Remember More: Personalize Your AI

Most professionals use ChatGPT the same way every time: open a new chat, explain who they are, describe their company, specify their tone, and then finally ask the actual question. That's four steps of setup before any real work gets done. Custom Instructions and Memory eliminate that friction entirely. Set them once, and ChatGPT already knows your role, your company, your preferences, and your style every time you open a new conversation. This reference guide shows you exactly how to configure both features and use them to work faster starting immediately.

7 Things You Need to Know Before Anything Else

1. Custom Instructions are available on ChatGPT Free, Plus ($20/month), and Team plans, they live in your account settings, not inside any individual chat.
2. Memory is a separate feature that lets ChatGPT remember facts across conversations, it's available on ChatGPT Plus and Team, and must be turned on manually.
3. Custom Instructions have two fields: one for background about you, one for how you want ChatGPT to respond. Each field holds up to 1,500 characters.
4. Memory stores specific facts ChatGPT learns during chats (like 'user manages a team of 12') and applies them in future conversations automatically.
5. You can view, edit, and delete every memory ChatGPT has stored, nothing is hidden from you.
6. Custom Instructions do NOT carry into ChatGPT's memory system and vice versa, they are two separate systems that work together.
7. Temporary chats (available under the chat menu) bypass both Custom Instructions and Memory entirely, useful when you need a clean slate.

Custom Instructions: Your Permanent Professional Profile

Think of Custom Instructions as a one-page briefing document you hand to a new assistant on their first day. Instead of re-explaining your job title, your company's industry, your audience, and your preferred communication style in every single conversation, you write it once and ChatGPT reads it automatically before responding to anything you ask. The result is that every chat starts with context already loaded. A marketing director at a B2B software company gets different default responses than a high school principal, because their Custom Instructions tell ChatGPT exactly who they are and what they need.

The two fields serve distinct purposes. Field one, 'What would you like ChatGPT to know about you?', is your professional profile. Include your job role, industry, team size, the types of tasks you do most often, and any constraints that matter (budget ranges, company size, geographic market). Field two, 'How would you like ChatGPT to respond?', controls output style. This is where you specify tone (formal vs. conversational), format preferences (bullet points vs. prose), length defaults (brief summaries vs. detailed breakdowns), and any language rules (avoid jargon, always use active voice, never use bullet points for emotional topics).

• Job title and department: 'I'm a Regional Sales Manager overseeing 8 account executives in the Midwest.'
• Industry and company size: 'We sell HR software to mid-market companies with 200–2,000 employees.'
• Common task types: 'I frequently write sales emails, prep for client calls, and summarize CRM data.'
• Audience details: 'My audience is usually HR directors and CFOs, they're busy and skeptical of vendor claims.'
• Tone preference: 'Write in a confident, direct tone. No corporate fluff. Short paragraphs.'
• Format defaults: 'Lead with a summary sentence, then use bullet points. Keep responses under 300 words unless I ask for more.'
• Restrictions: 'Never suggest tools that require IT approval or coding. Assume I'm working in Microsoft 365.'

Memory: ChatGPT That Actually Learns About You

Memory works differently from Custom Instructions. Where Custom Instructions are static text you write manually, Memory is dynamic. ChatGPT updates it automatically as you have conversations. Tell ChatGPT in passing that your biggest client is in the healthcare sector, and it may store that fact and reference it weeks later when you ask for help drafting a proposal. Ask for feedback on a presentation and mention your boss prefers data-heavy slides. Memory can log that and apply it the next time you ask for slide help. It's the difference between a tool and something that starts to function like a genuine professional assistant.

Memory is not passive or invisible. ChatGPT will sometimes tell you when it's saving something ('I'll remember that your team uses Salesforce'). You can also explicitly instruct it: 'Remember that I present to non-technical executives' or 'Don't remember anything from this conversation.' At any time, go to Settings → Personalization → Manage Memory to see every stored fact, edit inaccuracies, or delete anything you don't want retained. This level of control matters, especially if you discuss sensitive client information or want to keep certain work contexts completely separate.

1. Go to ChatGPT.com and click your profile icon in the bottom-left corner.
2. Select 'Settings', then navigate to 'Personalization'.
3. Toggle 'Memory' to ON, it may already be on if you have a Plus account.
4. Start a conversation and tell ChatGPT something specific: 'I manage a team of 6 customer success managers at a fintech startup.'
5. Watch for ChatGPT to confirm it's storing the information, or check Settings → Personalization → Manage Memory afterward.
6. Review the stored memory entry, edit it if the wording is imprecise or delete it if you don't want it retained.
7. In a new conversation, ask something related ('Help me write a message to my team') and notice ChatGPT uses the stored context without being told.

Click to enlarge

What ChatGPT Cannot Remember. And Why It Matters

Memory has real boundaries. ChatGPT does not retain the actual text of your past conversations, it stores summarized facts and preferences, not transcripts. If you had a detailed strategy discussion three weeks ago, ChatGPT won't remember the specifics of what was said, only any high-level facts it logged (like 'user is planning a product launch in Q2'). This means you cannot use Memory as a conversation archive or a searchable log of past work. For that, you need to save your own copies, paste important outputs into a Google Doc, Notion page, or email to yourself.

There's also a professional judgment call here. Memory works across all your conversations, personal, professional, and everything in between. If you use one ChatGPT account for both work and personal tasks, stored memories can bleed between contexts in unexpected ways. A memory logged from a personal conversation ('user is going through a career transition') could surface in a work context inappropriately. The cleanest solution: use separate ChatGPT accounts for personal and professional use, or use Temporary Chat mode for any conversation you don't want influencing your professional context.

Click to enlarge

Quick Reference: Custom Instructions + Memory Cheat Sheet

• Custom Instructions location: Settings → Personalization → Custom Instructions
• Memory location: Settings → Personalization → Memory (toggle on/off here)
• View stored memories: Settings → Personalization → Manage Memory
• Delete one memory: Manage Memory → click the trash icon next to any entry
• Delete all memories: Manage Memory → 'Clear All Memory' button
• Tell ChatGPT to remember something: Start with 'Remember that...' in any message
• Tell ChatGPT to forget something: 'Forget that I mentioned [X]' or delete manually in settings
• Block memory for one session: Start message with 'Don't save anything from this conversation'
• Use a clean-slate chat: Click the pencil/new chat icon → select 'Temporary Chat' from the menu
• Character limit per Custom Instructions field: 1,500 characters (roughly 250–300 words)
• Conflict resolution: In-chat instructions override Memory, which combines with Custom Instructions
• Data tip: Use job titles and anonymized descriptions, not real client names or employee data

Click to enlarge

Key Takeaways from Part 1

• Custom Instructions are a static profile you write once, they load automatically in every new chat and eliminate repetitive context-setting.
• Memory is a dynamic system that learns facts from your conversations and applies them going forward, it's separate from Custom Instructions and must be enabled manually.
• The two fields in Custom Instructions serve different purposes: Field 1 is your professional background, Field 2 is your output preferences.
• You have full control over Memory: view it, edit it, delete individual entries, or wipe everything at once through Settings.
• Temporary Chat mode bypasses both systems entirely, use it when you need a genuinely clean slate or are working with sensitive information.
• Never store specific client names, employee details, or confidential figures in Memory, use anonymized, role-based descriptions instead.

Now that you understand what Custom Instructions are and how to activate them, the real work begins: writing instructions that actually change how ChatGPT responds. Most people write vague, generic instructions and wonder why the output still feels off. The difference between mediocre and excellent Custom Instructions comes down to specificity, structure, and knowing exactly which field to use for which type of information.

1. Custom Instructions have two separate fields, they serve different purposes and should never be mixed up.
2. The 'About You' field shapes context; the 'Response Style' field shapes behavior, both must be filled for best results.
3. Specificity beats generality every time: 'marketing manager at a B2B SaaS company' outperforms 'I work in marketing.'
4. You can update Custom Instructions anytime, treat them as a living document, not a one-time setup.
5. Memory and Custom Instructions are separate features, they work together but don't replace each other.
6. ChatGPT doesn't ask clarifying questions by default, your instructions must pre-answer them.
7. Instructions that contradict each other cause inconsistent output, audit your fields for conflicts regularly.

Writing the 'About You' Field That Actually Works

The first field, 'What would you like ChatGPT to know about you?', is your professional context briefing. Think of it as the one-page background document you'd hand a new consultant before they started working with you. It should cover your role, your industry, your audience, your goals, and any constraints ChatGPT needs to respect. A sales director at a healthcare company should say exactly that, not 'I work in sales.' The more precisely you describe your world, the more relevant every response becomes. Include your company size, who you communicate with most, and what kind of outputs you produce regularly.

Don't treat this field like a résumé. You're not listing accomplishments, you're giving operational context. What decisions do you make every week? What problems do you solve repeatedly? Who reads your work? A school principal might write: 'I lead a K-8 public school with 400 students. I communicate daily with teachers, parents, and district administrators. I write policy memos, staff newsletters, and parent communications. My tone needs to be professional but warm, and I avoid educational jargon when writing to parents.' That's a briefing. That changes outputs dramatically compared to 'I work in education.'

• Your job title and department (be specific: 'Head of People Operations' not 'HR')
• Your industry and company type (e.g., 'mid-size regional accounting firm,' 'e-commerce startup')
• Your primary audience (who reads your emails, reports, proposals)
• Your recurring outputs (weekly status reports, client proposals, board presentations)
• Any domain-specific language you use, or want to avoid
• Geographic or cultural context if it affects your communication style
• Your biggest recurring challenge or time sink (so ChatGPT can anticipate it)

Click to enlarge

Writing the 'Response Style' Field With Precision

The second field, 'How would you like ChatGPT to respond?', is where you control behavior, not background. This is where you set tone, format, length, and communication rules. Most people write something like 'be concise and professional' and stop there. That's not a style guide, that's a vague hope. A strong response style field reads more like an editorial brief. It tells ChatGPT exactly how to structure answers, what to skip, when to ask questions versus when to just produce output, and how formal or casual the language should be for different situations.

The most effective response style instructions address three layers: format preferences (bullet lists vs. prose, headers vs. flowing text), length expectations (one paragraph vs. full document), and communication rules (always include a subject line when drafting emails, always give three options when I ask for ideas, never use passive voice). You can also set negative rules, things ChatGPT should stop doing by default, like adding lengthy disclaimers, over-explaining obvious things, or starting every response with a restatement of your question. Negative rules are often the most powerful edits you can make.

1. Set a default format: 'Use bullet points for lists of 3+ items; use prose for explanations and analyzis.'
2. Define length expectations: 'Keep responses under 200 words unless I ask for a full draft.'
3. Specify tone: 'Professional but direct, no filler phrases, no excessive politeness.'
4. Add communication rules: 'When drafting emails, always include a suggested subject line.'
5. Set idea-generation defaults: 'When I ask for options, give exactly three with a one-line rationale each.'
6. Add negative rules: 'Do not add disclaimers or caveats unless they are genuinely critical.'
7. Define when to ask questions: 'If my request is ambiguous, ask one clarifying question before proceeding.'
8. Specify vocabulary: 'Avoid jargon. Write at a level appropriate for a smart non-expert reader.'

How Memory Works, and What It Can't Do

Memory is ChatGPT's ability to retain information across separate conversations. Without it, every new chat starts completely blank. ChatGPT has no idea you spoke yesterday, last week, or ever. With Memory enabled, ChatGPT saves specific facts from your conversations and uses them in future sessions. It might remember that you prefer bullet-point summaries, that your key client is called Meridian Group, or that you're preparing for a board presentation in March. You can see exactly what ChatGPT has saved, edit any entry, and delete anything you don't want retained.

Memory works differently from Custom Instructions in one critical way: Custom Instructions are things you deliberately write and control. Memory is things ChatGPT learns by observing patterns in your conversations, unless you manually add memories yourself. That means Memory builds over time as you use ChatGPT. It's passive accumulation versus active configuration. Both serve the same ultimate goal, making ChatGPT feel like a tool that knows you, but they do it through different mechanisms. Smart users treat them as complementary: Custom Instructions for permanent preferences, Memory for evolving context and project-specific details.

Quick-Reference Cheat Sheet: Custom Instructions and Memory

• Field 1 (About You): Role + industry + audience + recurring outputs + key constraints, write it like an onboarding brief
• Field 2 (Response Style): Format + length + tone + email rules + idea-generation defaults + negative rules
• Character limit: 1,500 per field (~250 words), prioritize your highest-frequency tasks
• Specificity rule: 'B2B SaaS marketing manager targeting CFOs' beats 'I work in marketing' every time
• Negative rules are powerful: tell ChatGPT what to stop doing, not just what to start doing
• Memory = passive accumulation of facts across conversations; Custom Instructions = deliberate permanent preferences
• Check your Memory log regularly: Settings > Personalization > Memory, edit or delete anything inaccurate
• Never store sensitive client data, employee records, or confidential business information in either field
• Test Custom Instructions in a fresh chat, old conversations carry their own context that interferes with testing
• Update your Custom Instructions quarterly or whenever your role, projects, or priorities shift significantly

Key Takeaways from This Section

• The 'About You' field is a professional briefing document, operational detail produces dramatically better outputs than vague job titles
• The 'Response Style' field is an editorial brief, it controls format, tone, length, and communication rules simultaneously
• Negative rules (what ChatGPT should stop doing) are often more impactful than positive instructions
• Memory and Custom Instructions serve different purposes: use Custom Instructions for permanent preferences, Memory for evolving project context
• Both fields have hard character limits, ruthless prioritization is required, not optional
• Never include sensitive, confidential, or personally identifiable information in either field

Custom Instructions and Memory are not set-and-forget features. They require maintenance, intentional design, and the occasional reality check. Once you understand how to troubleshoot common failures, manage memory hygiene, and build a living instruction set that evolves with your role, ChatGPT stops feeling like a generic tool and starts behaving like a well-briefed colleague who actually knows your context.

7 Things Every Professional Should Know About Custom Instructions and Memory

1. Custom Instructions apply to every conversation, they are your standing brief, not a one-time setting.
2. Memory stores facts ChatGPT has learned across sessions. Custom Instructions store what you manually write.
3. Both features are available on ChatGPT Plus ($20/month), free users get limited or no memory.
4. You can edit, delete, or pause memory entries individually without wiping everything.
5. Custom Instructions have a character limit, prioritize your highest-impact context.
6. Temporary chats bypass memory entirely, use them when you need a clean slate.
7. Neither feature is private from OpenAI's systems, never store passwords, client PII, or sensitive financial data.

Troubleshooting: When ChatGPT Ignores Your Instructions

The most common complaint from professionals using Custom Instructions is this: "It's not following what I wrote." There are three likely causes. First, your instruction is too vague, "be professional" means nothing specific to an AI. Second, you've hit the character limit and the instruction is getting cut off silently. Third, a specific prompt in the conversation is overriding the instruction because explicit in-chat direction always takes priority over background settings.

The fix is almost always specificity. Replace "keep responses concise" with "limit all responses to three paragraphs or fewer unless I explicitly ask for more." Replace "I work in marketing" with "I manage B2B content marketing for a SaaS company targeting HR directors at mid-sized firms." The more concrete your instruction, the less room there is for ChatGPT to interpret it loosely. Treat your Custom Instructions like a job description, not a personality sketch.

• Vague instruction: "Write in a friendly tone" → Better: "Use a warm, direct tone, no jargon, no corporate-speak, short sentences."
• Vague instruction: "I'm in sales" → Better: "I'm a B2B account executive selling logistics software to operations managers at manufacturing companies."
• Vague instruction: "Help me be productive" → Better: "Always give me a recommended next step at the end of every response."
• Vague instruction: "Don't be too formal" → Better: "Avoid phrases like 'Furthermore,' 'It is imperative,' and 'As per our discussion.'"
• Vague instruction: "Know my audience" → Better: "My audience is non-technical HR managers who distrust jargon and respond to concrete examples."

Click to enlarge

Managing Memory: Keeping It Useful Over Time

Memory accumulates. After weeks of use, ChatGPT may have stored dozens of facts about you, some accurate, some outdated, some just wrong. A promotion, a new client base, a changed communication style, none of these update automatically. Left unmanaged, memory becomes a liability. ChatGPT might reference a project you finished six months ago as if it's still active, or address you by a role you no longer hold.

Build a monthly memory review into your workflow. Go to Settings, open the Memory panel, and read through every entry. Delete anything outdated. Correct anything inaccurate. You can also tell ChatGPT directly: "Please update your memory. I'm no longer working on the Henderson account. My current focus is the Q3 product launch." It will revise the stored entry. Think of memory management the same way you think about cleaning up your email folders, neglect it and it works against you.

1. Go to Settings → Personalization → Manage Memory to see all stored entries.
2. Delete entries that are outdated, incorrect, or no longer relevant to your work.
3. Correct entries by telling ChatGPT directly what has changed, it updates in real time.
4. Use Temporary Chat for sensitive conversations you don't want stored.
5. Tell ChatGPT what to remember: "Remember that my primary deliverable this quarter is the annual report."
6. Do a full memory audit every 30 days, schedule it like a recurring calendar task.

Building Instructions That Evolve With Your Role

Your Custom Instructions should be a living document, not a one-time setup. The professionals who get the most out of ChatGPT treat their instructions like an onboarding document they revise every quarter. When your responsibilities shift, when you take on a new client type, when your team structure changes, your instructions should reflect that. Static instructions create a growing gap between what ChatGPT thinks you need and what you actually need.

The most effective approach is to keep a running notes document, in Notion, Google Docs, or even a simple text file, where you draft and refine your Custom Instructions before pasting them into ChatGPT. This lets you version-control your instructions, compare what worked previously, and update systematically rather than reactively. Professionals who do this report dramatically more consistent output because ChatGPT is working from a precise, current brief every single time.

Quick Reference: Custom Instructions and Memory Cheat Sheet

• Custom Instructions = what you manually write; Memory = what ChatGPT learns from your conversations.
• Both features require ChatGPT Plus for full functionality ($20/month).
• Keep instructions specific, vague instructions produce generic results.
• Test instructions by asking ChatGPT to describe your role and preferences in a new chat.
• Audit memory monthly, delete outdated entries, correct wrong ones.
• Use Temporary Chat for sensitive topics you don't want stored.
• Never store passwords, client PII, or confidential data in instructions or memory.
• Update instructions after promotions, new projects, or major workflow changes.
• Draft and version your instructions in a separate document before pasting into ChatGPT.
• In-chat prompts override Custom Instructions, use explicit direction for one-off exceptions.

Key Takeaways

• Custom Instructions and Memory together create a persistent professional profile that eliminates repetitive context-setting.
• Specificity is everything, the more precise your instructions, the more consistent and useful the output.
• Memory requires active management; treat it like a professional profile you update quarterly.
• Troubleshooting almost always comes down to vague instructions or unmanaged memory conflicts.
• Your instruction set should evolve with your role, static instructions create a growing gap between what ChatGPT knows and what you need.
• Data hygiene matters. Temporary Chat exists for a reason; use it for anything confidential.