AI prompts for business plans: market research, financials and pitch decks

The best ChatGPT prompts for a business plan do not ask for "a business plan." They work one section at a time, feed the model your real facts, and force every estimate to show its assumptions so you can check them. Below is a section-by-section workflow with copyable prompts, a pitch deck outline, a "red team my plan" prompt, and the rules that keep AI from putting made-up numbers in front of a lender or investor.

Why "write me a business plan" fails

Ask for a full plan in one line and you get a generic document: a confident market size with no source, a competitor list that may include companies that do not exist or have closed, and revenue that grows 40% a month because nothing told the model otherwise.

The model is filling gaps with plausible text, and a business plan is mostly gaps it cannot fill: your costs, your customers, your local market, your margins. So you supply the facts, and the AI supplies structure, wording, and pressure testing.

Three rules for everything below:

• AI cannot know your market data. It has no access to your sales, your town's foot traffic, or last quarter's supplier prices unless you paste them in.
• Verify every number. Any figure the model produces is a draft to check against a real source, not a finding.
• Never present AI estimates as research. If a number came from a chat window, label it as an assumption or replace it with a sourced figure before anyone else reads the plan.

Step 1: build a fact sheet before you prompt

Every prompt in this article starts by pasting the same fact sheet, so the model works from your reality instead of its guesses. Spend 20 minutes on this first. It is the difference between a usable draft and a fictional one.

BUSINESS FACT SHEET
Business name: [name]
What we sell: [product or service, in one sentence]
Who buys it: [specific customer, e.g. "independent cafes in Portland with 1 to 3 locations"]
Where: [city, region or online]
Stage: [idea / pre-launch / operating for X months]
Current numbers (real, not estimated): [revenue, customers, pricing, costs you know]
Team: [who, relevant experience]
Funding goal: [amount and purpose, or "bootstrapped"]
What I do NOT know yet: [list honestly]
Reader of this plan: [bank loan officer / angel investor / myself / grant committee]

The "What I do NOT know yet" line matters. It tells the model where to ask questions or mark placeholders instead of inventing answers. If you are new to giving context like this, the basics are covered in how to write better ChatGPT prompts.

Step 2: prompts for each section of the plan

Work through the sections in order, one chat or one message at a time. Paste the fact sheet at the top of each.

Executive summary (write it last)

Draft the executive summary after the other sections exist, because it summarizes them.

[Paste fact sheet]
[Paste the finished sections below]

You are an experienced small business advisor. Write a one-page executive summary for [reader of this plan].
Cover: what we do, the problem, who the customer is, why now, how we make money, key numbers, and what we are asking for.
Rules: use only facts from the material above. Where a fact is missing, write [MISSING: description] instead of inventing it. Under 350 words. Plain language, no hype words.

Problem and solution

[Paste fact sheet]

Help me write the problem and solution section.
First, ask me up to 5 questions about the customer's problem: how they handle it today, what it costs them, and how often it happens. Wait for my answers.
Then write two short sections: "The problem" (150 words) and "Our solution" (150 words). Describe the problem from the customer's point of view. Do not claim the solution is the only or best option.

Asking the model to question you first is one of the most useful habits in this whole workflow. Your answers become the evidence the section needs.

Market sizing with clear assumptions

This is where AI does the most damage when unsupervised. Do not ask "how big is the market." Ask it to build a bottom-up model you can check.

Before:

What is the market size for a mobile dog grooming business?

After:

[Paste fact sheet]

Help me build a bottom-up market sizing for a mobile dog grooming service in [city].
Do NOT give me a final market size from memory.
Instead:
1. List the variables I need (e.g. number of households in the area, share with dogs, share using paid grooming, visits per year, average price).
2. For each variable, tell me what kind of source would have a reliable figure (census data, industry association, my own pricing research) and leave the value as [TO VERIFY].
3. Show the formula that combines them into a serviceable market and a realistic first-year share.
4. Flag which variables have the most impact on the result.

Why the after works: it turns the model from an oracle into a spreadsheet designer. You get the structure and the list of things to look up, and every number that ends up in the plan comes from a source you can name.

Competitor analysis

[Paste fact sheet]
Here are the competitors I have found myself, with notes from their websites and reviews:
[Competitor 1: what they offer, price, strengths, complaints in reviews]
[Competitor 2: ...]
[Competitor 3: ...]

Organize this into a competitor analysis. For each competitor: positioning, price level, strengths, weaknesses.
Then write a paragraph on where the gap is for us, based only on these notes.
If you think I am missing a type of competitor (e.g. DIY alternatives, doing nothing), name the category but do not name specific companies.

The last line prevents a common failure: invented or outdated company names. You do the competitor discovery; the AI does the organizing.

Go-to-market plan

[Paste fact sheet]

Draft a go-to-market plan for the first 6 months.
Budget available for marketing: [amount]. Time available per week: [hours].
For each channel you suggest, give: why it fits this customer, first 3 actions, cost range as an assumption, and one metric to track.
Limit to 3 channels. Rank them by how quickly they could bring the first 20 customers.

For channel-level copy and campaign prompts, see ChatGPT prompts for marketing.

Financial projections with an assumptions sheet

Never let the model produce a revenue forecast without a separate list of assumptions. The assumptions sheet is the part lenders and investors actually read.

[Paste fact sheet]

Build a 3-year financial projection structure in two parts.

Part 1: ASSUMPTIONS SHEET. A numbered list of every assumption the projection depends on: price per unit, units or customers per month, growth rate, churn, cost of goods, rent, salaries, marketing spend, payment terms. Use my real numbers from the fact sheet where they exist. For anything else, propose a conservative value, mark it [ESTIMATE], and explain in one line what it is based on.

Part 2: PROJECTION. Monthly for year 1, yearly for years 2 and 3: revenue, cost of goods, gross margin, operating expenses, net income, cash balance. Each line must reference the assumption numbers it uses.

Then list the 3 assumptions that change the result most, and show what happens to year 1 cash if each is 25% worse.

Copy the output into a real spreadsheet and rebuild the formulas yourself. Language models make arithmetic mistakes, especially across many rows, so treat the AI's table as a layout, not as math you can trust.

Risks

[Paste fact sheet]

List the 8 most serious risks to this business: market, operational, financial, legal, and personal (e.g. founder dependency).
For each: how likely (low/medium/high, with one sentence of reasoning), impact, an early warning sign, and a mitigation.
Be blunt. Do not soften risks to make the plan look better.

Pitch deck outline prompts

A pitch deck is a different document from the plan: shorter, visual, and built for a 10-minute conversation. Use the finished plan as the input.

[Paste executive summary and key sections]

Create a 12-slide pitch deck outline for [type of investor, e.g. "pre-seed angel investors"].
For each slide: a headline that states the point as a full sentence, 3 bullets max, and a note on what visual would support it (chart, photo, diagram).
Slides: problem, solution, why now, market, product, business model, traction, go-to-market, competition, team, financials, the ask.
If I have no traction yet, suggest what early evidence I could show instead (waitlist, pilot, letters of intent).

Then tighten it slide by slide:

Here is my draft for the [slide name] slide: [paste].
Rewrite the headline so a busy investor gets the point in 5 seconds.
Cut any bullet that does not support the headline.
Tell me what question an investor would most likely ask after seeing this slide.

Before:

Make my traction slide better.

After:

Here is my traction slide: "Growing fast. Lots of interest. Great feedback."
Real data: 140 waitlist signups in 5 weeks, 3 paid pilots at $400/month, 2 pilots renewed.
Rewrite it with a headline that states the strongest real number, and 3 bullets using only this data.

Why the after works: vague slides come from vague inputs. Once the real numbers are in the prompt, the model has something specific to emphasize and nothing to invent.

The "red team my plan" prompt

Once the plan is drafted, ask the AI to attack it. This is where it earns its place, because it has no ego invested in your idea.

[Paste the full plan]

Act as a skeptical [bank loan officer / experienced angel investor] reviewing this plan. You have seen hundreds of plans and most fail.
1. List the 5 weakest points, ordered by how likely they are to get the plan rejected.
2. For each, quote the exact sentence or number that is weak and explain why.
3. Identify every number that looks like an unsupported estimate.
4. List the 10 hardest questions you would ask in a meeting.
Do not suggest fixes yet. Do not compliment the plan.

Run it, fix the issues, and run it again in a fresh chat. To keep the reviewer persona consistent across sessions, set it up once as a system prompt or custom instruction; system prompt examples shows how.

A verification checklist before anyone sees the plan

Go through this list with the final document open:

1. Every market figure has a named source you checked yourself, or is labeled as your assumption.
2. Every competitor named exists, is still operating, and is described accurately.
3. The financial projection has been rebuilt in a spreadsheet with your own formulas.
4. Every [ESTIMATE] and [MISSING] tag has been resolved or deliberately kept as a stated assumption.
5. No sentence claims "studies show" or "experts agree" without a real citation.
6. Growth rates match the go-to-market plan: you can explain where each new customer comes from.

A faster way to build these prompts

The prompts above share the same structure: a role, the task, your context, a format, and constraints. Once you have written a few, you can reuse them for every new section. If you write prompts like this every day, a prompt generator such as Promptay builds that structure from one sentence describing what you need, and you then paste in your fact sheet.

Frequently asked questions

Can ChatGPT write a complete business plan?

It can produce a document that looks like a complete business plan, but the facts in it will only be as good as what you provide. Use it to structure sections, improve wording, and stress test your thinking. Supply your own data, and verify or label every number before sharing the plan.

Is it okay to use AI-generated market research in a business plan?

AI output is not market research. A model can suggest which variables to measure and where reliable data usually lives, but any figure it produces from memory may be outdated or wrong. Replace those figures with sourced data, or clearly label them as your own assumptions.

What is the best prompt for business plan financial projections?

The most reliable prompt asks for two parts: an assumptions sheet listing every input with its source or an [ESTIMATE] tag, and a projection that references those assumptions line by line. Ask for a sensitivity check on the three most important assumptions. Then rebuild the math in a spreadsheet, because language models can make arithmetic errors.

How do I use AI to prepare for investor questions?

Paste your plan or deck and ask the model to act as a skeptical investor, list the weakest points, and generate the hardest questions it would ask. Answer each question in writing, then ask the model to challenge your answers. Repeat in a fresh chat for a second perspective.