Does Claude Remember Other Chats? How to Check Its Sources

Yes, Claude can use information from other chats through saved memory and, where available, a separate past-chat search feature. That does not mean a new chat contains every earlier conversation or can quote an old approval accurately. For work that must stay exact, tools like fenbs keep the project record outside Claude so an assistant can read the approved decision rather than reconstruct it.

The useful question is not simply whether Claude remembers. Ask what it is drawing on, whether that source belongs to this project, and whether you can verify it. Our AI memory hub covers the broader distinction between remembered background and a record you deliberately maintain.

What allows information to cross between chats?

Anthropic distinguishes saved memory from past-chat search↗ (opens in a new tab). Memory stores topics; chat search looks up earlier conversations and can show citations. Availability and controls differ, so check your own settings rather than assuming a colleague has the same setup.

Say you ask for another customer update and Claude uses short paragraphs. That could be a remembered preference, an instruction you already set, or simply a reasonable writing choice. The familiar style alone does not prove that it searched last Monday’s conversation. Treat recognizable details as something to investigate, not evidence of access to all your history.

How do project boundaries affect recall?

According to Anthropic’s project guidance↗ (opens in a new tab), a project can use its own memory alongside account memory. Project memory can also inform outside chats; turning off Use account memory on the project keeps it separate. Memory from another project is not part of that project’s memory context.

Do not confuse those memory boundaries with search boundaries. Anthropic describes past-chat search as covering conversations outside projects, or conversations within the particular project you are using. A missing result may therefore reflect where the conversation lives rather than a failure to save it.

For example, keep two clients’ website projects distinct. A preference for concise updates may be useful in both, but an approved tagline belongs to one client. Name the client and source document in your request. Before reusing a remembered statement, check that it is an account preference rather than a conclusion from unrelated work.

How can you check what Claude is using?

1. Identify the conversation you want: its subject, the project, and a distinctive phrase. A request about “the launch” is ambiguous when you have several launches.
2. Ask Claude to find the earlier exchange and separate direct evidence from remembered background. Open any chat citations it provides.
3. Review saved Topics in Settings, then Memory. Compare a relevant topic with the source rather than relying on the assistant’s description of its own behavior.
4. Check account instructions, project instructions, and supplied files when a detail appears without a chat citation. It may already be available there.

Anthropic documents account and project instructions↗ (opens in a new tab) as separate ways to guide responses. This matters because turning a memory control off does not answer every question about why a particular phrase appeared. A source check is more useful than asking the same vague “do you remember?” question repeatedly.

For project decisions, fenbs keeps the agreed wording with the work, so your check can point to a particular recorded choice.

Why might Claude miss something you discussed?

First make sure you are testing the right thing. Asking about an exact paragraph is a lookup task; asking for your usual writing style is a background task. A useful preference does not guarantee access to every sentence that established it. A missing citation also does not prove a conversation has disappeared.

Narrow the request before supplying more history. Give the project, the decision you need, and any version label you know. If the answer remains uncertain, paste the relevant approved text or provide its source. Do not ask Claude to fill the gap with its best guess when that guess will shape a deliverable.

This article concerns the Claude chat experience. What Claude Code is explains the separate coding tool; do not apply chat-app settings to a coding session merely because both use the Claude name. Start with the product and account you are actually using.

Can you have a conversation without earlier memory?

Anthropic’s incognito chat guide↗ (opens in a new tab) says incognito conversations do not use existing memory, contribute new memories, or enter chat history and past-chat search. They still have service retention rules, and profile information such as preferences can remain available. Incognito is therefore a defined chat mode, not a promise that no information about you exists anywhere.

Choose a mode before entering information and read its description. A regular chat with memory off is a different choice: it can remain in history. For the separate question of what happens afterward, see whether Claude remembers deleted chats. Removing a conversation and inspecting remembered information deserve their own check.

Which work facts need something more exact?

Keep approvals, unresolved questions, and next steps explicit. “The client prefers a simpler design” is helpful background but does not say which layout was approved or whether accessibility testing remains. Write the decision, its scope, the person who made it, and what it changes. Keep a proposal clearly labeled until someone chooses it. In fenbs, Decisions and rules preserve the choices people made, with rules shown first in the context an assistant reads.

How to do this with fenbs

1. Create a free account and add the project you want a fresh conversation to understand. One person can start free, with no credit card.
2. Add the current tasks and record the approved choices in Decisions and rules. Name what is still undecided instead of turning a suggestion into a rule.
3. Use Integrations to connect Claude over MCP, a standard way for an assistant to use an outside tool, with browser sign-in.
4. Ask Claude to read the project context, explain its next step, and save what changed before you end the session. Check that its explanation matches the decision you recorded.

fenbs does not read or import your Claude chats. If your only need is a personal preference inside Claude, its own memory controls may be sufficient; fenbs is useful when the work record needs deliberate wording and a clear next step. For orientation before changing settings, see our plain guide to Claude.