Asking ChatGPT for your price: why the answer misleads

Ask a general-purpose AI what you should charge and you get a number within seconds. The problem is not that the number is wrong. The problem is that neither you nor the model can know whether it is right, because the four inputs a price is actually derived from are missing: your costs, your realistic volume, your customers' willingness to pay, and the prices you actually lose deals to.

This is not an argument against language models. I use them daily, and for several steps in pricing they are excellent. But the step where most founders reach for them, "tell me my price", is precisely the step they are worst suited to.

Why the number sounds plausible

A language model predicts the most likely next word. Ask it for a price for a B2B SaaS product and it answers with what appears most often in comparable text. The result is a market average, phrased as a recommendation.

Two things make that average dangerous.

It lands near your expectation. Your prompt almost always contains an anchor: the category, a competitor, sometimes a range you name yourself. The answer picks it up. You read a number that confirms your own guess, and you mistake that confirmation for a check.

It arrives without an uncertainty estimate. A consultant who does not know your numbers says "I cannot answer that yet". A model rarely does. It is as fluent at €49 as at €149, because fluency is a property of the text, not of the evidence.

The four questions it never asks

Next time you try it, watch what the model does not ask you.

What does a customer actually cost you? Without variable cost per customer and fixed costs per month there is no price floor to compute. That floor is not a recommendation, but it tells you where a price becomes a mistake. The break-even calculator does the arithmetic.

How many customers do you actually close per month? The same fixed costs spread over 20 customers instead of 100 produce a completely different floor. Volume is the input founders most often set too optimistically.

What does your customer pay today to solve the problem without you? That is the reference point for the ceiling: a spreadsheet plus hours, a competitor, an agency, or living with the problem. How to establish it without guessing is in how to find willingness to pay, in German.

Which three providers do you lose deals to? Not the list from a market report, but the names that actually come up in sales calls.

Those four answers produce a price corridor. Anything produced without them is a guess with clean formatting.

Two companies, one answer

The core problem shows up in a simple case. Two B2B SaaS providers ask the same AI for a price for a mid-market analytics tool. Both receive a recommendation around €99 per month, because that is a typical value for the category.

The first company has €12 variable cost per customer, €8,000 in fixed costs, and closes 100 customers. Its floor is €92. At €99 it is effectively working without margin, without noticing.

The second has the same variable cost but €3,000 in fixed costs and 250 customers. Its floor is €24. At €99 it gives nothing away, but it also has no idea whether €149 would have held, because nobody tested willingness to pay.

The same recommendation, two completely different mistakes. Both times the AI worked correctly: it reported the category average. A price simply is not a property of the category.

Where AI genuinely helps with pricing

The honest version of this argument is not a rejection of AI but a division of labour. Once the four answers exist, a language model is strong at several steps:

• Quantifying value: pulling the saved hours or avoided costs out of a customer interview.
• Weighing models: holding flat rate, tiers, usage-based and hybrid against each other with the consequences for your case. The overview is in the SaaS pricing models guide.
• Wording: announcing a price increase, justifying a quote, answering discount requests.
• Finding contradictions: checking whether your premium positioning matches a price in the bottom third of the market.

The difference is not the model, it is the order. Questions first, then arithmetic, then wording. Start with the wording and you get a well-written guess.

What this means for you

If you have had an AI confirm your price, you did nothing wrong. You asked a question that cannot be answered without your numbers.

The test is simple. Take the number you were given and answer the four questions above. If your number sits inside the corridor they span, you were lucky and now you know it. If it sits below, you are currently working on your own account. If it sits above, you are losing deals without knowing why.

That order is exactly what the 9-step framework encodes: product, customer, competition, value and costs first, then the corridor, then strategy and model, and only at the end the number. AI is used at every step, but always after you have answered.