Can ChatGPT write a business plan? Yes, but not the numbers: 9 prompts and where to get the data

Which business plan sections can ChatGPT write from your notes?

Five of the eight sections need nothing but what you already know. The SBA's Write your business plan guide lists eight sections for a traditional plan, and it says "Lenders and investors commonly request this plan." Its text promises nine sections. It lists eight. Sorted by what each one needs as input, they split like this:

Any chatbot can draft the five sections in the first row, since every fact in them comes from you. Market analysis is the hard one. It needs counts of customers and competitors that only public data has. The funding request and the projections need your prices, costs and loan terms, and the SBA wants a financial outlook "for the next five years."

Start with an interview, not a request for a finished plan. Paste this into ChatGPT, Claude or Gemini:

"I'm writing a business plan for [business, city] in the SBA's traditional format. Interview me one question at a time about the business, the customers, the team and the money. Don't draft anything until I type DRAFT. Then write one section per reply, in the SBA's order, using only what I told you."

The interview gets your facts into the chat before the model has a gap to fill. That's the whole trick.

What prompts write each section of a business plan?

Use one prompt per SBA section, in this order, and check each reply before you send the next. Write the executive summary last. The SBA says it tells the reader "what your company is and why it will be successful," and that's easier once the other seven exist. Fill the brackets.

Company description: "Write the company description for [business]. Name the problem we solve, the customers we serve ([list]), and the advantages in my notes. Don't invent awards, partners or credentials."

Market analysis: "Draft the market analysis for [business] in [city or county]. Leave every figure as [NUMBER, source needed] and name the public dataset that would supply it. Don't estimate any market size yourself."

Organization and management: "Write the organization and management section. We're a [sole proprietorship, LLC, S corporation]. Here are the people: [name, role, years of experience]. Add a text org chart and one line on how each person's experience helps."

Service or product line: "Describe what we sell: [products and prices]. For each one, say how it helps the customer, where it is in its life cycle, and any copyright, trademark or patent plans from my notes."

Marketing and sales: "Write the marketing and sales section. Our customers are [description]. We reach them through [channels] and close a sale by [process]. Show the steps from first contact to paid order, and list every assumption the financial projections will need."

Funding request: "Write the funding request. We need [amount] as [debt or equity] over [term]. Here is how we'll spend it: [items and amounts]. Check that the items add up to the total and tell me if they don't."

Financial projections: "Build a projection for five years as an Excel file: monthly for year one, then yearly. Put each assumption (price, units sold, growth, wages, rent) in its own input cell and make every total a formula that uses those cells. Use only these inputs: [list]. Mark any input I haven't given you as MISSING."

Executive summary: "Using only the sections in this chat, write a one page executive summary: mission, what we sell, who runs it, where we are, and how much funding we want and why. List any fact you had to assume."

Four of these prompts (company, market, projections, summary) close with a rule against invented facts. Add the same line to the other four. A blank is easy to spot, and an invented figure isn't.

What does ChatGPT get wrong in a business plan?

ChatGPT gets the numbers wrong, not the prose. A market size figure in a chat answer has no source unless the model searched for it. Even then, OpenAI's own search help page warns that "Search results and citations can be incomplete, outdated, or incorrect." A figure you can't trace doesn't belong in a plan a lender reads.

Projection tables break next. A table in a chat reply is plain text, so if you change your price, none of the totals move. OpenAI's data analysis page adds a second limit. The Python sandbox ChatGPT uses for calculations "cannot make external web requests or API calls," so it works only with numbers you upload or type.

Growth assumptions are the third gap. The SBA asks for five years of forecasts, and five years is long enough for many new businesses to close. The Bureau of Labor Statistics survival table tracks 767,573 private sector establishments opened in the year to March 2020. A year later, 80.9% were still open. By March 2025 it was 51.4%. A forecast that grows every year with no downside case leaves out half the story.

Old facts are the last trap. Without web search, a model answers from training data that stops at a cutoff date, so its competitor list can include shops that have closed. Turn search on, ask for a link per competitor, and open each one. Our AI hallucination statistics page collects the measured error rates.

Where do you get real market size and financial numbers?

Real market size and cost figures come free from US federal data, and the SBA's guide links most of these sources itself. Pull the figure, write down its year, then paste it into the chat with its link.

Here's a top down market size built only from BLS figures. The 2024 Consumer Expenditure Surveys count 135.76 million US consumer units (roughly, households), and they spent an average of $3,945 each on food away from home. Multiply the two and US households spent about $535.6 billion on food away from home in 2024. That counts only what households report to the survey, so business meals and tourists sit outside it. A cafe plan would swap in its own county's household count from the Census Bureau.

Then hand the model your figures, not the job of finding them:

"Here are my figures and where each came from: [figure, source, year]. Write the market analysis around them. Quote each figure exactly with its source and year, and don't add any number I didn't give you."

For the projections, ask for a file instead of a chat table. Our guide to cleaning up a spreadsheet with AI covers checking what comes back, and AI for market research goes deeper on competitor work.

Which AI is best for writing a business plan?

Claude is the best free pick for the financial projections, and Gemini for the market research. Anthropic's file creation page says you can "create financial models in Excel with working formulas," and it lists the Free plan. For research reports, Google's plans page lists Deep Research on Gemini's free plan, while ChatGPT's pricing page marks deep research Limited on Free and Claude's pricing page shows Research as No on Free. Each row below comes from the vendor's own pages, read on 12 October 2026.

ChatGPT is the tightest free option for a plan built from your own files. OpenAI's file uploads FAQ says "Free users are limited to 3 file uploads per day," and the ChatGPT pricing page lists data analysis and deep research as Limited on Free. Gemini's export page saves a table to a new Google Sheet, so check whether the totals arrive as formulas or as typed values.

Dedicated generators are the other route. Shopify's July 2026 roundup prices Venturekit from $8 a month, PrometAI from $55 a month and 15 Minute Plan at a one time $69, and the first two keep parts of the financials for paying users. Shopify says these tools "offer a more structured approach to making a business plan than LLMs."

Whizi, which publishes this page, can hand back a downloadable .xlsx from a chat, and a total the model writes as a formula stays a live Excel formula. Web search is on by default in a new chat. It loses on free use: a free account gets 10 messages a day, on Whizi AI, Auto and Claude Haiku 5.5 only, which won't carry an eight section plan. Founders weighing other tools can start at AI for founders.