How to use AI to price freelance projects from your own past work

Ask a chatbot how much to charge for a brand identity and you'll get a range pulled from blog posts and marketplace listings. It doesn't know how long your last three identities took, which clients paid late, or that the quotes you sent above $7,000 went nowhere. Those facts decide your price. The fix is to hand the AI your numbers, or to work them out yourself and use AI only for the wording and the line items.

Why a generic chatbot gets your price wrong

OpenAI's help center says ChatGPT can produce incorrect or misleading answers and may sound confident while being wrong. Anthropic's documentation for Claude gives the remedy that matters here: tell the model to use only the documents you provide, not its general knowledge, and still check anything important. For pricing, the documents are your own past quotes, invoices and time entries. Without them, the model fills the gap with averages from people who aren't you, in markets that aren't yours.

There's a second problem. A chatbot rounds toward the number you seem to want. If you write "the client's budget is $3,000", that figure tends to become the price, even when your history says the job takes 40 hours. Keep the client's budget out of the pricing step, and bring it back only when you decide whether to cut scope.

The four numbers to pull from your history

1. Similar jobs. Find at least 2 past quotes or invoices for the same kind of work. Match on what you delivered (a logo with a brand guide, a five-page website), not on the client's name or industry.
2. The median of what clients paid. Use the median, not the average: one rush job at double your usual price drags an average up and tells you nothing about a normal project. If fewer than two of those jobs were accepted or paid, use what you quoted and remember it's weaker evidence.
3. Your real hourly rate. Divide what each job paid by the hours you tracked on it. Fixed-price work often hides a rate well below the one you think you charge. If you don't track time on fixed-price jobs yet, start with tracking time on fixed-price projects.
4. Where acceptance drops. Pick a round number near your median and count how many quotes above it were accepted. If two of three quotes above $7,000 were turned down, a fourth one at $7,500 needs a reason, like a bigger scope.

Worked example: pricing a brand identity

Notice what the AI did and didn't do in that example. The numbers came from Priya's records and her own judgment. AI is useful for the next step: turning "brand identity, 3 logo concepts, 2 rounds, brand guide, 3 weeks" into clear line items, a scope paragraph and payment terms, with the price she chose.

A prompt that keeps the AI on your numbers

You are helping me write a quote. Use ONLY the numbers below. Do not suggest a different price, do not use market rates, and leave a line's price blank if I didn't give one.

Past similar jobs (what the client paid, hours tracked):
- [Job 1]: $[amount], [hours] h
- [Job 2]: $[amount], [hours] h
- [Job 3]: $[amount], [hours] h

Quotes I sent above $[round number] and whether they were accepted: [list]

The new job: [what I'll deliver, quantities, rounds of changes, timeline]
My price for this job: $[the price I decided]
Deposit: [percent] upfront. Valid for [days] days. Payment terms: [when the balance is due].

Write: a title, a 2-sentence scope summary, line items with quantity and unit, what's not included, and the payment terms. Then list anything a client might ask that the quote doesn't answer yet.

How SoloWOS does the same thing

In Proposals, the Draft with AI panel shows "Your history" next to the box where you describe the job. It updates as you type, without AI and without using your quota. It looks at the quotes you sent and the invoices you sent or got paid in the last 3 years, in your workspace's currency, and finds the ones about the same kind of work. Matching works on the words you used, with English and Spanish synonyms grouped ("logotipo" counts as "logo"), and the client's name never counts as a match. Generic words like "design" or "project" weigh less and are never enough on their own. An invoice made from a quote counts as the same job, not two.

The panel shows up to 5 similar jobs, their range and median (of what was accepted or paid when there are at least two), average tracked hours, your real hourly rate and how quotes above a round number near the median did. It sums that up in one sentence, such as "You charged $6,000 to $9,000 for your 4 most similar past jobs (median $7,500). They took 11 h on average, a real rate of $640/h. 2 of 3 quotes above $10,000 weren't accepted." With fewer than 2 similar jobs it says there isn't enough similar past work yet and suggests nothing.

When you click Draft with AI, your description, that history and a one-word summary of how the selected client pays go to Anthropic (Claude models; under Anthropic's commercial terms that data isn't used to train their models). The draft can only use prices you wrote or amounts from your history. Lines it can't price stay empty and are flagged, and if the client has paid late or missed a meeting before, the draft suggests a 50% deposit with a note that you decide.

When you click Send, a quick check runs first, without AI. It flags a total 40% or more below or above the median of your similar jobs, a missing deposit for a client who pays late, a validity date that's passed or is less than 3 days away, a total of 0, a missing client email and missing payment terms. You can fix it or choose Send anyway. Check with AI is optional and uses one AI draft: it lists up to 5 things a client might ask, like scope, rounds of changes or what's not included, and never gives pricing or legal advice.

You can also ask the SoloWOS AI assistant "How much have I charged for logos?" and it answers from the same history, or tells you there isn't enough data. The same question works from Claude or ChatGPT through the connector.

Mistakes to avoid

• Pricing from a rate survey alone. Surveys are a sanity check, not a price. Use the freelance rate calculator to see the floor your costs set, then price the job from your history.
• Letting a rush job or a friend's discount into the median. Leave out jobs that weren't normal, or keep enough jobs that one outlier doesn't move the middle.
• Ignoring unbilled revisions. If a past job took 12 hours of extra rounds you never charged, its real rate is lower than the invoice says.
• Putting the client's budget in the pricing prompt. Price the work first; decide what to cut afterward.
• Sending the AI's draft without reading it. Read every line item and the terms. The model can misread a quantity or drop a round of changes.

For the full method behind a project price, read how to price a freelance project. If you want the price to track value rather than hours, see value-based pricing. To turn the price into a quote clients accept online, send proposals clients accept online, or start from the copyable prompt in writing a quote with ChatGPT. More in AI for freelancers.