How to Create a Business Plan Using ChatGPT
Ask ChatGPT to write you a business plan and you’ll get one in ninety seconds. It will be well structured, confidently written, and largely fiction — because the market size, the competitor analysis, the growth rates and the industry benchmarks will all be invented, and it will present them with no indication that it made them up.
Used the other way round, though, it’s genuinely one of the best planning tools available. Not writing the plan for you, but interrogating you about your business until the gaps in your own thinking become obvious.
That’s the method here: AI as interviewer and editor, you as the only source of facts. It produces something considerably more useful than a generated document, and it won’t embarrass you in front of a bank.
Step 1: Decide what the plan is actually for
Three different documents share one name
This determines everything about length, tone and content, and getting it wrong means writing the wrong document well.
Which plan are you writing?
| For yourself | The most valuable and least formal. Purpose is to find out whether the numbers work before you commit money. Can be five pages of honest arithmetic. No polish required. |
| For a lender | They care about one thing: can you repay. Cash flow forecast, security, your own stake, realistic assumptions, and what happens if trade is 30% below forecast. Format matters; ask them for their template. |
| For an investor | They care about scale and exit. Market size, growth mechanism, why you, what the money buys. A different document entirely, and much harder to write honestly. |
If you’re not sure, write the first one. It’s the one that tells you whether to proceed, and both others are easier to produce once it exists.
Step 2: Gather your own numbers before you open anything
Every figure you don’t supply will be invented
Collect these first, from real quotes and real research rather than estimates: your setup costs itemised, your fixed monthly costs, your variable cost per unit or job, your intended price, how many you can physically deliver per week, and how much cash you have to start with.
Then the honest ones people skip: how long you can survive with no income, what you need to draw personally each month to live, and how long it realistically takes to get your first ten customers.
If you can’t fill these in, that’s the real finding — and it’s more useful than a plan built on placeholders.
Step 3: Get interviewed rather than written for
The technique that makes this worth doing
This is the centrepiece. Instead of asking for a plan, ask to be questioned — the gaps it finds are the gaps a lender will find, and better to meet them now.
Do not write the plan. Interview me.
Ask me questions one section at a time, starting with the business model. After each answer, do three things:
1. Tell me if my answer was vague, and ask again more specifically
2. Point out anything my answer implies that I might not have noticed
3. Only move on when the answer would satisfy a sceptical reader
Cover in order: what exactly I’m selling and to whom, how I’ll reach customers, my costs, my pricing and margin, my capacity, competition, what could go wrong, and what I need to be true for this to work.
Rules:
– One question at a time. Wait for my answer.
– Do not supply any market data, statistics or industry figures — I’ll research those myself
– If I give you a number, ask where it came from
– Be sceptical rather than encouraging. I want the holes found now.
Work through this properly. It takes an hour or two and it’s where the actual planning happens — most people discover two or three things they hadn’t thought about, which is the entire point.
Step 4: Build the financials with your assumptions, its structure
AI is good at the model and bad at the inputs
AI can build you a sound financial structure — the categories, the arithmetic, the format a lender expects. What it cannot do is know your costs, and it will happily fill them with plausible numbers if you let it.
SETUP COSTS: [itemised]
FIXED MONTHLY COSTS: [rent, insurance, software, phone, vehicle, etc — itemised]
VARIABLE COST PER [unit/job]: [amount]
MY PRICE: [amount]
REALISTIC CAPACITY: [how many per week I can actually deliver]
STARTING CASH: [amount]
WHAT I NEED TO DRAW MONTHLY: [amount]
Produce:
1. A month-by-month cash flow for 12 months, showing money in, money out and closing balance
2. My break-even point — how many sales per month to cover costs including my drawings
3. The month my cash position is lowest, and how low
4. What happens if sales are 30% below what I’ve assumed
5. What happens if I get paid 30 days later than I’ve assumed
Then list every assumption you had to make, and flag any figure I gave you that looks optimistic for this kind of business.
Points 4 and 5 are the ones lenders actually probe. A plan that has already answered them reads as competent; one that hasn’t reads as hopeful.
Do this instead. Use AI to work out what you need to find, then find it yourself: Perplexity for sourced answers you can verify, ONS and government data for UK statistics, Companies House for competitor filings, and trade bodies for sector figures. Anything you can’t source, either cut or state plainly as your own estimate with your reasoning shown.
Every number in a lender-facing plan should have an answer to “where did this come from?” — and “ChatGPT said so” ends the conversation badly.
Step 5: Research the market properly
Use AI to design the research, not to be it
Do NOT give me any figures or statistics. Instead tell me:
1. The five things I genuinely need to know about this market for the plan to be credible
2. For each, exactly where I could find that information — name specific sources: government datasets, trade bodies, public filings, published surveys
3. What I could establish myself through direct research — how many competitors within a radius, what they charge, how busy they look
4. Which questions can only be answered by talking to potential customers, and what I should ask them
5. What claims about this market a sceptical lender would immediately challenge
Point 3 matters more than people expect for local businesses. Counting competitors, ringing them for prices and observing how busy they are produces better evidence than any downloadable report, and it’s free.
Step 6: Write the sections — with you supplying the substance
Now the drafting is safe, because the facts exist
With the interview done, the numbers modelled and the research sourced, drafting becomes straightforward.
Here are my verified figures with sources: [paste them]
Here is my cash flow model: [paste the key outputs]
Draft the [section name] section of my business plan.
Rules:
– Use only the information above. Do not add any fact, figure or claim I haven’t supplied.
– Where something is my estimate rather than a sourced fact, say so in the text
– Plain British English. No “poised to disrupt”, no “rapidly growing market”, no “unique opportunity”
– Concrete rather than aspirational: what I will do, when, and what it costs
– [Length appropriate to your reader]
Afterwards, list anything you’d normally expect in this section that I haven’t given you.
That closing instruction is your gap-finder. Run it on every section and you’ll build a list of what still needs researching, rather than discovering it when someone else reads the plan.
Step 7: Attack your own plan before anyone else does
The step that saves you the embarrassment
Read it as a sceptical bank lending officer who has seen hundreds of these and rejects most.
Tell me:
1. The three weakest claims, and why you don’t believe them
2. Every number that has no stated source
3. Where my assumptions are optimistic, and what the realistic version looks like
4. What I’ve left out that you’d need before lending
5. The single question you’d ask that I’d struggle to answer
6. If you were lending your own money, would you? Give me a straight answer and your reasoning.
Do not soften this. Do not compliment the plan. I want it to fail here rather than in the meeting.
Then fix what it finds. And if the honest answer to point 6 is no, that’s worth knowing before you’ve signed a lease — this is the cheapest business failure you’ll ever have.
Before showing the plan to anyone
- Every figure has a source you could state out loud.
- No AI-supplied market statistics remain in the document.
- Your own estimates are labelled as estimates.
- The cash flow shows your lowest month, not just the annual total.
- A 30% shortfall scenario is included.
- Your personal drawings are in the costs — plans that omit them aren’t plans.
- Prices tested against what competitors actually charge, verified by you.
- Someone who’ll be blunt has read it — ideally an accountant.
Frequently asked questions
Can a bank tell if my plan was written with AI?
They increasingly can, and more to the point they don’t need to — what they notice is the pattern that gives it away. AI-written plans share recognisable features: unsourced market statistics, uniformly smooth prose, generic competitive analysis that would fit any business in the sector, and financials that are internally tidy but disconnected from the narrative. A lending officer who has read hundreds spots that quickly, and the problem isn’t the AI, it’s that the plan doesn’t demonstrate you understand your own business. What they’re actually assessing is whether you know your numbers and have thought about what could go wrong. So use AI for structure, drafting and interrogation, then make sure you can defend every figure in conversation without the document in front of you. If you can answer “where did this number come from?” for everything in the plan, how it was typed is irrelevant.
How long should a business plan be?
Shorter than most people write, and it depends entirely on the reader. For your own use, five to ten pages of honest arithmetic and clear thinking is genuinely enough — the value is in the working, not the wordcount. For a lender, ask them directly what they want, because many have their own template and a Start Up Loans application or bank format will specify exactly what to include; supplying a beautifully written twenty-page document when they wanted their four-page form is a waste of your week. For an investor, ten to fifteen pages plus a financial appendix is typical, though many prefer a short deck first. What’s consistent across all three: a strong plan is dense with specifics rather than long on description. If a section could be cut without losing a fact, cut it. Padding reads as uncertainty, and lenders in particular are looking for evidence you’ve thought concretely.
What’s the most common mistake in an AI-assisted plan?
Optimistic capacity, closely followed by omitted drawings. Capacity first: people model revenue on working every available hour at full utilisation, which no business achieves — you’ll lose time to quoting, travel, admin, chasing payment, illness and gaps between jobs, and a realistic figure is often half your theoretical maximum. AI won’t correct that because it doesn’t know your trade, so it takes your number and multiplies. Second, plans routinely omit what the owner needs to live on, which produces a business that looks profitable and can’t pay you. Put your personal drawings in the fixed costs and see whether it still works. Third, and specific to AI: growth curves that assume steady month-on-month increases from launch. Real small businesses have a slow, lumpy start, and a plan showing smooth exponential growth from month one signals inexperience more clearly than almost anything else.
Should I use ChatGPT or Claude for this?
Either works, with a mild preference by task. For the interview in Step 3 and the pressure test in Step 7, both are strong, though you may find one pushes back more willingly than the other — and if either starts agreeing with everything, tell it explicitly to be sceptical, because agreeableness is the failure mode that ruins this exercise. For the long drafting in Step 6, Claude tends to hold tone and structure better across several thousand words. For the financial modelling, either will handle the arithmetic, but check it yourself: language models make occasional calculation errors and a cash flow with a wrong subtotal undermines the whole document. If the numbers matter — and they do — build the final version in a spreadsheet where you can see the formulas, and use AI for the structure and the sanity check rather than as the calculator.
The bottom line
Don’t ask AI to write your business plan — ask it to interview you, then draft from what you supply. The gaps it finds in an hour are the gaps a lender would find in a meeting, and finding them now is the whole value of the exercise. Gather your real costs and honest capacity before you start, never let it supply market statistics, and put a source against every number in the finished document. Model the bad month as well as the good year, include what you need to live on, and finish by asking it to reject the plan as a sceptical lender would. If it can’t be defended there, it can’t be defended anywhere.
Nothing here is financial, legal, tax or investment advice. AI-generated market data, statistics and financial figures are frequently inaccurate or fabricated — verify every figure against a citable source before including it in any document shown to a lender or investor. Have financial projections reviewed by a qualified accountant before relying on them.