Make Claude start every reply with your name: a one-line canary alarm for long chats

Ever notice that after a long chat, Claude seems to forget what you told it at the start? Here is my trick for catching it: add one line to your settings, "start every reply with my name". The day it stops, I know the earlier rules are slipping too, and it is time for a new chat. This guide gives you: 1) the setting line (one way for Settings in claude.ai, one for CLAUDE.md in Claude Code), 2) the handoff-summary prompt to paste when the name goes missing, and 3) why this happens and what not to over-trust. Every image below comes from our Reel, starring our orange block mascot.

01What it is: a canary alarm that starts every reply with your name

Miners used to carry a canary into the mine: when the bird went quiet, they knew the air was bad. Same idea here. I ask Claude to start every reply with my name. If it does, it is still paying attention to my rules. The day it stops, the alarm is ringing.

Orange block mascot beside a desk; the monitor shows three Claude replies, each starting with "Vivian," highlighted — Reel opening: three different questions, and all three replies start with "Vivian,".

These three answers are real Claude output: on 19 Oct 2026 we asked Claude Haiku one question each, with a single instruction line, "start every reply with 'Vivian,'". To be clear, the chat page below is our own mock layout, not a claude.ai screenshot, but the text is the unedited answer.

Our own demo chat page with three questions and three replies, each reply starting with a highlighted "Vivian," — Full text of the three answers (19 Oct 2026, Claude Haiku, our own demo layout).

You will get: one setting line, one handoff-summary prompt for switching chats, and which situations are nothing to worry about. Read on; the setup takes 2 minutes.

02Step 1: write one line in claude.ai Settings

"Instructions for Claude" is an account-wide setting: whatever you write there applies to all of your future conversations.

Click your initials in the lower left corner.
Select "Settings".
Find "Instructions for Claude", paste the line below, and save.

Setting line (paste into Instructions for Claude, replace the [ ] with your name)

Start every reply with "[your name],".

Official Claude help page screenshot with the "Instructions for Claude" heading highlighted and setup steps — Official help page: Instructions for Claude in Settings (checked 20 Oct 2026).

Help page with a sticky note "start every reply with Vivian" and checklist item 1 ticked — In the Reel: write one line into Instructions and tick checklist item 1.

The official help page says it clearly: instructions you add here apply to all of your conversations with Claude. Menu labels may change between versions; if you cannot find it, look for an "Instructions" field in Settings.

03Step 2: using Claude Code? Put it in CLAUDE.md

Claude Code does not read your claude.ai Settings; it reads the CLAUDE.md file in your project. Put the same line in there.

Open CLAUDE.md in the project root; create it if it does not exist.
Paste the line below and save.
Start a new Claude Code session and ask a question to see whether it uses your name.

One line for CLAUDE.md (replace the [ ] with your name)

# Rules
Start every reply with "[your name],".

Official docs page on a dark wall with a CLAUDE.md paper card that reads the rule "start every reply with Vivian" — The official docs say a project-root CLAUDE.md is re-read after /compact.

Two points from the official docs (checked 20 Oct 2026): CLAUDE.md and auto memory are loaded at the start of every conversation, but Claude treats them as context, not enforced configuration; and a project-root CLAUDE.md is re-read from disk after /compact.

04Why the name disappears: not lying, the start gets crowded out

People often say "Claude starts lying in long chats". My take: it is not on purpose. As the chat grows, the rules you set at the start become easier to overlook.

Navy wall: a chat-window box fills with messages and the gold rule card at the top is pushed out — Reel metaphor: new messages keep coming and the earliest rule is pushed out. An illustration only.

Anthropic's own engineering article (29 Sep 2025) calls this context rot: as the number of tokens in the context window increases, the model's ability to accurately recall information from that context decreases.

Screenshot of the Anthropic engineering article with the "context rot" sentence highlighted — Anthropic engineering article; the highlighted sentence defines context rot (checked 20 Oct 2026).

05The name is gone: new chat, with a handoff summary first

When the name stops appearing, do not just re-explain everything. This is better than simply starting a new chat: have the old chat write a summary first, then carry it over.

Our demo chat page: the first reply starts with Vivian, the second has a blank gap where the name should be — Demo: without the setting line there is no name. A separate run of ours, not a drift caught in a long chat.

In the old chat, paste the handoff prompt below and let it write the summary.
Copy the summary it gives you.
Open a new chat, paste the summary, and carry on with your next question.
Check that the first reply in the new chat uses your name.

Handoff-summary prompt (paste in the old chat)

Please turn this conversation's goal, decisions made, my rules and the next step into a handoff summary.

Opening of the new chat (replace the [ ] with the summary you copied)

Here is a handoff summary from a previous chat. Please continue from it:
[paste the summary]

Yellow wall: old chat greyed out, a "handoff summary" card pasted into the new chat, checklist all ticked — The Reel's last step: old chat greyed out, summary pasted into the new chat, checklist complete.

06Things to watch out for

A missing name is only a warning sign. It does not mean Claude is lying or prove it is wrong, and a name that keeps showing up does not mean nothing else was forgotten.
This is not an official feature; it is a community trick. The docs do not say what the name detects, so we treat it only as a reminder.
The three answers above are real Claude Haiku output. The "name missing" one is a separate demo we ran, not a drift caught in a long chat.
Never put passwords, ID numbers or bank details into Instructions, CLAUDE.md or a handoff summary.
Menus and help pages change; the setting locations above were checked against the official pages on 20 Oct 2026. You can also tell Claude to "remember this" so it saves it to memory, but memory works differently from Instructions.