continue your work.

Use the search and reference chats toggle in your settings. Its exact location varies by app version, so check the help center if you cannot find it.

Help pages have described it as a paid-plan feature. Availability can change, so confirm in the current help center for your plan.

Memory is a lasting summary of useful context applied in the background. Chat search is a lookup of specific details from past conversations that you request.

Use topic words, such as a project name, a person, a product, or a distinctive detail, rather than words about talking or time.

Ask for recent chats in a time window, then narrow with a topic. Combining time and topic works best.

The help center has described deleting the chat as the sure way. Incognito chats are not searched, so use them for private topics in the first place.

Try different words, check that the toggle is on, and remember searches are limited by project. The chat may also have been incognito or deleted.

Use it as a guide, but open the source for important details, and ask Claude to separate your decisions from its suggestions.

Yes. When you search from inside a project, the search stays within that project’s conversations, which keeps work separate.

No. Only the setting changes it. Claude can point you to the toggle, but you need to switch it yourself.

Searching past chats turns a pile of separate conversations into a history Claude can look through on request. It is controlled by a setting you manage, it may depend on your plan, and it works within boundaries such as projects and excludes incognito and deleted chats. It is not the same as memory, which keeps a lasting summary, while search retrieves specific details. For the best results, ask with topic words and distinctive details, read what Claude found, open the source for anything important, and keep your decisions separate from Claude’s suggestions. Keep your history tidy with clear titles and focused chats, and use deletion, incognito chats, and projects to protect what should stay private.