You told ChatGPT last month that you manage a five-person ops team, prefer bullet lists, and hate vague “synergy” advice. This morning it drafts a customer email in the wrong voice, forgets your product name spelling, and invents a meeting cadence you never agreed to. You did not break the product. You just treated three different “memory” features as if they were one sticky brain.
This is Part 5 of Learn ChatGPT from scratch: how custom instructions, memory, Projects, and (on some paid tiers) scheduled tasks actually differ. Parts 1 through 4 covered what ChatGPT is, plans without hype, first thirty minutes on web and mobile, and the desktop Chat / Work / Codex split. Part 6 moves into voice, images, and files. Here we map the personalization stack so you stop pasting the same company brief into every new chat, and so you stop leaking sticky details you never meant to keep.
What you will learn
- How custom instructions differ from memory (you write one; the product can accumulate the other)
- Where to find, edit, and delete memories, and why Free accounts feel more limited
- When a Project beats a long chat thread (files, scoped instructions, project memory)
- What scheduled tasks are good for, and that they are not on every plan
- A starter kit for one work Project and one personal Project without mixing streams
- Privacy habits: consumer training opt-out vs Enterprise-style defaults, Temporary Chat, and what not to store as memory
Product labels move. Ground truth for this post is OpenAI’s Help Center articles on custom instructions, memory, Projects, scheduled tasks, and data controls. Re-check those pages before you write team policy. UI paths below match the Help Center at research time (Settings → Personalization on web and desktop; Customize ChatGPT on mobile for some toggles).
Three boxes, not one brain
People say “ChatGPT remembers me” the way they say “the laptop knows my Wi-Fi.” Under the hood you are managing separate stores with different rules, edit surfaces, and failure modes.
Picture three boxes side by side, plus a fourth clock for tasks:
- Custom instructions. Text you write on purpose. Applied across chats when customization is enabled. Best for standing style and role (“I’m a data analyst; prefer tables; no fluff openers”).
- Memory. Context the product can keep so you do not re-explain yourself every session. You can review, edit, and delete. Free plans can feel more limited in how much sticks and how often personalization fires. Sensitive facts should not live here by accident.
- Projects. A workspace that groups chats, reference files, and project-level instructions for one ongoing goal. Project memory can stay focused on that project’s history instead of everything else you ever typed.
- Scheduled tasks (some paid tiers). One-off or recurring prompts ChatGPT runs later and notifies you about. Useful for reminders and light monitoring. Not a substitute for a real job scheduler or your CRM workflows.
Rule of thumb: Custom instructions are the laminated card on your desk. Memory is the sticky notes the product keeps rewriting. Projects are the labeled binders. Scheduled tasks are the calendar invites. Treat them as four tools, not one magic recall.
Custom instructions: standing prefs you control
Custom instructions are the feature for people who are tired of pasting the same preamble. OpenAI’s Help Center describes them as guidance ChatGPT should consider in its responses, applied when customization is on. You edit them; they do not silently invent new “facts” about your company from a side chat you had at 11 p.m.
Where they live (typical paths)
- Web and desktop: Settings → Personalization → Custom Instructions (enable customization, then write the text)
- iOS and Android: Settings → Customize ChatGPT (toggle enable customization, then enter instructions)
Updates apply to future behavior right away. Old chat transcripts still show whatever the model said then. If a past thread is polluted, delete or archive that chat rather than assuming an instruction rewrite rewrites history.
Character limits (plan-dependent)
At research time, OpenAI documented roughly 1,500 characters for Free (and Go) users and up to 5,000 characters for Plus, Pro, Enterprise, Business, and Education. Treat those numbers as Help Center facts to re-check, not eternal constants. The practical advice stays the same: shorter instructions that are true beat a novel that contradicts itself.
What to put in (and what to leave out)
Good global instructions sound like a style guide, not a vault of secrets:
- Role and audience (“I write for non-technical managers; define jargon once”)
- Output shape (“lead with the answer; then bullets; then risks”)
- Hard “don’ts” that are not secrets (“no emojis; no fake citations; flag uncertainty”)
- Light durable prefs (“US English; metric units unless I say otherwise”)
Leave out SSN, passwords, API keys, card numbers, private customer lists, raw HR files, and anything your employer classifies as restricted. Instructions are convenient, not a secure vault. OpenAI also notes that if you use third-party plugins or tools, relevant bits of instructions may be shared with those tools. Only use integrations you trust.
Example instruction block you can adapt
Role: I am an analytics lead at a mid-size B2B company.
Audience: busy managers who want clear numbers and plain English.
Format: start with the answer in 2 sentences, then bullets, then risks or unknowns.
Style: concrete examples; no hype; no fake studies; say when you are guessing.
Defaults: US English; prefer tables for comparisons; ask one clarifying question if the metric grain is unclear.
Do not: invent tool screenshots, invent legal advice, or invent SQL that I did not ask to verify.That block is short enough for Free-tier character limits if you trim a line or two, and useful enough that every new chat starts less random. For project-specific tone (marketing launch vs personal tax notes), use project instructions instead of stuffing everything into the global field.
Memory: sticky context you should audit
Memory is the feature people mean when they say “it finally stopped forgetting my job title.” When memory is on, ChatGPT can keep useful context from chats (and related sources, depending on the current product generation) so personalization continues next week. OpenAI’s Memory FAQ is the place to re-read when the UI shifts between “saved memories,” “memory summary,” and related controls.
Editable is the point
You are not stuck with whatever the model inferred. Typical controls (again, confirm in Settings → Personalization → Memory):
- Turn memory on or off entirely
- Review a memory summary or saved memory list
- Edit or delete specific items
- Ask in chat what it remembers, then correct it in plain language
- Use Temporary Chat when you want a one-off conversation that should not create or use memories
Deleting a chat does not always erase every memory that chat produced. Treat memory cleanup as its own chore, the way you clean browser cookies after a research binge. If something sensitive appeared in memory, remove it from memory and from the chat history if you still care about the residue.
Free plan limits (practical view)
Memory availability and depth can feel thinner on Free. That is not a moral failing of your workflow. It is a product tier choice: consumer Free is for trying the product and light daily use, not for building a second brain of your entire career. Paid consumer plans usually unlock more headroom and features around personalization. Exact caps change; if Free feels forgetful, either upgrade for volume you actually use or keep durable truth in Project files and short instructions instead of expecting infinite recall.
Memory vs custom instructions
| Question | Custom instructions | Memory |
|---|---|---|
| Who writes it first? | You, on purpose | You can force it; product can also extract useful bits |
| Best for | Standing style, role, hard don’ts | Facts that recur (role, tools, goals, constraints) |
| Edit surface | One instruction field | Summary / saved list / chat corrections |
| Risk if messy | Contradictory style rules | Stale or sensitive facts that keep resurfacing |
| Off switch | Disable customization | Disable memory; Temporary Chat for one-offs |
OpenAI’s own FAQ line is simple: put explicit standing guidance in custom instructions; let memory (when on) carry conversational details you do not want to retype. When the two disagree, fix the source of truth. Do not argue with the model in five different chats.
What not to let memory keep
- Credentials, one-time codes, private keys
- Customer PII dumps, full HR packets, medical details you would not put in a shared Google Doc
- Unverified gossip about coworkers
- Temporary project codenames that must not bleed into personal chats
- Anything your company policy marks as “not for consumer AI tools”
If you need a sandbox for messy drafts, Temporary Chat (when available) is designed so chats are not saved to history the usual way, do not create memories, and are not used for training in the standard consumer training sense described in Data Controls help. Confirm the Temporary Chat FAQ for the current retention window (commonly discussed as about 30 days for abuse monitoring).
Projects: binders for long-running work
A long sidebar full of “Untitled” chats is how people lose decisions. Projects fix the organizational problem first: one place for chats, files, and instructions that belong to a single goal. OpenAI documents Projects for free and paid users (you must be logged in). Business, Enterprise, and Edu get richer team sharing and admin controls; consumer plans can share projects too, with plan-based collaborator and file caps that you should re-check in Help.
What a Project holds
- Chats created inside the project (or moved in)
- Files you upload as reference (PDF, sheets, docs, images, pasted text; limits depend on plan)
- Project instructions that apply only inside that project and override global custom instructions
- Project memory so context stays focused (including project-only options useful for sensitive or long-running work)
- Optional saved responses as project sources when a draft is worth reusing
Project instructions override global ones
This is the feature people miss. Global custom instructions still matter for everyday chat outside a project. Inside a project, project instructions win. That is how you can keep a calm, concise default for personal use and a brand-voice, checklist-heavy style for “Q3 launch brief” without rewriting the global field every Monday.
Project: Q3 launch brief
Instructions:
- Act like a careful product marketing partner.
- Always ground claims in the uploaded brief and competitive notes.
- Flag anything not in the files as a guess.
- Prefer short headlines, then bullets, then open questions.
- Never invent pricing or ship dates.Project-only memory vs default memory
When you create a project, you may be able to choose project-only memory versus a default setting. Project-only means chats lean on context inside that project and do not pull your broader personal saved memories the same way. Shared projects are typically forced into project-only style boundaries so teammates do not inherit your private life from other chats. Exact Enterprise vs consumer matrix tables live in the Projects Help article; if you do sensitive work, prefer project-only and keep personal memories out of the binder.
File limits (re-check, but plan for caps)
OpenAI documents plan-based file counts per project (examples at research time: Free around 5 files; higher tiers up to 25 or 40). Storage and upload rate limits also apply across chats and Projects. When you hit a wall, delete stale files, split into multiple projects, or put the canonical doc in a company drive and paste only the sections you need (Part 6 covers upload hygiene in depth).
Project starter kit (20 minutes)
- Name by goal, not by tool. “Weekly ops review,” not “ChatGPT stuff.”
- Add one brief file (goals, audience, definitions, non-goals). One honest page beats ten half-related PDFs.
- Add one or two example outputs you already like (a past email, a report section). Models copy shape better than vibes.
- Write 3 to 8 lines of project instructions with role, format, and hard don’ts.
- Chat only inside that project for that goal for a week. Move old related chats in if the UI allows.
- Save strong answers back into project sources when the product offers “Save to project.”
- Review memory and files monthly. Delete junk. Update the brief when reality changes.
Starter ideas that match OpenAI’s own examples: school study guide with PDFs and study mode, marketing launch plan, personal event planning. For AMS readers, strong work patterns include “metric definition cleanup,” “customer interview synthesis,” and “SQL review notes for team onboarding,” each with a short glossary file and two good sample answers.
Scheduled tasks: paid automation, not magic ops
Scheduled tasks let you ask ChatGPT to run a prompt later: a reminder, a recurring checklist, or a monitoring-style check that notifies you when something meaningful changes. OpenAI documents them for Plus, Pro, Business, and Enterprise (availability and active-task caps vary by plan). Free users should assume they do not get the full task system until the Help Center or their account says otherwise.
What they are good for
- Personal reminders with a useful draft attached (“Sunday 6 p.m.: draft my weekly priorities from this checklist”)
- Light recurring research digests on public topics
- Monitoring-style prompts when the product supports “notify only on change”
What they are bad for
- Mission-critical SLA monitoring (use real observability tools)
- Anything that needs guaranteed access to private project files (OpenAI notes tasks created in a project may not access those project files)
- Voice chats and Custom GPT-driven tasks (documented unsupported combinations exist; re-check Help if you rely on them)
- Secrets in the task instruction text (the prompt will run unattended)
Create tasks from the Scheduled page in the sidebar (web, mobile, desktop Chat experience) or by asking in chat. Edit, pause, and delete from the task UI. Active task limits (examples from Help at research time: Plus around 5, higher tiers higher) mean you should prune zombie reminders the way you prune calendar spam.
Worked example: one personal stream, one work stream
Maya uses ChatGPT on a Plus plan. She writes global custom instructions once: plain English, tables for comparisons, never invent citations. She turns memory on for convenience, then every Friday she opens Memory and deletes anything that looks like a password, a customer name dump, or an old “I’m interviewing at Company X” note she no longer wants floating around.
She creates two Projects:
| Project | Files | Instructions vibe | Memory choice |
|---|---|---|---|
| Ops weekly review | Metric glossary PDF, last week’s outline | Skeptical analyst; flag missing grain | Project-only if available |
| Home renovation budget | Contractor quotes (redacted phones) | Practical planner; no fake permits advice | Default or project-only; no work secrets here |
When she needs a throwaway “is this joke funny?” chat, she uses Temporary Chat so it does not pollute memory. When she wants a Monday 8 a.m. nudge to list open tickets from a pasted summary she will re-supply, she uses a scheduled task with no credentials in the prompt. When her company later moves her to Enterprise, she re-reads admin policy: workspace training defaults and retention can differ sharply from her old consumer account.
Privacy and training: consumer opt-out vs work defaults
Personalization features sit next to data controls. They are related but not identical.
- Consumer plans (Free, Plus, Pro, and similar): you can typically opt out of training via Data Controls → turn off Improve the model for everyone. History can still exist; training is the separate question. Re-check the Data Controls FAQ for signed-in vs signed-out paths.
- Business / Enterprise / Edu: OpenAI documents stronger enterprise privacy defaults (including not training on business workspace content by default in the usual Enterprise privacy story). Admins control feature toggles, retention, and sharing. Your consumer habits do not automatically become company policy.
- Projects training note (consumer): Free/Plus/Pro project content may be used for training when improve-the-model is on. Shared projects have additional rules about when training applies across contributors. Read the Projects Help section on data, not just the marketing page.
None of this makes a consumer chat a HIPAA vault. If your employer issues a workspace, use it for work. If you only have a personal account, keep work data off it unless policy explicitly allows the tool and the data class.
Common mistakes
| Mistake | What goes wrong | Fix |
|---|---|---|
| One mega custom instruction for every life context | Contradictions; brand voice leaks into personal chats | Short global style + project instructions per goal |
| Never reviewing memory | Stale job titles, old goals, sensitive residue | Monthly memory audit; delete aggressively |
| Putting secrets in instructions “so it remembers” | Broad exposure surface | Password manager for secrets; redacted briefs only |
| Doing all work in one endless chat | Context mud; hard to find decisions | Project per goal; move or start fresh inside it |
| Assuming Free equals full personalization stack | Frustration and bad mental models | Match workflow to tier; re-check plan limits |
| Scheduled task with private file assumptions | Task cannot see project files / wrong data | Paste minimal safe inputs; use real automation for critical jobs |
| Mixing personal and employer accounts casually | Policy and training mismatches | Separate streams; prefer company workspace for work |
How to practice this week
- Open Settings → Personalization. Write or trim custom instructions to under one screen.
- Open Memory. Delete anything you would not paste into a shared team doc.
- Create one Project named for a real goal. Add a one-page brief and two example outputs.
- Run three chats only inside that Project. Save one good answer as a source if the UI allows.
- If your plan includes tasks, create one low-stakes recurring reminder with no secrets in the prompt. Pause it after you confirm notifications work.
- Open Data Controls. Decide consciously whether Improve the model for everyone is on. Match that choice to how sensitive your everyday prompts are.
Next in the series: Part 6, Voice, images, and files for everyday tasks, including the paste/upload ladder (never SSN/passwords/cards; high risk for customer/HR; careful with internal docs; safer with public or synthetic samples). For privacy depth later in the track, Part 8 covers work rules and when not to use ChatGPT at all. If you are comparing stacks, the Claude Code memory layers post on this site is a sibling mental model for agent tools, not a substitute for OpenAI’s ChatGPT settings.
Quick recap
- Custom instructions = standing prefs you write and can disable
- Memory = sticky context you must edit and prune (more limited on Free)
- Projects = scoped chats + files + project instructions (and project memory options)
- Scheduled tasks = some paid tiers; good for reminders and light monitoring, not critical ops
- Consumer training opt-out lives in Data Controls; Enterprise privacy is a different contract
- Never store credentials or regulated dumps as “so it remembers”
Sources
Research and further reading used for this article (OpenAI Help Center; re-check for UI and plan changes):
- OpenAI Help: ChatGPT Custom Instructions (enable paths, character limits by plan, data notes)
- OpenAI Help: Memory FAQ (how memory works, editing, Temporary Chat contrast, instructions vs memory)
- OpenAI Help: What is Memory (capability overview entry point; cross-link from product docs)
- OpenAI Help: Projects in ChatGPT (files, project instructions, project-only memory, sharing, plan file limits, training notes)
- OpenAI Help: Scheduled Tasks in ChatGPT (who can use tasks, caps, monitoring tasks, unsupported combos)
- OpenAI Help: Data Controls FAQ (Improve the model for everyone, Temporary Chats, export pointers)
- OpenAI Help: Temporary Chat FAQ (no history/memories/training behavior as documented)
- OpenAI Help: ChatGPT Capabilities Overview (memory, Projects, scheduled tasks in one map)
- OpenAI: Enterprise privacy (how business workspace data is treated vs consumer defaults)
- Analytics Made Simple: Learn (related learning paths on this site)
