MTPE has gone from interesting-if-you-squint to fairly standard translation practice in the space of about three years as tech has improved.
In 2026, the interesting questions now are not so much about whether MTPE can produce outputs equivalent to full human translation but how to use machine translation post-editing, effectively, where it has the most limitations, and how to make sure the cost savings on offer do not counter-intuitively cost more than they save.
This blog is unashamedly aimed at those who might procure MTPE services - translation and localisation managers, heads of marketing, content and ops leads with international scope, and anyone in a regulated or competitive sector being asked to do more in more markets without much extra budget.
For completeness, and only briefly: when we talk about MTPE we mean a process where a human linguist, ideally a subject-matter expert, reviews and improves machine-translated content.
Done to the right (ICS-translate… ahem..) standard, it considers meaning, accuracy, terminology, tone, formatting, sector-specific language, regulatory sensitivity, and - increasingly - multilingual SEO and CRO requirements.
Done poorly or with the wrong motivations (i.e. saving cost at any.. erm.. cost..), it means tidying up some Google Translate or ChatGPT (or similar) output and hoping nobody notices. Aiming for 'good enough' as the best possible outcome is a risky strategy, however.
The strongest case for MTPE is rarely "we want to spend less money", even if that is how the conversation often starts. The stronger case is the ability to cover more tasks at an acceptably good level, and achieve more measurable results than would be possible with 'full' human translation… particularly in multilingual marketing and product/software launches.
Many marketing and localisation teams have a demoralisingly-large wishlist of things they would localise into more languages if they could.
This isn't always the most 'essential' or cleverest content, but content that helps customers self-serve, find their own way and which helps give a broader picture of the brand and its products or services.
For example, moderately-trafficked help centre/FAQ articles, lower-priority product pages, knowledge base content, internal documentation, older campaign material that is no longer worth fully retranslating but is also too useful to leave broken.
For multilingual marketing and customer engagement this type of content can have individually small impact but be collectively significant - especially when we consider long tail, non-brand keywords in international SEO.
Done properly, MTPE lets teams reach a wider footprint with the same headcount and reserve fully human translation for the work that needs it - regulated content, brand-led campaigns, high-conversion landing pages, anything where mistakes are expensive or where there are substantial opportunity costs to getting things wrong.
On a purely practical and partially anecdotal point, MTPE can also help with removing the risk of team burnout.
This is particularly the case where internal subject-matter experts are being pulled in to sense-check everything that moves because they happen to speak the language.
If those people are spending their afternoons fixing low-stakes copy because nothing else has scaled, that can be limiting and frustrating - MTPE can provide a better way to have a more comprehensive and opportunity-focused/prioritised approach to translation, that allows internal teams to focus on their day jobs.
I came back from Learning Technologies at ExCeL a couple of weeks ago thinking about how much edtech and L&D content now lives on screens with very specific spatial constraints.
Adaptive learning platforms, microlearning apps, mobile-first courseware, gamified training, AI tutors and so on are being built for global rollout from day one but in a sector where user engagement can be lacking to say the least, MTPE can help.
By that I mean edtech product users may be 'forced' to absorb course content in their workplace and this content is rarely just words on a page. Excellently localised content that maps to UX can help propel users through the process - and detail-oriented post-editing can account for the quirks of software.
Character limits, UI layouts, branching dialogue, age-appropriate language, regional pedagogy, character voice, tone consistency across hundreds of strings and the like are not areas a generic MT engine handles reliably, which means the brief, the glossary and the sector experience matter much more than they would for a help centre article.
Light post-editing tends to fail badly with screen content where copy might need to be re-evaluated more substantially. Full post-editing by linguists who actually know software, games or edtech can work very well - and it is one of the areas where rigorous, friendly process makes the most obvious difference to the people on the receiving end.
Most MTPE work falls into two categories: light post-editing and full post-editing. Light post-editing aims for "understandable, accurate enough, free of major errors" and tends to suit internal use, lower-risk support content, and material where speed matters more than polish.
In some highly competitive paid media contexts, brands also accept lighter editing on rapidly iterated ad copy where functional performance matters more than linguistic perfection - for example a Spanish-language casino ad that will only be live for a tiny period of time needs to convert, not win awards.
Full post-editing aims for something much closer to professional human translation, and is the right choice for published content, customer-facing material, and anything that needs to do real persuasive or commercial work in the target market. For screen-based content, full post-editing is almost always the right starting point given the UX considerations.
It is also worth being clear with the linguist before work begins. If a post-editor is briefed to "just check it quickly" and then judged against full human translation standards, the result is a grumpy linguist, an unhappy client, and a translation that sits awkwardly between the two. Worth avoiding.
"MTPE pricing" is one of the most-searched topics in the translation space, and one of the more tricky to answer responsibly and accurately.
What we can say is that light post-editing tends to come in around 40-50% of the equivalent fully human per-word rate, and full post-editing tends to land between 50% and 70%.
Some providers price hourly, particularly when MT quality is unpredictable. Hourly rates for experienced post-editors typically sit between £50 and £70, though this varies by language pair, expertise (do they know SEO for example?) and seniority.
Those numbers hide a lot, particularly when it comes to revision as the actual final cost depends on source text quality, MT output quality, language pair, sector, volume, the level of post-editing iterations required, and any additional services like SEO review, CRO adaptation or compliance sign-off.
If, as a customer, a quote arrives without anyone asking about most of those factors, that is itself arguably a useful signal about the level of expertise-focus.
Inevitably, if the machine translation output is poor enough, post-editing can take longer than translating from scratch, because the linguist first has to work out what has gone wrong.
In those cases, the apparent cost saving is completely illusory and a fully human approach is both safer and probably cheaper.
Anyone offering MTPE as a flat percentage discount on translation, without first looking at the source material or the MT quality, is either being wildly optimistic or hoping you will not notice the underperforming content that is shared back to you.
"Full post-editing brings the result much closer to human quality, but a skilled linguist working intensively on a poor machine translation output is not the same as that linguist writing from scratch, and the time and cost saving might be smaller than clients expect. For high-stakes content, where accuracy, register and tone genuinely matter, choosing full post-editing or fully human translation is a sound investment."
Sara Robertson, CEO of the Institute of Translation and Interpreting
Generally yes, but it depends. Light post-editing tends to come in at around 40-50% of a full human per-word rate, and full post-editing at 50-70%. Savings depend heavily on source text quality, MT output quality, language pair and domain. Where MT output is poor, the saving can disappear entirely - or even reverse, if the linguist ends up retranslating from scratch.
ISO 18587 is the international standard for full post-editing of machine translation output. ISO 17100 is the broader translation services standard. Reputable MTPE providers should be able to confirm which standards they work to, and ideally without too much hesitation. Some ISO 17100-certified agencies (like ICS-translate) don't hold ISO 18587 accreditation but have internal quality procedures for highly competitive sectors that are worth evaluating.
Generally not. Legal agreements, contracts, medical instructions, patient-facing healthcare content and pharmaceutical material with regulatory implications tend to be safer with fully human translation, given the risk of harm or liability if a translation is wrong. The savings on offer rarely justify the exposure.
Often, with care and with extra focus on the 'post-editing' part. Screen-based content - games, edtech, software, apps - tends to depend heavily on context, character limits, UI constraints and tone. Light post-editing tends to fail here. Full post-editing by linguists with actual sector experience can work well, particularly where the brief includes glossaries, character voice notes and UI context.
It's faster, in most cases. Human translators typically produce around 2,000 words a day, while a well-functioning MTPE workflow can deliver 3,000-5,000 words a day depending on content type, MT quality and post-editing depth. The variance matters - a poor MT output can wipe out the speed advantage entirely.
Sometimes, but with caveats. MTPE on its own tends to translate metadata, keywords and on-page content fairly literally, which often misses the right local search intent. For SEO-sensitive pages, MTPE tends to work best alongside multilingual keyword research and an SEO review pass, rather than in isolation. The translation can be technically correct and still rank for nothing.
If MTPE is on your roadmap and the workflow is starting to show strain, it is probably worth a conversation.
ICS-translate works across iGaming, finance, health, legal, pharma, edtech, software, mobile and video games, with ISO 17100-certified translation processes and direct integration with our multilingual SEO, CRO and paid media teams at ICS-digital.
For the practical side - how to brief MTPE projects properly and how to spot when your current provider is not pulling its weight - have a look at our companion piece, When to Switch MTPE Agency: A Practical Guide and 2026 Red Flags.
You can also have a look at our iGaming, crypto and blockchain, ecommerce and travel and tourism sector pages for a closer look at how this works in practice.
