The translation market, the people behind it, and what AI is actually changing

2025 was the year the translation industry stopped debating whether AI would matter and started dealing with how it reshapes pricing, quality, and accountability. The clearest takeaway is that the market didn’t collapse — it re-segmented:

• High-volume, low-risk translation accelerated towards automation and post-editing, pulling prices down.
• High-stakes translation (immigration, legal, medical, regulated submissions, brand-critical work) became more clearly defined around human responsibility, verification, and data handling — because fluency is not the same as accuracy.

Below is our evidence-based review of what happened in 2025 and what it means for clients and linguists going into 2026.

1) Market snapshot: growth, but with a different shape

The industry is still growing, but expectations have shifted from “fast compounding” to slower, more linear growth as automation changes unit pricing. Nimdzi’s 2025 market estimate puts language services at USD 71.7bn in 2024 with a projection of USD 75.7bn in 2025, while explicitly adjusting long-term growth assumptions downward compared to pre-AI forecasts.

At the same time, buyer-side indicators show more translation volume, not less. Smartling’s 2024 State of Translation findings report volumes up 30% year on year, alongside widespread intent to adopt generative AI.

What this means: demand is resilient, but the value is moving toward providers who can deliver speed plus governance (quality controls, terminology discipline, traceability, confidentiality).

2) Buyer behaviour in 2025: “AI-first” for scale, “human-first” for risk

Across procurement, we saw three dominant buying patterns:

A) “More languages, more often”

AI lowered the perceived cost of expanding languages, which pushed up volume.

B) MTPE became mainstream (and is now a default ask)

Nimdzi reports a big jump in MTPE adoption: in 2024, 62.6% of LSPs had more than 30% of projects as MTPE (up from 29.1% in 2022), and 45.2% used MTPE for at least half of their projects.

C) Official submissions still demand verifiable certification

UK authorities remain clear: if you submit a document not in English or Welsh, you typically need a translation that can be independently verified, including a statement of accuracy, date, translator name/signature, and contact details.
That requirement is operationally incompatible with “AI-only” translation because someone must be accountable for correctness.

3) How 2025 felt for translators: pressure, polarisation, and a skills shift

The human impact is real and uneven.

• The UK Society of Authors reported in 2024 that 36% of translators had already lost work to generative AI, and 43% said income had decreased in value.
• CIOL’s Translators Day 2025 survey found 37% reporting less work, while others reported similar or more — pointing to a split by niche, language pair, and client type.
• ITI’s member survey highlights caution at the professional level: 83% not using generative AI in their work, with only 17% incorporating it — contrasting with larger providers investing heavily in AI capability.

What changed in practice: many translators were pushed from “translation” into “correction” (post-editing), often with tighter rates, tighter turnarounds, and more fatigue. The Financial Times captured this shift starkly through translator accounts and the rise of proofreading machine output.

4) Quality in 2025: MT got better, but “looks right” became the problem

AI output is often impressively fluent — and that is exactly where risk increased.

Post-editing is not automatically faster

Nimdzi’s MTPE analysis frames an “efficiency gap” and documents the speed/effort tension as MTPE share rises.
In other words: the industry is doing more MTPE, but efficiency gains depend on domain, source quality, and how quality is measured.

Why this matters for clients

Errors in official, legal, medical, and compliance content are not “typos” — they can trigger refusals, delays, or liability. For that category of work, it’s not enough that text is readable; it must be correct, complete, and defensible.

Best-practice response: treat MTPE as a defined service with defined competence requirements — not a vague “quick check”.

5) Standards and what “good” looks like (human and AI-enabled)

Two standards remain central reference points:

• ISO 17100:2015 sets requirements for the core processes and resources behind a quality translation service.
• ISO 18587:2017 sets requirements for full, human post-editing of MT output and post-editor competence.

Even if a provider isn’t formally certified, these standards are useful as a practical checklist: define scope, assign accountability, ensure appropriate competence, and build QA into the workflow.

6) Regulation and compliance: 2025 brought real rules, not just opinions

EU AI Act: obligations began to bite

The European Commission published guidelines on obligations for general-purpose AI model providers in the run-up to requirements entering application from 2 August 2025 (with later enforcement milestones).
Reuters covered the compliance direction of travel: transparency requirements for foundation models, plus evaluation, risk mitigation, and reporting expectations for systemic-risk models.

UK data protection: AI and personal data stayed in the spotlight

The UK ICO’s guidance on AI and data protection remains a key reference point, and the ICO notes it is under review due to legislative changes coming into law in 2025.

Translation has a special privacy profile

Translation routinely includes personal data and “incidental” sensitive details. The EUATC/ATC guidance on GDPR and personal data in translation explains the practical implications and includes risk assessment thinking by content type.

What clients should take from this: if you are translating passports, bank statements, medical letters, contracts, or HR files, you should ask not only “Is it accurate?” but also “Where is the data going, who can access it, and what’s retained?”

7) The pipeline problem: language skills are shrinking

A quiet but critical 2025 development is the decline in formal language learning in the UK. HEPI warns of a “vicious cycle” where low uptake drives university course cuts, worsening skills shortages for employers.
This matters because high-quality specialist translation depends on deep language and domain expertise — and that expertise takes years to develop.

8) What this means for 2026: the likely direction of travel

Based on 2025 signals, we expect:

1. More segmentation by risk
   Buyers will separate “understanding” translation from “submit to authorities” translation more explicitly.
2. Transparency becomes non-negotiable
   Clients will increasingly require disclosure of AI use, data processing locations, and retention policies.
3. Quality proof beats quality claims
   Expect more sampling, independent review, and terminology governance — especially in regulated industries.
4. Human accountability becomes the premium product
   Not “we used AI”, but “we can stand behind this translation”.

9) What Locate Translate is committing to (and why)

Locate Translate’s position going into 2026 is simple: use technology to improve speed and consistency, but keep humans accountable for the final meaning — especially when documents are used for UK authorities.

For certified translations used with UK bodies, we follow the verifiable certification expectations reflected in UKVI/Home Office guidance (accuracy statement, date, signature, contact details).

For sensitive documents, we emphasise:

• confidentiality and controlled handling of personal data, aligned with established GDPR risk thinking for translation content
• clear service scoping (human translation vs MTPE vs bilingual review) using standards as reference points

If you’re a buyer, the most useful question you can ask any provider in 2026 is: “Who is responsible for the final text, and what is your process for proving it?”

Resources (for further reading)

Market and industry

• Nimdzi 100 (2025): market size estimate and growth assumptions
• Smartling: 2024 State of Translation (volume and adoption trends)
• Nimdzi: MTPE adoption and the “efficiency gap” Nimdzi

Professional bodies and workforce impact

• Society of Authors (UK): survey findings on lost work and income pressure
• CIOL: Translators Day 2025 survey reflection on workload changes
• ITI: generative AI adoption snapshot among members
• Financial Times: reporting on translators at the sharp end of AI adoption

Standards

• ISO 17100:2015 (translation service requirements)
• ISO 18587:2017 (post-editing of MT output requirements)

Regulation and data protection

• European Commission: GPAI provider guidelines and obligations timeline
• Reuters: EU AI Act guidance and compliance expectations for powerful models
• UK ICO: guidance on AI and data protection (status and updates)
• EUATC/ATC: GDPR and personal data in translation (risk framing)

UK authority requirements for certified translations

• UKVI/Home Office: translation content requirements for non-English/Welsh documents

Language skills pipeline

• HEPI: 2025 report on decline in language learning and the “vicious cycle”