If you own or run a boutique translation agency, this message is for you.
Last week, an LSP client of mine raised a question that many language service providers are wrestling with: If you use AI in your translation workflow, do you need to tell your clients?
Based on my information, most LSPs are saying they would never deliver machine-translated work, even with human review, without disclosure.
But a few take a different position: “If we stand behind the quality, why does the client care how we do it?”
I understand both sides. Clients pay us for results, not a running commentary on every tool we use.
But my answer starts with a distinction: Are we talking about raw machine translation, or a human-led workflow that uses AI as one component?
Of course, don’t deliver raw MT without disclosure.
Imagine that a client discovers a significant error. You explain that you use an advanced AI workflow, but that some mistakes can still get through.
Their next questions are predictable: “You’re using AI? We didn’t know that. We didn’t authorize it. And are we paying the same price as before?”
Now you have two problems. One is the translation error. The other—and potentially much bigger one—is a breach of trust.
Human translators make mistakes, too. But a human workflow gives you a kind of plausible deniability: “My best linguist checked this. I believed it was ready to deliver. I’ll investigate what went wrong. I’m sorry.” You had someone responsible for reviewing the translation, and you reasonably expected that review to catch the problem.
With unreviewed AI output, you can’t make the same argument. You chose a workflow that you knew is error prone, without a linguist checking the result. If the client only learns that after finding a mistake, the conversation shifts from “How did this error happen?” to “Why didn’t you tell us what we were buying?”
That’s why raw MT should be sold explicitly as raw MT.
But human-reviewed AI translation is a more nuanced question.
Whether to deliver human-post-edited MT or AI translation—i.e. MTPE or AIPE—without telling your clients is a harder decision.
If you start with AI and deliver a human-post-edited translation, you may not need to disclose it, provided you’re honoring your client agreements and confidentiality requirements. That argument becomes especially compelling if you take the savings and invest them in a more rigorous QA process, so the end result is actually better than anything you’ve delivered before.
This is entirely doable right now. I can recommend ways to upgrade your QA process with AI—not someday, but today.
Besides, many of your translators are probably already using AI in their own workflows. You may be delivering “shadow MTPE”; you just don’t know which jobs it’s happening on. The question, then, isn’t simply whether to introduce AI. It’s whether to bring that work into a deliberate, managed process where you can put the productivity gains toward a comprehensively better solution.
But I think the disclosure debate, by itself, is too small a frame for the business decision ahead.
The growth question is bigger than disclosure.
Pricing pressure is relentless. If your response is only, “We only do high-quality, full-human translation,” you’re going to lose clients.
But if you simply focus on “How do I deliver the same work for less?” you risk turning every productivity improvement into another price concession.
That’s why the better question is: “How do I deliver more value for the same budget?”
Your clients already have their budget and they probably have content they would translate if they could fit it into that budget. They may want more languages, more frequent updates, or access to larger volumes of internal information that never justified full translation rates.
Some of that work is already going through free tools. Some is being handled inconsistently by internal staff. Some simply remains untranslated.
If you offer only one service at one price point, much of that demand stays outside your business. And when clients go elsewhere for their lower-stakes needs, you create an opening for them to forget about you for all of their other work, too.
I see the strongest path to growth in having the capacity to deliver a full spectrum of service levels and then helping clients get way more content done into way more languages.
That is the purpose of my Three-Tier Service Model for LSPs.
Three tiers are a starting point—not the limit.
Premium Human Translation, MTPE, and Premium Raw MT give clients a simple framework. As a customizable pricing baseline, I suggest your standard rate for Tier 1, around 50% for Tier 2, and 10–15% for Tier 3.
These are starting points, not universal promises about costs or margins. Your language pairs, project requirements, review effort, and operating economics determine the actual offer.
In practice, we have access to a whole spectrum of quality/cost tradeoffs—not just three fixed options. AI-assisted QA can help focus human review where it matters most, catch document-wide inconsistencies, and support linguists inside their CAT tools. You can use those capabilities to reduce costs—or reinvest the savings in better quality.
But why would clients pay for something they can get “free”?
Because free translation tools do not eliminate the cost of getting useful work done.
Someone still has to prepare files, move text between systems, repair formatting, manage terminology, assess data-handling arrangements, and decide whether the output is suitable for its intended use, including making sure everything is on-brand and consistent.
Those tasks consume staff time…. and require skills most clients don’t have.
Not every tool has the same limitations or privacy terms. But clients still need someone to evaluate the options and take responsibility for a coherent workflow.
Your pitch is not, “Our machine translation costs less than a free chatbot.”
It is, “We can manage this language workflow for less than it costs you to manage it yourself.”
And that is what your clients really want: They just want just someone to take the multilingual problem off their hands so they can get back to their main business.
That shifts the conversation from the price of generating words to the value of solving a business problem.
It also makes the disclosure question easier. Instead of quietly changing production behind an undifferentiated service, you help the client choose among clearly described options.
I’ve put the full Three-Tier Service Model, along with answers to common questions about pricing, cannibalization, quality expectations, security, and getting started, here:
