ChatGPT Translator Contract: Per-Word Rates + Memory

See the canonical entry point at Using AI to Generate Professional Freelance Documents: The Complete Guide (Contracts, Proposals & Invoices).

You have a job booked and the client wants terms, so you ask ChatGPT for a translation contract. Ten seconds later you have a clean document with scope, a rate, a deadline, and an IP line. It reads like a real agreement.

The trouble is that a generic model does not know how translation work is actually priced and owned. It writes the rate as a vague per-word figure, it says nothing about the translation memory you build along the way, and it hands the finished work to the client on delivery. Those gaps are how a job quietly loses money on repeated text or surrenders an asset you could have reused for years. This post gives you a translator-specific prompt, then the three clauses you rewrite before sending.

The prompt that drafts a translator's contract

Paste this into ChatGPT, Claude, or Gemini, using placeholders for the real client details.

You are an expert at drafting freelance translation contracts. Draft a
contract from the details below.

PROJECT: [language pair, document type, word count]
RATE: [e.g. "$0.12 per SOURCE word"; minimum charge; rush fee]
CAT DISCOUNTS: [your match-tier schedule, or "none"]
DEADLINE + DELIVERY FORMAT: [...]

Rules:
1. Pricing: state the per-word basis (SOURCE or target), the rate, a
   minimum charge for small jobs, and the CAT-tool match tiers (100%
   match, fuzzy bands, no-match) if discounts apply. No undefined
   "per word" rate.
2. Translation memory: the Translator RETAINS ownership of the
   translation memory, termbase, and glossary created during the job,
   unless separately purchased.
3. IP: copyright in the translation transfers to the Client on receipt
   of FULL payment, not on delivery. Until then the work remains the
   Translator's.
4. Plain English. Flag any clause that depends on my jurisdiction.

Output the contract, then list the decisions you made that I should
confirm.

The rules are doing the work. Strip them out and the model writes an undefined rate, ignores the memory entirely, and signs the work over on delivery. Even with the rules, confirm the three clauses below, because they are where a translation job protects its margin and its assets.

Clause 1: define the per-word basis and the match tiers

The pricing clause is where a translation job leaks margin, and generic AI writes the leakiest version: a single per-word number with no basis attached. The first thing to pin down is whether you count words in the source or the target language. Source-word pricing is the cleaner default, because the original is fixed and countable before you start. As Tomedes puts it:

When you intend to charge per source word, then you'll be able to give your client an accurate quotation for the translation up-front.

Source: Tomedes

The second piece is the CAT-tool discount schedule. Translation tools grade repeated text into match tiers, a 100% match, high fuzzy matches in the 95% to 99% and 85% to 94% bands, and brand-new no-match words at the full rate, and the contract should say which tiers get which rate if you offer discounts at all. The exact percentages are yours to set; as Ciprian Iovu notes, published pricing grids are illustrative rather than fixed, so the schedule in your contract should be the one you actually use. A minimum charge and a rush surcharge belong in the same clause, and AI routinely skips both. The minimum keeps a fifty-word job worth your setup time; the rush surcharge prices a tight turnaround instead of giving it away free. The benchmark rates to anchor against live in the translation pricing report; here the job is making the prompt define the basis, the rate, a minimum charge, and the tiers instead of one floating number.

Clause 2: keep your translation memory

The clause generic AI omits completely is who owns the translation memory, and that omission favors the client every time. The memory, the termbase, and the glossary you build during a job are reusable assets that make your future work faster, so handing them over without thought gives away an asset you paid to create. The new model contract from the American Translators Association sets the default clearly on the linguist's side:

The translator will retain rights to any termbases, glossaries, translation memories (TMs), or other files created while the translation is produced.

Source: Danielle Maxson, ATA Chronicle

So unless a client is specifically paying you to surrender the memory, the contract should keep it with you. The prompt has to instruct the model to add that clause, because it will not raise the question on its own. The fuller picture, including how agency contracts quietly capture the memory through the CAT tool you bought, lives in the per-word and translation-memory contract guide.

Clause 3: transfer the translation on payment, not delivery

The IP clause is where a linguist loses leverage, and generic AI drafts it backwards. It assigns the finished translation to the client on delivery, which means the work is theirs the moment you send the file, even if the invoice goes unpaid. A translation can carry its own copyright, so the transfer should be conditioned on payment in full. You retain the rights until the client has paid, the client holds a limited license to review the draft in the meantime, and ownership moves only when the money clears. This is the same default trap that shows up in every AI-drafted freelance contract: the model reaches for vague assignment language without pinning down the moment ownership changes hands. Vague contracts are not a small risk: 44% of freelancers have been stiffed by a client, and among them, 37% blame vague or poorly written contracts. A transfer-on-payment clause is exactly the specificity that prevents it.

Here is what to rewrite:

Read it once, because AI invents legal specifics

The rewrites are the predictable gaps. The other rule is to read the whole document, because AI will state a legal specific with total confidence that is simply wrong. In Mata v. Avianca (2023), lawyers were fined $5,000 for filing a brief ChatGPT had stuffed with fabricated cases. You are not in court, but a contract that governs who owns your translation and your memory deserves the same read-through. As Pactly notes, any text generated by an AI must be treated as a draft and requires final legal oversight.

Or start from a contract built for translation work

The prompt works, and the translator-specific version above is far better than a generic request. The catch is the same as every AI draft: you fill in the rules, fix the clauses, and reformat the output for every new client.

If you would rather skip that and start from a contract where the per-word basis, the retained translation memory, and the transfer-on-payment clause are already set, FreelanceDesk builds it with those choices baked in and generates locally in your browser, so client and document detail never leaves your machine. It is free.

To go deeper, the per-word and memory contract guide covers what agency contracts capture and how to counter it, and the translation pricing report gives the rate benchmarks to anchor your numbers. The generic AI-contract prompt covers the base this one builds on, the content-writer version applies the same idea to written work, and the contract essentials post covers the clauses every freelancer needs underneath the translation-specific ones.

References

• ATA's New Model Contract for Translators : ATA Chronicle
• Charge by Source or Target Word Count : Tomedes
• Understanding Translation Pricing Grids : TranslateCluj
• Freelancer Statistics : TeamStage
• Mata v. Avianca, Inc. : Wikipedia
• Can I Use ChatGPT to Write a Legal Contract? : Pactly
• Sensitive Data in ChatGPT : Cyberhaven