To use ChatGPT as a to-do list, keep one long-running chat, paste your whole brain dump into it, and ask for one structure back: one line per task, the first physical step for anything vague, and three items for today. Each morning paste yesterday’s list back in and ask what moves. This works well for a day or a week. It stops working as a store, because ChatGPT memory, in OpenAI’s own words, “does not retain every detail from every conversation.” Use the chat for the rewriting, and keep the list itself somewhere that cannot forget.
That split, rewriting versus holding, is the whole argument of this page. Almost every guide to this topic is a list of prompts. Prompts are the easy part. The part that decides whether you are still doing this in March is what happens to the list while you are not looking at it, and that is a question about ChatGPT’s memory, its scheduled tasks and its connected apps, all of which OpenAI documents and almost nobody quotes.
The method, in five moves
Using ChatGPT as a to-do list is not a feature you switch on, it is a short routine. Five moves, each argued in full below:
- Keep one chat. One conversation that you return to, named something you will recognise in your history, instead of a new chat each morning.
- Dump everything in once. Unpunctuated, out of order, mid-sentence. The dump is the input, not the output.
- Ask for a structure, not an opinion. One line per task, projects split into a first step, blocked items marked, vague items flagged.
- Ask for today separately. A list of everything and a plan for today are two different requests, and merging them gets you neither.
- Close the loop in the evening. Paste what you did not finish and ask for it to be moved, in one request rather than item by item.
None of those five require a paid plan or any setup. What follows is why each one is shaped that way, and the three OpenAI features (memory, scheduled tasks, connected apps) that decide how far the routine can go.
1. One chat, not a new one every day
Starting a fresh chat each morning is the most common way people lose a ChatGPT to-do list, and it feels like the tidy option. It is not: a new chat starts with whatever memory happens to surface, not with your actual list, so by Thursday you have four partial lists in four conversations and no way to tell which one is current.
One long chat solves that with no features involved. The current list is the last message in a conversation you can find in two seconds, and the chat carries the shape of how you like the list formatted without being told again. Rename it to something blunt, so that you are picking it out of a history list, not reading previews.
The one cost of a long chat is that it eventually gets slow and the earliest part of it stops influencing anything. That is the point to start a second chat deliberately, pasting the current list in as the first message, rather than letting it happen by accident.
2. Dump it out, then ask for structure rather than sympathy
The reason a brain dump into ChatGPT works better than a brain dump into a notes app is that the dump does not have to be usable. You can write “chase the invoice thing, also the tax thing, mum’s birthday, that client who went quiet, the website is still broken” and get back five separate items, which is exactly the step that usually does not happen because it is boring.
What matters is the instruction you attach to the dump. Ask ChatGPT what you should prioritise and you get a confident paragraph of generic advice. Ask it for a specific transformation and you get the transformation. Four things are worth asking for every time:
- One line per task, with no preamble, no encouragement and no reordering into themes you did not ask for.
- A first physical step for anything that is really a project. “Sort out the tax thing” becomes “find last year’s return in the email archive”, which is startable.
- A mark on anything blocked on another person, because those items are the ones that silently make a list feel like failure.
- A flag on anything too vague to act on, rather than a guess at what you meant. A tool inventing detail in your list is worse than a tool admitting it cannot tell.
That shape is not an accident of prompting. The OpenAI research paper on how people use ChatGPT found that about two thirds of all writing requests ask ChatGPT to modify text the user already wrote rather than to produce new text, and “Practical Guidance” sits at roughly 29% of conversations. Rewriting what you already said is the thing this tool is genuinely reliable at, and a brain dump is exactly that job.
3. Ask for today separately from everything
A dump returns twenty eight items, and twenty eight items is not a plan. The second request is the one that makes the day work: give ChatGPT the structured list and ask for three items for today, chosen against a constraint you state out loud. The constraint is the part people skip.
Useful constraints are concrete and slightly unflattering. “I have about four usable hours and two meetings.” “Two of these have external deadlines this week, the rest do not.” “I have not slept and I need the smallest version of today.” Without a constraint, a model picks the three that look most important, which is nearly always three large items, which is how you end a day having started all three and finished none.
Ask for the three to be ordered by what has to happen first, not by importance, and ask for one of them to be small. The rest of the list is not deleted, it is simply not today’s problem, which is the mechanism a long list destroys and a short one restores.
What ChatGPT memory actually does, in OpenAI’s own words
Here is where a chat to-do list stops being a matter of technique. ChatGPT memory comes in two kinds, and neither is storage. Saved memories are details you explicitly ask ChatGPT to remember. Reference chat history lets ChatGPT draw on past conversations, and OpenAI notes that information from chat history “can change as ChatGPT updates what is most useful to remember.” The documentation is direct about what that means:
“Memory does not retain every detail from every conversation. ChatGPT decides which available information is relevant to a response.”
Read that as a product decision rather than a flaw, because that is what it is. Memory exists to make answers feel personal, not to be a database you can audit. A database that quietly drops rows it judged uninteresting would be a broken database. A memory that does exactly that is working as designed.
The practical consequence for a to-do list is a single rule: anything you cannot afford to lose has to be in the chat window you are looking at, not in memory behind it. That is why the routine above pastes the list forward each morning instead of asking “what was on my list?” The paste costs ten seconds and removes the entire question.
OpenAI also notes that memory features and controls vary by plan, region, platform and workspace settings, so what a colleague’s ChatGPT remembers is not evidence about yours.
Scheduled tasks, and the caps nobody mentions
Scheduled tasks are the closest thing ChatGPT has to a reminder. ChatGPT can run one-time or recurring prompts, monitor for changes and notify you when they run, which makes a daily morning check-in genuinely practical: “ask me what the three things are, and tell me what I said yesterday.”
The caps are the part worth knowing before you build a system on it. OpenAI documents a maximum number of active scheduled tasks per plan, and the numbers are small.
| Plan | Maximum active scheduled tasks |
|---|---|
| Free and Go | 3 |
| Plus | 5 |
| Business and Edu | 10 |
| Pro and Enterprise | 15 |
Free accounts are limited further: a recurring scheduled task can run no more than once a day, while paid plans allow hourly. OpenAI also notes that scheduled tasks do not support voice chats or GPTs, and that a task created inside a project cannot access that project’s files.
So scheduled tasks are a good fit for one or two standing rituals and a poor fit for per-task due dates. Five active tasks is a morning check-in, an evening sweep, and three spare. A working week has more than three things in it that need to resurface on a particular day, which is the structural reason a chat cannot replace a list with dates in it.
Projects, as a container for a recurring list
If the list you keep in ChatGPT is about one thing (a client, a move, a dissertation), Projects is the feature that makes it less fragile. OpenAI describes Projects as keeping “related chats, files, and instructions together”, and project instructions “apply only within that project and override your global custom instructions.”
That last part is the useful bit for a to-do list. The formatting rules you would otherwise repeat every morning, one line per task, no commentary, flag the vague ones, go into the project instructions once and apply to every chat inside it. A project can also be set to project-only memory, which keeps a client’s details out of the rest of your ChatGPT use.
Project file limits are documented per plan: 5 files on Free, 25 on Go and Plus, 40 on Edu, Pro, Business and Enterprise, with a maximum of 10 uploaded at once. Useful if the list leans on a brief or a spreadsheet, and irrelevant if it is just text.
Projects still does not make the chat a store. It makes the container tidier, and the list is still the last message you pasted.
When ChatGPT can actually edit a real list
There is one configuration where ChatGPT stops being a thinking surface and starts changing your real list: when the app holding that list is connected to ChatGPT. OpenAI states that connected apps let you “search and reference information from connected services” and “take supported actions, such as creating or updating information”, after you sign in and authorise the connection.
“ChatGPT may ask for approval before reading information or completing an action.”
Note the shape of that sentence, because it is the right shape and it is also the ceiling. An assistant that writes into your list should ask first, and OpenAI’s permission settings include an “Always ask” option. What you do not get is a preview of the exact change in the way a diff shows you one: approval is per action, and the action is described rather than laid out as a before and after.
Two further limits are worth knowing. App availability depends on the app, your plan, region, workspace, role, model and the interface you are using, so a connector a friend has may simply not be there for you. And an app can only do what it exposes, so “update the due date on everything I missed” may or may not be a supported action in your particular tool.
The broader version of this question, what any assistant should be allowed to do to your list and what it should have to ask about, is the subject of our piece on AI that asks first.
Four places a chat to-do list breaks
Each of these shows up in week two or three, not on day one, which is why the approach feels better than it is at the start.
- Nothing comes back to find you. A chat is a place you go. A list with dates is a thing that arrives. Without scheduled tasks, a ChatGPT to-do list only works on days you remember to open the chat, and the days you forget are exactly the days it mattered.
- The list has no single location. After a fortnight the current version is in the chat, an older one is in a memory, a third is in a message you sent yourself. A to-do list whose job is to stop you holding things in your head cannot have an ambiguous location.
- Re-pasting is a tax that grows. Ten seconds a day is nothing. Ten seconds a day plus the judgement call about whether this paste is current is the thing people quit over, usually without being able to say why.
- Nothing is shared or checkable. You cannot hand a chat to the other person in your household, and you cannot glance at it on your phone lock screen. Both are ordinary requirements of a real list.
None of those four is a reason to stop using ChatGPT for the rewriting. All four are reasons not to make it the store.
Why writing it down anywhere works, and what that implies
It is worth knowing why this helps at all, because the mechanism tells you which part is doing the work. Getting an intention out of your head and into something external is called cognitive offloading, defined by Risko and Gilbert as “the use of physical action to alter the information processing requirements of a task so as to reduce cognitive demand” (Trends in Cognitive Sciences, 2016).
The size of the effect is large and specific. A review of intention offloading by Gilbert and colleagues reports that in one of their studies participants forgot about 45% of intentions when relying on their own memory, against around 5% when using external reminders, which the review describes as reducing the forgetting rate “by almost an order of magnitude.” The reminder is not a small optimisation on top of memory. It is most of the result.
The second mechanism is why a plan feels different from a list. Masicampo and Baumeister ran six studies on unfulfilled goals and found that making a specific plan removed the intrusive thoughts and the performance interference that an unfinished goal produces.
“Once a plan is made, the drive to attain a goal is suspended, allowing goal-related cognitive activity to cease, and is resumed at the specified later time.”
Put the two findings next to each other and the implication for ChatGPT is direct. The value of asking ChatGPT to turn your dump into a plan is real and it is measured: plan making is what quiets the noise. The value of leaving that plan inside a chat window is not, because the offloading benefit depends on the external store being reliable. Asking a system that “decides which available information is relevant” to be your reminder is asking the wrong component to do the one job the research says matters.
The three prompts worth keeping
Three prompts cover the whole routine. Paste them somewhere you can reach, not into a chat you will lose.
1. THE DUMP, STRUCTURED Here is everything in my head. Do not advise me and do not reorder into themes. Return one line per task. For anything that is really a project, replace it with its first physical step. Put [WAITING] in front of anything blocked on someone else. Put [VAGUE] in front of anything you cannot turn into a startable step, and do not guess what I meant. <paste the dump> 2. TODAY, AGAINST A CONSTRAINT From that list, pick three for today. My constraint: <hours I actually have, and what is already in the calendar>. Order them by what has to happen first, not by importance. One of the three must be small enough to finish in 20 minutes. Leave the rest of the list alone. 3. THE EVENING SWEEP I finished <these>. I did not finish <these>. Rewrite the full list with the finished ones removed and the unfinished ones carried forward, and tell me which unfinished item has now been carried forward twice.
The last line of the third prompt is the one that earns its place. An item carried forward twice is almost never a scheduling problem, it is a task that is too big, not actually yours, or not actually going to happen, and naming it is the only thing that resolves it.
Chat versus a task manager, job by job
The comparison that matters is not feature counts, it is which component is trustworthy for which job.
| The job | ChatGPT in a chat | A task manager |
|---|---|---|
| Turning a dump into items | Strong. This is the thing it is reliably good at | Usually manual, one item at a time |
| Breaking a vague task into a first step | Strong, and it will admit when it cannot | Rare outside AI-native tools |
| Holding the list next week | Weak. Memory decides what is relevant | This is the entire product |
| Reminding you on the right day | Caps at 3 to 15 scheduled tasks by plan | Unlimited dates, and they arrive |
| Rescheduling everything you missed | Strong as a rewrite, then you paste it somewhere | Varies. Often twenty manual edits |
| One current version, on every device | Ambiguous after a fortnight | Guaranteed |
| Sharing with one other person | Not really possible | Standard |
When to keep going, and when to move off it
Keep using ChatGPT as your to-do list if the list is short-lived and bounded. A week of getting through a crisis, a one-off move, a project with an end date: a chat is genuinely the fastest tool for that and setting up anything else is procrastination wearing a productivity hat.
Move the list somewhere durable when any two of these are true: something was genuinely forgotten rather than postponed, you are not sure which chat is current, you are re-pasting more than once a day, or another person needs to see it. Those are not preferences, they are the failure signature of using a thinking surface as a store.
What you should not do is give up the rewriting when you move. The reason a chat felt better than your old list app is that the chat did the restructuring work your list app made you do by hand. Moving to a plain list app that cannot do that is how people end up back in a chat six weeks later.
Where Taskly fits
Taskly was built around the split this page describes: an AI-native task manager where the rewriting happens inside the list rather than in a separate window, and where the list is a store rather than a transcript.
- The brain dump goes straight in. Paste or dictate the whole mess and ask Otto to split it into separate items, turn the ones that are really projects into steps, and propose dates. Same transformation as the first prompt above, and the result is the list rather than a message about the list.
- Nothing changes without your approval. Otto shows every proposed change as a before and after that you accept or discard. Not an action described to you, the actual diff.
- Rescheduling is one request. “Move everything I did not finish to Thursday” is one approval instead of twenty edits, and the result stays in the list instead of needing to be pasted anywhere.
- Dates arrive. Every item can carry a date, with no cap on how many, and do dates are kept separate from real deadlines so a slipped intention does not read as a missed commitment.
- Ready for two. A shared list for a household or a client works without a team pricing story attached to it.
What Taskly cannot do, plainly. Taskly has no offline mode, so it needs a connection. There are no native mobile apps yet, it runs in a browser. There is no calendar layout, so if seeing the day as a timeline is the thing you need, Taskly is not that and a visual planner will serve you better. The integration surface is small, so Taskly is not yet the hub that pulls tasks out of ten other tools. And Otto is a good rewriter, not an oracle: it will propose a plan from what you gave it, and the judgement about whether that plan is right stays yours.
If the assistant part is what you are weighing up, the AI to-do list page covers what to check before trusting any tool with your list, and what an AI task manager is covers the category rather than the product.
The short version
Use ChatGPT for the job it is reliably good at: turning an unusable dump into one line per task, a first step for the vague ones, and three things for today against a constraint you name. Keep one chat, paste the list forward, and do not ask memory to hold anything you cannot afford to lose, because OpenAI says plainly that it decides for itself what is relevant. When the list starts needing dates, a second person or a single current version, the chat has run out of road, and the thing to carry across is the rewriting rather than the habit of starting over.
Frequently asked questions
Can ChatGPT be my to-do list?+
Does ChatGPT remember my tasks between chats?+
Can ChatGPT remind me about a task?+
Can ChatGPT add tasks to my to-do list app?+
What is the best prompt for turning a brain dump into a to-do list?+
Is it safe to put my whole to-do list into ChatGPT?+
When should I stop using ChatGPT as a to-do list?+
Is ChatGPT better than a to-do list app?+
Every limit described on this page was checked against OpenAI’s own help documentation on 5 October 2026. Plan caps, file limits and feature availability change, so check the linked pages before relying on a number here.
