# The Death of Per-Word Pricing? How AI Is Changing What Translators Charge For

Source: https://traduq.com/blog/death-of-per-word-pricing · Updated November 7, 2026

## Key takeaways

- A per-word rate was always a stand-in for effort. Once a machine writes the first draft, the word count says little about how much work the text still needs.
- The alternatives in use are hourly or per-project fees, prices set by what's at stake, edit-distance pricing for post-editing, and fixed program fees for large buyers.
- Each has a catch. Hourly needs trust, edit distance can't see the time spent checking a sentence that was already right, and outcome-based pricing is hard to measure.
- Per-word rates will stay in freelance work and small jobs for a long time. The real fight is over post-editing discounts set before anyone has seen the machine output.
- Ask for a sample before you agree to a post-editing rate, quote review and transcreation by the hour, and put a minimum fee in your terms.

Why a per-word rate fits AI-assisted translation badly, which pricing models are showing up instead, what each one gets wrong, and what I'd ask for as a translator.

Count the words, multiply by the rate, send the invoice. That's how most of us have priced translation for as long as anyone remembers, and it worked well enough. The client knew the price before the job started. The translator knew roughly what a day was worth. Anyone could compare two quotes.

Machine translation and AI didn't break that formula, but they bent it. When a machine produces a full draft before you've finished your coffee, the number of words in the file stops telling you how much work is left. This post is how I see the change: what's wrong with per-word pricing now, what's being tried instead, and what I'd ask for as a translator.

## Why per-word pricing worked

The per-word rate was never really about words. A client doesn't want 8,000 Spanish words. They want a manual people can follow, a contract that holds up and an ad that doesn't embarrass them in front of a new market. The word count was just the easiest way to estimate the effort.

And it was a decent estimate, because the effort was fairly even. A thousand words of a manual took about as long as the next thousand. Hard texts got a higher rate, repeated text got a discount through the fuzzy-match grid, and short jobs got a minimum fee. Every one of those is a patch on the same idea, and the idea held.

It also gave both sides one number to talk about, and it's still the default. Slator [reported](https://slator.com/will-language-service-providers-finally-change-pricing-models-in-2025/) that 87% of language service providers in a 2024 survey by the Association of Language Companies still priced by the word, and 68% of the people in its own reader poll were still charging or paying that way.

## What changed with MT and AI

Machine translation is fast. Translated [says](https://translated.com/hyper-adaptive-mt-model) its ModernMT engine can translate about 3 million words a minute, which is a little faster than me on a good day. With MT or a large language model in the workflow, the same 1,000 words can need almost nothing, or a full rewrite. A plain product description in a well-supported pair might need a handful of fixes. A legal clause the engine misread with total confidence needs more work than translating it from scratch, because first you have to notice that it's wrong. Fluent and wrong is the most expensive kind of wrong.

The usual answer is a post-editing discount: the client pays a share of the full rate because "the machine did most of it". Sometimes that's fair. Often the discount is decided before anyone has opened the file, and "light post-editing" turns out to mean that the only light thing is the pay.

Translators feel it. In a [Society of Authors survey](https://europeanwriterscouncil.eu/soa-survey-uk-ai-2024/) from early 2024, 36% of translators said they had already lost work to generative AI, and 43% said their income from that work had dropped. The [European Language Industry Survey 2025](https://knowledge-centre-translation-interpretation.ec.europa.eu/sites/default/files/ckeditor5-files/ELIS-2025_Report.pdf) found the downward price trend "became more visible in 2024", and 53% of participants said the move to "discounted and less-valued post-editing" made them question their future as freelancers. Meanwhile, the post-editing standard, [ISO 18587](https://www.iso.org/standard/62970.html), is being revised under a new title about "non-human translation output". Even the standards body has stopped assuming there's a machine on the other side.

## What's replacing per-word pricing

None of these models is new, but they come up more often now that the first draft is cheap.

**Hourly or per-project fees.** Already normal for review, transcreation, terminology work and anything where the word count lies. In Slator's [2024 market report](https://slator.com/resources/how-much-does-translation-cost/), 64% of LSPs said they charge per hour for some work, and 14% per document. A "quick review" of 40 pages is never quick, and an hourly rate is the only price that admits it. The catch is trust: the client wants an estimate and a cap, and you need to track your time honestly.

**Pricing by what's at stake.** Some buyers sort content by risk instead of volume. A drug label, a contract or a brand campaign goes to a qualified human with review, and costs more. An internal FAQ or user reviews go through a lighter, mostly automated process, and cost less. I like this one, because it prices the consequence of getting it wrong, which is what clients care about anyway.

**Edit-distance pricing.** The price depends on how much the post-editor changed the machine output, measured after the job. Jost Zetzsche [suggested it in the ATA Chronicle](https://www.atanet.org/resources/rethinking-payments/) back in 2016, and by 2024 only 12% of LSPs told [Slator](https://slator.com/resources/how-much-does-translation-cost/) they charged this way. It sounds fair, and for high-volume, low-risk content it can be. Its blind spot is that reading a sentence, checking it against the source and deciding it's correct takes time and leaves an edit distance of zero. Pay purely by edits and you reward people for finding something to change. That's not an incentive I'd want in a reviewer.

**Program fees and subscriptions.** Large buyers pay a fixed monthly or yearly amount for a defined scope: MT, human review, terminology and QA. It's how companies already buy software, and it makes budgets predictable. For a freelancer it mostly shows up as a retainer, which is a lovely word for "regular client who pays on time".

**Outcome-based pricing.** In theory the price follows results, such as sales in the new market. In practice it's hard to prove that the translation, and not the product, the timing or the weather, moved the number. Where the term does show up, it can point the other way: in 2026 the translation AI company ModelFront [announced outcome-based pricing](https://modelfront.com/news/outcome-based-pricing-2026) that, in its words, "makes it straightforward to shift budget from manual post-editing to AI". So the outcome being measured there is a smaller post-editing bill.

## What this means for translators

The worry is easy to state. If a machine does the "first 80%" and you do the rest, will you be paid for the judgment that makes the text usable, or for 20% of the words? That depends on how the change is negotiated, and right now a lot of it is being negotiated by the side that buys.

Under a pure per-word model, speed was king: the faster you translated, the better your real hourly rate. With AI in the workflow, the value is in the parts that don't scale with speed. Knowing the field. Catching the confident mistake. Knowing when a literal sentence will land badly with a reader in Mexico City or Madrid. A medical translator who spots a wrong dosage in a machine draft isn't being paid for the three words they changed.

Per-word rates won't disappear. They'll stay for years in freelance work, small jobs and pairs where MT is still weak, because they're simple and everyone understands them. What I'd change is how they're used. I charge 50% of my translation rate for review and 60% for post-editing. A share of my own rate is simpler than a separate price list: when my rate changes, everything else moves with it, and there's one number to agree on instead of three.

- **See the output before you agree to a post-editing rate.** Ask for a sample of the actual machine output, not a promise that "the engine is very good".
- **Quote review, transcreation and messy files by the hour,** with an estimate and a cap.
- **Price by risk.** A contract and a newsletter aren't the same job, even at the same length.
- **Keep a minimum fee,** so a 50-word urgent job "for yesterday" still pays for opening the file.
- **Put it in writing:** your rates, your match grid and whether you use MT at all.

## How TraduQ fits

The free [word count and quote tool](https://traduq.com/tools/translation-quote) I built into TraduQ lets you price a job per word, per hour or as a fixed fee, set your own share of the rate for each match band, and add a rush surcharge or a minimum fee. It's free, like everything on TraduQ, and its counts run in your browser.

## Sources

Pages checked on 7 November 2026:

- [Slator: Will LSPs finally change pricing models in 2025 (December 2024)](https://slator.com/will-language-service-providers-finally-change-pricing-models-in-2025/): ALC survey, per-word share, reader poll
- [Slator: How much does translation cost (July 2024)](https://slator.com/resources/how-much-does-translation-cost/): hourly, per-document and edit-distance pricing among LSPs
- [Translated: ModernMT Hyper Adaptive (October 2023)](https://translated.com/hyper-adaptive-mt-model): MT speed
- [European Writers' Council: Society of Authors AI survey (May 2024)](https://europeanwriterscouncil.eu/soa-survey-uk-ai-2024/): translators losing work and income
- [European Language Industry Survey 2025 (report, PDF)](https://knowledge-centre-translation-interpretation.ec.europa.eu/sites/default/files/ckeditor5-files/ELIS-2025_Report.pdf): price trend and post-editing
- [ISO 18587:2017, Post-editing of machine translation output](https://www.iso.org/standard/62970.html): revision in progress
- [ATA Chronicle: Rethinking payments, Jost Zetzsche (2016)](https://www.atanet.org/resources/rethinking-payments/): edit distance
- [ModelFront: Outcome-based pricing (June 2026)](https://modelfront.com/news/outcome-based-pricing-2026)

## About the author

[Andrés Lacruz](https://traduq.com/translators/andres-lacruz-6ae1gbuo) — EN↔ES translator and creator of TraduQ

## Related posts

- [Translation Rates 2026: Per Word, Per Hour & Budget Guide](https://traduq.com/blog/translation-rates-guide) — English to Spanish runs about $0.10 a word, or $100 per 1,000 words. Rates per word, per hour and per page by language pair, with the surveys behind them.
- [How to Open memoQ Files (MQXLIFF, MQXLZ, MQOUT) Without memoQ](https://traduq.com/blog/open-memoq-files-without-memoq) — What .mqxliff, .mqxlz and .mqout files are, how to open and translate them without a memoQ licence, and what your project manager needs back.
- [How to Open an SDLPPX (Trados Package) Without Trados](https://traduq.com/blog/open-sdlppx-without-trados) — What's inside a Trados .sdlppx package and how to open and translate it without Trados, in memoQ, Wordfast Pro, Smartcat or free in the browser.

## Tools for this

- [Word count & quote](https://traduq.com/tools/translation-quote) — Count words by match band against your TM or CAT report, price the job and download the quote PDF.

## Glossary

- [CAT analysis](https://traduq.com/glossary/cat-analysis)
- [CAT tool](https://traduq.com/glossary/cat-tool)
- [Fuzzy match](https://traduq.com/glossary/fuzzy-match)
- [Minimum fee](https://traduq.com/glossary/minimum-fee)
- [Project package](https://traduq.com/glossary/project-package)
- [SDLXLIFF](https://traduq.com/glossary/sdlxliff)
- [XLIFF](https://traduq.com/glossary/xliff)
