Blog · 6 November 2026 · Guest post
Your AI tools cost money per client — put it in your rate
Freelancers now run a surprising share of client work through AI APIs: drafting, code assistance, research, image generation. The work feels free because there is no invoice per call — but every call bills per token, and the meter runs on your account, not the client's. It is a real cost of doing the job, exactly like stock photos or a software seat, and it belongs in your rate. The freelancers who absorb it silently are subsidising their clients' AI use out of their own income.
Small per job, real per month
Take a freelance developer running a code-assistant workflow per client project: roughly 500,000 input tokens and 200,000 output tokens on gpt-6.1-sol, which OpenAI lists at $2.00 input and $10.00 output per million tokens. That is $1.00 + $2.00 = $3.00 of API spend per project. At twenty projects a month, that is $60 a month — a software subscription you never signed up for. A heavier agentic workflow on the flagship gpt-6-astra ($10.00 / $50.00 per million) burns 2 million input and 500,000 output tokens per project: $20.00 + $25.00 = $45.00 per project. Same client, same deliverable, fifteen times the hidden cost — decided by which model the workflow calls. Run your own volumes through an AI API cost calculator before you assume the number is negligible, and estimate your monthly API spend the way you would any other business expense.
Two honest ways to charge for it
1. Pass it through as a line item. Add an “AI tooling” line to the invoice, priced from the project's actual usage. It is transparent, the client sees exactly what the tooling cost, and heavy-AI clients pay for heavy-AI work. This suits project pricing, where the scope — and the token burn — is known upfront.
2. Bake it into your rate. Take your average monthly API spend, add it to the expenses field of the freelance rate calculator, and let it lift your hourly rate by the right amount. A $60 monthly API bill spread over 120 billable hours is 50 cents an hour — invisible to any client, but it stops the cost leaking out of your pay. This suits retainer and hourly work, where per-project usage is hard to predict.
Track it per client, not per month
A single monthly total tells you nothing about which client is expensive. Log API spend per project for one month — most providers show usage by API key, so one key per big client is enough. You will usually find the same pattern as with admin hours: one or two clients drive most of the cost, and they are rarely the ones paying the most. That is the information that lets you re-price the right relationship instead of raising your rate on everyone. API spend is also a business expense, so it lowers taxable profit — the same reason your rate should already account for quarterly taxes, and the same kind of unbilled time that the admin quarter quietly eats.
Bottom line
AI tooling is a cost of goods now, not a free perk. Measure it per client for one month, then either invoice it as a line item or fold the monthly average into your rate. The amount is usually small — which is exactly why it is easy to ignore, and why it compounds into real money over a year of projects.
Estimate only — not tax advice. Token volumes above are illustrative examples, not measurements of any real workflow. Model prices are OpenAI Standard-tier rates per million tokens, checked 6 November 2026 on the official pricing page; run your own usage through a calculator for your number.