Claude has memory now. It's been quietly building a profile of who you are for months. Most of that profile is half-wrong. The exact audit prompt that fixes it in 10 minutes.
Anthropic shipped persistent memory across all Claude users in March 2026. It means Claude no longer starts every chat from scratch. It remembers who you are, what you do, what tools you use, what your goals are.
That sounds great. The catch is that Claude can't tell the difference between a fact you stated about yourself and a one-off question you asked at midnight out of curiosity. It treats both like real information. So a few months in, Claude is working off a profile of you that is half right and half stale.
This guide gives you the exact 3-step audit prompt I run on my own Claude every couple of months. It surfaces what Claude thinks it knows, lets you confirm or correct each item, and updates the memory in real time.
When Claude generates something for you that feels almost-right-but-not-quite, the cause is usually one of two things: a bad prompt, or stale memory.
Bad prompts are obvious to debug. You can see the prompt. You can rewrite it. Stale memory is invisible. You don't see what Claude is pulling from its memory of you, so when an output uses a slightly outdated assumption about your role or your priorities, you can't easily figure out where it came from.
The audit fixes that. It makes the memory visible. You get to read what Claude thinks it knows about you, line by line, and decide what stays and what goes.
Paste this into Claude. It will run the full audit in three rounds. About 10 minutes if you do it properly.
Copy this prompt
You are about to run a memory audit on yourself, with me. The goal is to clean up everything you have stored about who I am, what I do, and what I care about, so your future outputs reflect the actual current me rather than an outdated or partial version. Run this in three rounds: ROUND 1, AUDIT WHAT YOU REMEMBER Write out everything you currently believe about me based on your memory and our chat history. Organize it under these headings, one line per item: - ROLE & WORK (job title, company, what I actually do day-to-day) - TOOLS & STACK (the specific tools I use, with versions or plans where relevant) - PRIORITIES & GOALS (what I'm trying to accomplish in the next 90 days) - WORKFLOWS & HABITS (how I work, when, with whom, in what order) - PERSONAL CONTEXT (family, location, communication preferences, anything that affects how you talk to me) - EXPLICIT INSTRUCTIONS (anything I've told you to always or never do) For each item, mark the confidence level: HIGH (stated explicitly), MEDIUM (inferred from multiple chats), LOW (inferred from one-off or might be stale). ROUND 2, INTERVIEW ME ON THE LOW-CONFIDENCE ITEMS Go back through your list and interview me, one item at a time, on every LOW and MEDIUM confidence item. Ask short specific questions. Multiple choice where possible. Wait for my answer before moving to the next. Examples of how to phrase: - "I have you down as a [thing]. Is that still accurate? (Yes / No / Partially)" - "I have us using [tool] for [purpose]. Are you still using it, or did you switch?" - "I have you focused on [goal]. Is that still a top-3 priority, or has something replaced it?" After each answer, summarize what changed about your understanding in one line. ROUND 3, UPDATE AND REPORT Once we're done with the interview, update your memory with the corrected picture. Then give me a one-page report with: - THE UPDATED PROFILE: a clean current version of what you know about me - WHAT CHANGED: the specific items that were stale or wrong - THREE RECOMMENDATIONS: based on the updated profile, the three highest-leverage workflow improvements you'd suggest I make in the next 30 days End by asking me if I want to save this as a recurring quarterly skill so we can rerun it on a schedule.
Run it quarterly
I run this audit once a quarter and the outputs from Claude noticeably get sharper in the week after. The cumulative drift between audits is real. Calendar it. The first Saturday of every quarter, 10 minutes, done.
If you use Claude Cowork or Claude Projects, save the audit as a reusable Skill called Quarterly Memory Audit. Then you can invoke it with one command instead of pasting the full prompt every time.
Setup: in Claude Cowork, go to Skills, create a new skill, paste the prompt above as the system instructions, and give it a one-line description. Trigger it any time by typing /quarterly-memory-audit or saying 'run my memory audit.'
Bonus move: set a recurring scheduled task in Cowork to remind you to run it. First Saturday of every quarter is what works for me.
Each time you run the audit, save the final one-page report to a file called claude-memory-YYYY-QQ.md in your Claude Project, your Drive, or wherever you keep working files.
Two months from now, when an output feels off, you can pull up the most recent report and verify whether the issue is fresh drift or something the audit already caught. That history compounds.
After three or four audits, you also start to notice patterns about how you change over time. Which workflows actually stick. Which priorities keep showing up. Which tools you abandon. That's a small but real piece of self-knowledge most people never write down.
The wider rule
Any AI system that has memory of you needs this same audit. The version above is written for Claude. The same idea applies to ChatGPT's memory feature, Perplexity's profile, or any agent that builds a persistent profile. Run it on each, on the same schedule. The goal is that no AI is using a 6-month-old version of you to make today's decisions.
Go All In
If this guide helped, but you’re looking to go deeper, I got you!! My 30-Day Challenge takes you from saving AI tips you never use to actually building with AI, step-by-step.
I show you exactly how I automated two e-commerce brands, my social media, and most of my personal life, then hand you the agents, workflows & systems to do the same. I’m teaching you every single thing I know with one lesson and one build a day.
Join the Challenge →Doors close June 21. We start June 22.
© 2026 Mariah Brunner. All rights reserved.