Bottom-up, not top-down
Ask two models to size the market independently from your assumptions. When their numbers differ by ten times, you have found the assumption to check with real data before anyone else does.
Most business plans are written once, for a bank or a form, and never opened again. The useful kind is short, honest about what you do not know yet, and updated as you learn. AI makes that kind cheap to write and cheap to keep current. This page shows what goes in it, how several AIs make each section sharper than one would, and how to keep the plan alive in a Project instead of in a forgotten PDF.
Free to start. The latest models matter most for the financial and market sections.
Write them in this order. Each is one conversation in the Project. If a section takes more than a page, it is hiding a decision you have not made.
| Section | The question it answers | Done looks like |
|---|---|---|
| Summary | What is this, for whom, and why now? | Three sentences a stranger understands. |
| Problem | What do customers do today and what does it cost them? | Evidence from real conversations, not assumptions. |
| Customer | Who exactly buys, and who decides? | A named type of person with a job and a budget. |
| Offer and business model | What do they get, and how do you get paid? | One offer, one price, one reason it is worth it. |
| Market | How many of these customers exist and how many can you reach? | A bottom-up number with each assumption on its own line. |
| Competition | What do they use instead, and why would they switch? | A table with real prices and real complaints. |
| Product and delivery | What will you build or deliver first? | A version-one scope you can finish in weeks. |
| Marketing and sales | Where do the first fifty customers come from? | Two channels you can name, and the first message. |
| Financials | What does it cost, what comes in, when does it break even? | A twelve-month spreadsheet at three prices. |
| Risks and milestones | What would make you stop, and what are the next three milestones? | A stop rule and dated milestones. |
Any AI can draft a business plan. The value is in the sections where a single confident answer is most likely to be wrong.
Ask two models to size the market independently from your assumptions. When their numbers differ by ten times, you have found the assumption to check with real data before anyone else does.
One model builds the twelve-month sheet; another lists what is missing. Payment fees, taxes, refunds, the month you will need to pay for help. The second pass is where plans become honest.
Upload competitors' pricing pages and their worst reviews to the Project. Ask for the pattern. That pattern is your positioning, and it belongs in the plan in the customers' own words.
A plan in a Project can be asked questions: "Given this plan, what should I do this week?" A plan in a PDF cannot. Update the sections as you learn, and every later conversation starts from the current version.
If a bank or grant body needs a plan, they usually have a template. Paste it into the Project and ask for the plan in that structure, drawn from the sections above. Do not let the template make you invent numbers; write "not yet known, will be tested by [date]" where that is the truth. Reviewers trust a plan that admits gaps more than one that fills them with guesses.
For a partner or first hire, the summary, the customer, the offer and the milestones are enough. Generate a one-page version and a short deck from the Project; they will read the same facts as the long plan because they come from the same place.
Drafting a summary is easy for any model. Sizing a market bottom-up from real assumptions, finding the cost lines a first-time founder forgets, and arguing whether a price is too low: these are reasoning tasks, and the latest models do them noticeably better than older or smaller ones. If the plan is going to a bank or into your own savings, that is where to spend.
Newer models hold your whole Project in view instead of the last few messages, reason through trade-offs instead of picking the first plausible answer, and are wrong less often and with less confidence. For the questions on this page, that is the difference between advice that sounds right and advice you can act on.
Pro from $20 a month, cancel any time. The free plan stays free.
Add the idea and anything you already know: customer notes, competitor pages, a rough budget.
It can draft every section, and with your notes in the Project the draft will be specific to you. What it cannot do is know your customer. The plan is only as good as the conversations and evidence you feed it.
Ten sections of a page or less. Longer plans hide decisions. If someone needs a longer document, generate it from the short one.
You need the customer, the offer, the first channel and the stop rule. The rest can be filled in as you learn. Writing those four down is the plan.
Banks care about the numbers and whether you understand them. A plan whose assumptions are visible and whose financials are bottom-up does well. A plan full of invented figures does not, whoever wrote it.
A template gives you headings. This gives you the questions, the evidence standard for each section, and several AIs checking the answers. Then it keeps the plan alive in a Project.
Your first Project, file uploads and the major model families. The latest models, generated spreadsheets and documents, and sourced research are on paid plans. See plans.
Customer, offer, first channel, stop rule. Put them in a Project and let several models build the plan around them. Free to start.
Also: validate the idea · start a business with AI · create a marketing plan
Continue learning
Run the same prompt through ChatGPT, Claude, Gemini and Grok before trusting one answer.
Core featureLet several models draft, challenge and verify each other instead of trusting one answer.
ProjectsKeep files, instructions, code and chats together so every model works from the same context.
FeaturesModels, Projects, collaboration, verification, Studios and images in one workspace.
HubIs the idea good, who else does it, what must it do, what will it cost: answered, then built in a Project.
ValidationA five-rung ladder with pass marks, from scoring to pre-selling.
Market researchSourced market size, segments, pricing and competitors with Perplexity Sonar Pro.
Online businessA real problem, a small product or service, a demand test, honest costs.
IdeasSixteen narrow, startable ideas with who pays and what you build.
StartupsTwelve vertical ideas with a wedge and a why-now.
FoundersOne Project for research, code, decisions and launch work.
LaunchThe four weeks before, the day, the week after.