What we learned in 2026 from orders that combine AI translation and human review

What share of our orders now combine AI translation and human review?

About one in five. In 2026, 21.5% of our global orders included at least one line for machine translation post-editing (MTPE), our AI Translation - Post-Editing service, or our Managed AI Translation Service, up from 4.1% in the same period of 2025.

In 2025, MTPE was the only product line for this work. In 2026 we added Managed AI Translation Service and AI Translation - Post-Editing. The 2026 figure therefore includes work that had no product line a year earlier.

I believe the rise shows demand for AI translation that a human reviewer stands behind, and the sections below explain why I am careful about how far to generalize it. 

Is AI translation with human review widespread among our clients?

No, about 8% of our global clients placed an AI-assisted order in 2026, and one client accounts for roughly two-thirds of those orders. The top three clients account for 78.1% of AI-assisted orders, and the top five account for 84.2%.

The client share was 3.3% in 2025, so adoption is growing from a small base. I would not describe this as how most businesses buy translation today.

I think a high-volume buyer with repeatable content is a natural fit for post-editing, because the process can pay off across many similar documents. 

What happens to the numbers when we remove our largest AI-assisted client?

Without that client, the AI-assisted share of our remaining orders was 8.4% in 2026, roughly double the 4.1% we saw in 2025.

I would use that figure to describe adoption among our other clients. I think it is the more useful number for a buyer weighing the same choice.

As far as I understand, a service that starts with a few high-volume buyers often looks like this. I would ask any translation vendor that quotes an adoption figure how much of it comes from its largest account.

How did the share change month by month in 2026?

The share of orders with an AI-assisted line rose from 15.1% in January to 26.6% in March, then fell to 15.8% in September.

Where do our clients rarely choose AI translation with human review?

Legal, medical, and certified documents show the lowest AI-assisted shares in our orders. Most legal categories sit between 5.6% and 7.3% of orders, most medical categories between 2.4% and 6.4%, and certificates between 1.2% and 1.4%.

AI-assisted share of orders by content category. 

In Tomedes findings from 2025, raw AI reached roughly 85 to 90% accuracy on complex legal text, and a human-in-the-loop workflow brought it to 99% or more. I think clients in these categories choose more human work because one wrong clause costs more than post-editing saves.

Website content, UI strings, and technical material sit above 22%. I suspect part of that comes from the same large buyer, though I have not checked.

What does the human step cover in a post-editing order?

ISO 18587:2017 gives requirements for the process of full, human post-editing of machine translation output and for the competences of the post-editor. Tomedes is certified to that standard.

I explain how AI post-editing differs from traditional MTPE in our comparison of AIPE and MTPE.

I believe the reviewer's accountability matters more than the model that wrote the first draft, which is the idea behind human-in-the-loop AI translation.

What should a business buying translation take from these numbers?

I would take three checks from our own data, and I would apply the same three to any vendor.

1. Ask who owns the final file, and whether the vendor can name the reviewer.
2. Ask how much of the vendor's AI-assisted volume comes from its largest client.
3. Match the level of human review to the cost of an error, with more review for legal, medical, and certified documents.

Our guides to human translation, MTPE, and raw AI and to when to use human-in-the-loop go deeper on the third check. I expect the AI-assisted share outside our largest client to keep rising, though I could be wrong on the pace.

If you want AI translation with a named human reviewer on your next project, talk to Tomedes about AI translation with human review.

FAQs

Q: What is machine translation post-editing?
A: Machine translation post-editing, or MTPE, is the process where a human linguist reviews and corrects machine translation output. ISO 18587:2017 sets requirements for full human post-editing and for post-editor competences. I consider the reviewer's ownership of the final file the point of the process.

Q: What are the downsides of AI translation?
A: Raw AI translation can contain errors that read as confident. In Tomedes findings from 2025, raw AI reached roughly 85 to 90% accuracy on complex legal text, and a human-in-the-loop workflow brought it to 99% or more. The gap matters most in legal, medical, and certified documents, where an error carries a cost.

Q: Is there proofreading in translation?
A: Yes, proofreading is a separate step where a second linguist checks a finished translation for errors in grammar, spelling, and consistency. It differs from post-editing, which corrects raw machine output before the translation is considered complete. Tomedes sells proofreading as a separate service alongside post-editing.

Q: What is the difference between human translation and machine translation?
A: Human translation is written by a linguist who reads, interprets, and renders the source text. Machine translation is produced by software, now often by AI translation models, and is faster and cheaper but varies in quality. AI translation with human review combines the two, with a linguist checking the machine output before delivery.