How to Turn a ChatGPT Conversation Into a Project Plan (2026): Five Lists, Prompts and a Worked Example

Key Takeaways

• Storyflow is the strongest destination for a plan that starts as a ChatGPT or Claude conversation: the assistant builds the decisions, the task wall, the schedule table and a deck directly on one board, and can later read back what changed and who changed it.
• A conversation is not a plan. It is a log in which the current answer sits below five older answers to the same question. The work is the collapse, not the export.
• Collapse into exactly five lists: decisions (with what each one replaced), tasks (verb first), owners (one name per task), dates (stated or derived, never invented) and open questions (with a default and a person who answers).
• Ask for the collapse before you ask for a plan. A plan requested straight from a long chat quietly mixes superseded ideas with current ones.
• The Storyflow connection needs ChatGPT Plus or above with Developer mode, or any Claude plan (Claude Free allows one custom connector). It runs on the assistant's own model and spends no Storyflow AI credits.
• Asana, Notion and Trello all run their own MCP servers, so an assistant can write to them too. Asana wins when tasks need assignees and dependencies; Trello's server cannot permanently delete anything.
• Storyflow has no dependency engine, no calendar tool, no mobile or desktop app, and it is paid-only during early access. The section on when not to use it is specific.
• Budget one sitting for the whole job: collapse, check, build, share. The checking is the part you cannot delegate.

Why a ChatGPT Conversation Is Not a Project Plan

You spent three evenings talking a project through with ChatGPT. It was useful: the scope got sharper, a bad idea died, a launch date appeared. Then a teammate asks "so what are we doing?" and the honest answer is a link to 100 messages. The thinking happened. Nothing kept the result.

The current answer is buried under the old ones

A chat appends. When you change the price from three tiers to two in message 61, messages 14, 22 and 40 still say three. A human skimming the thread, or an assistant summarising it carelessly, has no way to know which one stands. We wrote about the mechanics of that drift in Why ChatGPT Loses the Plot After the Third Reply. A plan has one value per decision. A conversation has every value you ever considered.

Nothing in the chat has an owner or a date

Conversations produce intentions: "we should probably get the photos reshot." A plan needs that sentence turned into a task that starts with a verb, belongs to one person and has a day attached. Chats almost never do this unprompted, because nobody in the chat asked who and when.

The people who do the work were not in the room

The designer, the developer and the client did not read your conversation and will not. Sending them the thread hands them your thinking process when they need your conclusions. It is not a sharing problem. It is a format problem: the chat is shaped like a transcript, and the people doing the work need something shaped like a board.

The assistant cannot see what happened after

Once work starts, reality moves away from the conversation. Tasks get done, dates slip, someone adds a blocker. The chat knows none of it. A plan that lives somewhere the assistant can read means next week's question ("what moved, and what is late?") gets answered from the plan, not from memory. That gap is the subject of ChatGPT for Creative Projects: What It Cannot Do.

The Five Lists: What You Extract From the Chat

Every project conversation, however long, reduces to five lists. If a line from the chat does not belong on one of them, it is context, and context stays in the chat.

Decisions: record what each one replaced

A decision without its history gets reopened. Ask for every decision as a pair: the current position and the earlier position it replaced. "Two service packages (replaces the three-tier idea from early in the chat)" stops the third tier from walking back into the plan. Expect a long conversation to contain more reversals than you remember.

Tasks: verb first, small enough to finish

"Website copy" is a topic. "Write the three service page drafts" is a task. Ask the assistant to start every task with a verb and to split anything that needs more than one person or more than a few days. A plan of 20 to 30 small tasks is easier to run than one of 8 vague ones, because each small task has an obvious done state.

Owners: one name, or the word "unassigned"

Chats rarely name owners, so this list will have holes. That is the point. Tell the assistant to write "unassigned" where the conversation did not name a person, and never to guess from job titles. A visible gap gets filled in a two-minute conversation. A wrong guess gets discovered three weeks later.

Dates: stated, derived or missing

There are only three honest kinds of date. Stated: you said "we launch on 1 December". Derived: the copy must be final before design signs off, so it follows from another date. Missing: nobody said. Ask the assistant to label each date with which kind it is. A plan with six real dates and four labelled gaps beats a plan with ten confident dates, four of which are fiction.

Open questions: each one gets a default

An open question without a default blocks everything downstream. "Do we migrate the old blog posts? Default: yes, the 12 with traffic; Dana decides by 16 October." Now the project can proceed on the default, and the question has a deadline of its own.

How We Evaluated Where the Plan Should Live

Once you have five lists, they need a home. We compared five destinations people commonly reach for: Storyflow, Asana, Notion, Trello and Google Docs. The job being scored is narrow: a plan that starts as an assistant conversation and has to be run by people who were not in it.

The criteria and their weights

How the scores add up

The scores are our editorial judgment against vendor documentation read on 10 October 2026, not a lab measurement. Change the weights and the order changes: put follow-through at 30% and Asana wins, which is the honest reading of what Asana is for.

Where Storyflow loses points

Storyflow scores 4 on follow-through and 5 on cost, the two lowest cells in its row. It has no dependency engine and no free plan during early access. If your project is 60 interlocked tasks across a department, those two cells matter more than the weights here give them, and you should read the Asana review first.

What we did not score

We did not score general project management depth (portfolios, workload, time tracking), because the question here is the handoff from chat to plan. For the wider ranking, see The 10 Best MCP Servers for Project Management in 2026.

Comparison Table: Where a ChatGPT Conversation Can Become a Project Plan

Sources for each row are linked in the reviews below.

How to Turn a ChatGPT Conversation Into a Project Plan, Step by Step

Nine steps. The first six happen in the chat and work with any destination. The last three build the plan on a board.

Step 1: Stop adding to the conversation

Pick a moment and freeze the thinking. If you keep brainstorming while you extract, the lists go stale as you write them. Tell the assistant plainly: "We are done exploring. From here, do not suggest new ideas. We are turning this conversation into a plan."

Step 2: Ask for the decisions first, with what they replaced

Decisions come first because every task depends on them. Prompt: "List every decision we reached in this conversation. For each, give the final position, the earlier position it replaced if there was one, and a short quote from the conversation as evidence. Mark anything we discussed but did not settle as OPEN, not as a decision."

The evidence quote matters. It forces the assistant to ground each decision in what was said, and it lets you check the list in a minute instead of rereading the thread.

Step 3: Check the decisions yourself

Read the list against your own memory. You are looking for three errors: a superseded idea presented as current, a suggestion from the assistant that you never agreed to, and a decision you remember that is missing. Correct them in the chat ("Decision 4 is wrong, we dropped the referral scheme") before you go further. This is the one step where the assistant's speed does not help you. It summarised the chat; only you know what you meant.

Step 4: Ask for tasks, owners and dates as one table

Prompt: "From the confirmed decisions only, list every task needed to finish this project. One row per task: task (verb first), owner (a name from the conversation, or 'unassigned'), date (and whether it was stated, derived or missing), and which decision it serves. Do not invent owners or dates."

Asking for the decision each task serves exposes orphan tasks, the work left over from ideas you killed. If a task serves no surviving decision, delete it.

Step 5: Ask for the open questions with defaults

Prompt: "List every question this conversation left unanswered. For each: the question, a sensible default if nobody answers, who should answer it, and which tasks are blocked until it is answered."

Long project chats often leave more open questions than their author expects. Seeing them in one list is often the most useful output of the whole exercise.

Step 6: Work back from the deadline

If the conversation fixed an end date, ask the assistant to derive the milestones backwards from it: "The launch date is fixed. Working back from it, propose milestone dates for each phase, on weekdays, and tell me which stated dates conflict with that schedule." Derived dates get the label "derived". You approve or move them.

Step 7: Connect the assistant to Storyflow

You do this once. The address is https://mcp.storyflow.so/mcp.

• Claude: Customize, then Connectors, then Add, then Add custom connector. Name it Storyflow, paste the address, continue, sign in.
• ChatGPT (web): you need ChatGPT Plus or above with Developer mode. Turn Developer mode on under Settings, then Security and login. Open chatgpt.com/plugins, click the plus button, name it Storyflow, choose Public endpoint, paste the address, create it, sign in.

At sign-in you choose which projects the assistant may use, whether it may make changes, and whether it may use documents outside projects. For this job, pick the one project and allow changes. The full walkthroughs are in How to Connect ChatGPT to Your Project Board, How to Connect Claude to Your Project Board and the connection help article.

Step 8: Have the assistant build the board, one section at a time

Ask for sections, not "a plan". Prompt: "In my Storyflow project Studio Relaunch, build a board from the five lists we confirmed. Section 1: a Decisions wall, one note per decision with what it replaced. Section 2: a kanban of To do, Doing, Waiting and Done columns, one ticket per task, with the date on the ticket and the owner as its label. Section 3: a Schedule table with milestone, date, owner and status, rows in date order. Section 4: an Open Questions table with question, default, who answers and blocked tasks."

In Storyflow a kanban is a row of column walls holding tickets, and the assistant builds each wall with its tickets in one call. Open the board in a browser tab while it works: what the assistant adds appears live, so you can stop it early if a section goes wrong. In clients that support MCP apps, the board can also show inside the chat.

Step 9: Share it and make the board the source of truth

Send the board's view-only link to everyone who has to act. They open it with no account, and sign in to comment. Then say one sentence to your team and to yourself: the board is the plan now, and the chat is the archive. Next week, ask the assistant "what changed on the Studio Relaunch board in the last week, and who changed it?" and it answers from the board.

The Five Destinations Ranked for Turning a ChatGPT Conversation Into a Plan

Each review has the same shape: verdict, who it suits, price, why it ranks where it does, and what it cannot do.

1. Storyflow

Verdict: The best destination when the plan has to keep the decisions, the tasks, the dates and the open questions visible together, and when the assistant should build it for you.

Best for: Founders, agency leads, producers and marketers who think a project through in ChatGPT or Claude and then have to hand it to two to ten people.

Price: Plus $7.99 a month billed annually ($9.99 monthly), Pro $14 ($19 monthly), Max $39 ($49 monthly), as of October 2026. No per-seat pricing. Paid-only during early access; a Free plan launches before the end of 2026, and anyone a paid member invites to a board joins free now. See pricing.

Storyflow is a visual project planning workspace: an infinite canvas board that holds notes, tables, documents, images, mind maps, kanban, charts and slide decks. Its MCP server, launched in October 2026, gives an assistant tools that match the five lists closely. build_section creates a wall and every card in it in one call, which is how the decisions wall and each kanban column get built. Tickets carry a title, a description, a date and one label. Tables hold the schedule and the open questions. create_deck builds a slide deck on the same board for the kickoff meeting, and create_mindmap turns the scope into a map. On the read side, get_board_summary describes the board and get_board_changes reports what changed and who changed it over the last day, week or month.

Why it ranks first: The five lists are different shapes, and Storyflow is the one destination here where each gets its own shape on a single surface. Decisions sit as notes beside the tasks they justify. The schedule is a table. The open questions are a table with a default column. A newcomer sees the whole project at one zoom level. And because the assistant can read the board back, the conversation continues against the plan instead of against a memory of it.

Strengths:

• The assistant builds placed, structured elements, not a pasted block of text.
• Reads changes and who made them, in fixed windows of a day, a week or a month.
• View-only links open with no account; invited collaborators edit in real time on any paid plan.
• No MCP limit on Plus, Pro or Max, and no Storyflow AI credits spent, because the work runs on the assistant's model.

Limitations:

• No task dependencies: moving one date does not move the tasks after it.
• A person cell in a table points to a real member of the board. The assistant can fill one only with the name of someone who has already joined the project; any other name is kept as plain text. For people who are not on the board yet, owners go in as a text column or as the ticket's label.
• No calendar tool, and the canvas does not reorder a table for you: the assistant writes rows in date order and you keep them that way.
• Browser only, paid-only during early access. In ChatGPT it is a custom app behind Developer mode, not a one-click install.

2. Asana

Verdict: The best destination when the conversation produced a long list of tasks that depend on each other and belong to different people.

Best for: Teams already running work in Asana, and any plan where follow-through matters more than context.

Price: Personal is free for up to 2 users; Starter is $10.99 per user a month billed annually ($13.49 monthly); Advanced is $24.99 ($30.49 monthly), per Asana's pricing page on 10 October 2026.

Asana's MCP server runs at https://mcp.asana.com/v2/mcp over Streamable HTTP and signs you in with your Asana account. Asana's developer documentation describes creating and managing tasks and projects through natural language, generating reports and summaries, and names the Asana app in Claude and the Asana app in ChatGPT. Its examples include creating a task in a project and assigning it to you, listing a project's sections and checking a project's status.

Why it ranks second: A task, an assignee and a due date are exactly what Asana stores, so the tasks, owners and dates lists land cleanly. It drops behind Storyflow on this job because two of the five lists have no natural home: decisions and open questions end up in task descriptions or a separate document, away from the work they explain. If you weigh follow-through at 30% instead of 10%, Asana comes first.

Strengths:

• First-party apps in both Claude and ChatGPT, so setup is short.
• Real task follow-through for work that spans a team.
• A free Personal plan for up to 2 users.

Limitations:

• Decisions, references and visuals live outside the task list.
• The docs tell you to check the live tool list, so confirm what your client can set (for example due dates) before relying on it.
• Per-user pricing once a third person joins.
• The older endpoint at mcp.asana.com/sse is deprecated; the docs give a shutdown date of 05/11/2026.

The workspace comparison is in Storyflow vs Asana.

3. Notion

Verdict: The best destination when the plan is mostly writing and your team already reads Notion every day.

Best for: Teams with an existing Notion workspace, a tasks database and a habit of documenting decisions in pages.

Price: Free; Plus is $10 per member a month billed annually ($12 monthly) and Business is $20 ($24 monthly), per Notion's pricing page.

Notion's server is at https://mcp.notion.com/mcp over Streamable HTTP, with OAuth sign-in. Notion's get-started guide says a connected client can read and update content you can access in the selected workspace, and names tools for creating and updating pages. Notion's help page lists Claude, Cursor and ChatGPT Pro among the apps that connect, and does not say which Notion plans include it, so check your own workspace.

Why it ranks third: A decision log is a document, and Notion is very good at documents. A page titled "Relaunch plan" with decisions on top and a linked task database below is a respectable plan. It ranks below Asana and Storyflow here for one reason: the task half is only as structured as the database you already built. If you have none, the assistant creates a page of bullets, which is a nicer transcript, not a plan.

Strengths:

• Writes into the pages your team already opens.
• Strong for the decisions list and its reasoning.
• A free plan to start on.

Limitations:

• Task structure depends on your existing databases.
• A page reads top to bottom, so the schedule, the tasks and the questions are never in view together.
• Notion's help page names ChatGPT Pro; check what your own ChatGPT plan allows.

See Storyflow vs Notion for the workspace comparison.

4. Trello

Verdict: The simplest and safest destination for a small plan that is honestly just a list of cards.

Best for: Freelancers and teams of two or three with fewer than 30 tasks and no need for a decision log.

Price: Free for up to 10 collaborators and 10 boards per Workspace; Standard is $5 per user a month billed annually ($6 monthly); Premium is $10 ($12.50 monthly), per Trello's pricing page.

Trello's MCP server is at https://mcp.trello.com/v1. Atlassian's support article names ChatGPT, Claude, Cursor, Visual Studio Code and Gemini CLI, and says the server works on any Trello plan. You approve Read, Write and Search separately. The assistant can create boards and create, update, move and archive cards, attach labels, assign card members and read activity history. The article states that destructive delete operations are not supported.

Why it ranks fourth: For the tasks list alone, Trello is fine, and its separate Read, Write and Search approvals are the most careful permission design in this group. It ranks fourth because a card holds little, and a plan made of cards has no place for a schedule with milestones, or for decisions with their history. You end up with a "Decisions" list of cards nobody opens.

Strengths:

• Works on the free plan.
• No permanent deletes through the assistant, by design.
• A board everyone already understands.

Limitations:

• Atlassian's article lists viewing and moving lists, not creating them, so set up your columns first.
• Shallow structure for anything multi-phase.
• Custom fields need Standard or above.

See Storyflow vs Trello.

5. Google Docs

Verdict: The right destination when what you need is a written summary to send, not a plan to run.

Best for: A one-page recap for a client or a manager, or a project small enough that a list of bullets is the plan.

Price: Google Docs comes with a personal Google account at no charge. For business accounts, see Google's Workspace pricing page.

In Claude, the Google Workspace connectors are available to all users (on Team and Enterprise plans an owner enables them first). Per that article, Claude can create new Google Docs, Sheets and Slides and can edit a Doc live in a beta pane beside the chat. It reads text and cannot process images embedded in documents. We could not confirm on OpenAI's help pages what ChatGPT's Google Drive app can write, so from ChatGPT we treat this route as copy and paste.

Why it ranks fifth: Everyone can open a Doc, which is why it scores 9 on readability, the highest in the table. But a document has no tasks, no owners you can filter by, and no dates that mean anything to software. The five lists become five headings. That is a clear summary and a weak plan: nothing in it changes state when work gets done.

Strengths:

• Universal: nobody needs a new tool.
• Good for the decisions list, which is prose anyway.
• No cost on a personal account.

Limitations:

• No task tracking of any kind.
• Goes stale the day work starts.
• Claude's live editing is a beta, and its article tells you to check the result when Claude edits a busy file.

An Illustrative Example: 112 Messages Into One Studio Relaunch Board

This is an illustrative example. The studio, the people and every count below are invented to show the method at a realistic size; none of it is a measurement.

The conversation

Dana runs a three-person design studio. Over nine days, across several sittings, she talks a relaunch through with ChatGPT: a new website, new service packages and a relaunch date. The thread reaches 112 messages. Along the way the number of packages changes twice, a podcast idea appears and dies, and the launch date moves from "mid November" to Tuesday 1 December 2026. Idris (design) and Mei (a freelance developer) have seen none of it.

What the decisions pass returned

Dana asks for decisions with what each replaced. The assistant returns 14. On checking, she corrects two: one was the assistant's own suggestion she never accepted, and one superseded idea was listed as current.

Seven of the fourteen are shown. Recording decision 7, the thing they are not doing, matters as much as the rest: it stops the podcast from coming back in week three.

The tasks, owners and dates table

From the 13 surviving decisions the assistant lists 23 tasks. Nine have an owner named in the chat; 14 come back "unassigned". Six dates are stated or derived; the rest are missing. An extract:

Two orphan tasks surface ("draft podcast trailer script", "research microphones"). Both serve the dead podcast idea. Dana deletes them.

The schedule, worked back from 1 December

The backward pass exposes one conflict the chat never noticed: Dana had said the photos could be reshot "in late November", which is after design sign-off. The reshoot moves to the week of 2 November and stays flagged as needing an owner.

The open questions

Nine questions come back. Three block tasks:

The board

Dana connects Storyflow to her assistant, picks only the Studio Relaunch project and allows changes. She asks for four sections plus a short deck. The assistant builds a Decisions wall with 13 notes, a kanban of four column walls holding 21 tickets (each with its date and an owner label, "Unassigned" where needed), a Schedule table with six rows in date order, an Open Questions table with nine rows, and a six-slide kickoff deck. She watches it land in a second tab and fixes two ticket titles by hand.

What happened next

Dana sends the view-only link to Idris and Mei, then invites both to the board so they can edit. In a 20-minute call they assign the 12 "Unassigned" tickets by changing labels. A week later Dana asks the assistant what changed on the board in the last week and who changed it, and gets a list: tickets Idris moved to Done, a date Mei pushed by two days, a new blocker ticket. The conversation that started the project is never opened again.

What it cost

Dana pays for Storyflow Plus at $7.99 a month billed annually. Idris and Mei joined by invite, so they pay nothing. The assistant work ran on Dana's own ChatGPT plan and spent no Storyflow AI credits.

Prompts You Can Copy for Each Stage

Paste these in order, in the same conversation you want to convert. Replace the project and board names with yours.

Why the prompts are separate

One giant prompt ("turn this chat into a plan") gets one giant answer you cannot check. Seven small prompts give seven short lists, each of which you can correct before it contaminates the next. You spend a little longer prompting and far less time untangling a plan built on a wrong decision.

If the conversation is very long

Assistants tend to be least reliable on the oldest parts of a long thread. Two defences work. First, the evidence quotes: a decision the assistant cannot quote is a decision to distrust. Second, split the job: ask for decisions from the first half of the conversation, then the second half, then ask which second-half decisions overturn first-half ones.

If the conversation spans several chats

Run the decisions prompt in each chat, paste the lists into one new conversation, and ask the assistant to merge them and flag contradictions by date. The newest position wins unless you say otherwise. Then continue from the tasks prompt.

ChatGPT or Claude: Which Assistant for the Conversion

The method is identical in both. The setup and one practical detail differ.

If the conversation happened in ChatGPT, convert it in ChatGPT

The assistant that had the conversation already holds it. Run the collapse there. To build on Storyflow from ChatGPT you need ChatGPT Plus or above with Developer mode, and you add Storyflow as a custom app on the web.

If you cannot use Developer mode, move the lists to Claude

The five confirmed lists are small. Copy them into Claude, where the custom connector setup is shorter and Claude Free allows one custom connector, and ask Claude to build the board. You are moving the conclusions, not the transcript, so nothing is lost. Claude's pricing page lists Pro at $17 a month billed annually ($20 monthly) if you need more usage.

If your team is split between assistants, the board is common ground

One Storyflow address works for ChatGPT, Claude, Gemini and Grok. A teammate on a different assistant connects to the same board and asks it the same weekly question. We compare the two main assistants for this kind of work in Claude vs ChatGPT for Creative Project Planning.

What It Costs: Solo and a Team of Three

Prices are from each vendor's page on 10 October 2026, billed annually. The assistant subscription is extra in every row and the same in every row, so it is left out. For ChatGPT plan prices, see OpenAI's pricing page.

How to read the table

Storyflow is not the lowest number here. Three of the four alternatives cost a solo user nothing. What Storyflow's flat price buys is that the second and third person do not add to it, and that an invited free account can also connect its own assistant, with 25 actions a week (every read and change counts as one, resetting Monday UTC).

The cost that is not in the table

The expensive part of a bad conversion is not a subscription. It is the week a team spends building from a superseded decision because the plan was a pasted summary nobody checked. Twenty minutes on Step 3 costs less than any subscription in the table.

Decision Rules: Which Destination for Which Conversation

Match the destination to what the conversation produced, not to the tool you opened most recently.

The chat produced decisions, tasks, dates and questions in roughly equal measure

Use Storyflow. This is the common case for a launch, a campaign, a rebrand or a shoot, and it is the case where the lists need to sit together. If the project started as a diagram instead of a chat, How to Turn a Mind Map Into a Project Plan covers the same conversion from the other side.

The chat produced 40 or more tasks with a clear order

Use Asana. Put the decisions in the project description or a linked document and let the tool do what it is good at.

The chat was mostly reasoning, and the output is a position

Use Notion or Google Docs. Some conversations do not produce a project. They produce a strategy, a policy or a brief. Write it down as prose and stop.

The chat produced fewer than 15 tasks for one or two people

Use Trello, or a to-do list. Do not build a board for a weekend's work.

You are not sure the project will happen

Build nothing. Ask for the decisions and open questions lists only, save them, and convert the rest when the project is real.

Common Mistakes When Turning a Chat Into a Plan

Mistake 1: Asking for "a plan" in one prompt

The assistant will produce one, confidently, from all 112 messages at once, including the ideas you dropped. Ask for the lists first.

Mistake 2: Skipping the check on decisions

Every later list inherits the errors in this one. If you check only one thing, check this.

Mistake 3: Letting the assistant fill the gaps

Blank owners and missing dates are information. A plan where the assistant assigned everyone and dated everything looks finished and is wrong in ways you will find late.

Mistake 4: Exporting the transcript

Pasting the whole conversation into a document or attaching it to a task moves the problem without solving it. Nobody reads it. Move conclusions.

Mistake 5: Keeping the chat as the place where the plan changes

After the conversion, changes happen on the board. If you keep revising in the chat, you now have two plans, and the team is reading the stale one.

Mistake 6: Leaving out the non-goals

What you decided not to do is part of the plan. Without it written down, dead ideas come back, usually from the person who was not in the conversation.

Mistake 7: Giving the assistant every project

When you connect an assistant to any tool, scope it to the project at hand. In Storyflow that is a choice at sign-in. The reasoning is in Are AI Connectors Safe? A Permissions Guide.

Mistake 8: Attaching images in the chat and expecting them on the board

A picture you attach in ChatGPT or Claude does not reach Storyflow. The assistant gives you an upload link for the board, or adds pictures from web addresses. The connector also cannot generate images or read comment threads.

When Storyflow Is Not the Right Choice

Storyflow ranks first on the weighting above. It is the wrong choice in these five situations.

When moving one date must move twenty others

Storyflow has no dependency engine. Dates live on tickets and in table cells, and each one is independent. A build schedule where every slip cascades belongs in Asana or another tool with dependencies.

When tasks must notify and chase their owners

For anyone who has not joined the board, the assistant writes the owner as a ticket label or a text column, because a real person cell needs a member of the project. If your team relies on a tool assigning work to named accounts and chasing them, start in Asana.

When you need a free tool today

Storyflow is paid-only during early access. Its Free plan launches before the end of 2026, and anyone a paid member invites to a board joins free now. If nobody on the project can pay, Trello's free plan works with its MCP server, and Notion has a free plan.

When you work on a phone or without a connection

Storyflow runs in the browser. There is no mobile or desktop app and it does not work offline.

When the plan needs a calendar, approvals or billing

Storyflow has no calendar tool, does not run formal approvals with an approve button, and does not invoice or handle contracts. Pair it with tools that do, or pick one that includes them.

Template: The Conversation-to-Plan Board

Copy this structure whatever the project. It is the board the build prompt above produces.

The board layout

The ticket fields

Each ticket carries four things: a title that starts with a verb, a one-line description of what done looks like, a date in the ticket's date field when one was stated or derived, and one label. Use the label for the owner on small teams ("Dana", "Unassigned") or for the phase on larger ones ("Copy", "Design", "Build").

The Waiting column earns its place

Most project boards have three columns. Add the fourth. Tasks blocked by an open question go in Waiting with the question in the description, which makes the cost of every unanswered question visible on the board.

Starting points if you would rather not begin from a chat

If the source is a document, turn a document into a project board. If you have only a goal and a deadline, the AI project plan generator and the AI kanban board generator start from a sentence. For a site relaunch like the example, the website planner has the sections laid out. The wider approach is described on the visual project planning page.

The Bottom Line

The conversation was the thinking. The plan is what survives it: five short lists, checked by you, placed somewhere other people can read and the assistant can come back to. Do the collapse in the chat, in order, decisions first. Then stop revising in the chat.

For most projects that start this way (a launch, a campaign, a relaunch, a shoot), build the plan on a Storyflow board, because it is the one destination here where the decisions, the task wall, the schedule and the open questions sit on a single surface that ChatGPT or Claude both builds and reads. If your plan is a long chain of dependent tasks, send it to Asana. If it is an argument, write it in Notion or a Doc.

Take the longest project conversation you have open right now and run the decisions prompt on it. If the list surprises you, you have your answer about whether the chat was ever a plan. Then connect your assistant to Storyflow and build the board.