By Anthony Neal Macri, CMO at LanguageCheck.ai
For most of my career, I sat on your clients’ side of the table. I’m a marketer by trade: I spent years buying translation and localisation while launching products across international markets, and today I run marketing at LanguageCheck.ai, which builds AI-powered translation quality assurance technology. From both seats, there is one sentence I’ve heard more than any other these past few years:
“The AI version is fine.”
Every language service company has heard it too, usually just before a budget conversation goes the wrong way. It’s worth taking that sentence seriously, because the client who says it is right about what they can see, and wrong about what they can’t.
Fluency is doing the hiding
Ten years ago, bad translation announced itself. Clunky phrasing was the smoke alarm: a buyer with no knowledge of the target language could still sense something was off, and quality had a visible advocate in every review cycle.
Modern AI output has been engineered precisely against that alarm. It reads smoothly almost regardless of whether it is accurate, appropriate, or persuasive. And readers take smoothness as the signal of quality: University of Maryland researchers studying trust in machine translation found that people reacted strongly against disfluent output but were “surprisingly, much less concerned” when fluent output was simply wrong. Fluent-but-flat, fluent-but-off-brand, fluent-but-inaccurate; all of it sails through the only check most buyers can actually perform.
The failure hasn’t disappeared. It has moved somewhere the client isn’t looking.
Where the damage actually shows up
Marketing copy has one job, and it isn’t to be understood. It’s to move someone: to click, sign up, trust, buy. That job depends on exactly the things that travel worst through a statistical system: idiom, rhythm, emotional register, cultural reference, the small choices a copywriter agonised over in the source language.
The commercial stakes are well documented. CSA Research’s long-running Can’t Read, Won’t Buy series, covering 8,709 consumers across 29 countries, found that 76% of buyers prefer to purchase products with information in their own language, and 40% won’t buy at all from websites in other languages. Language isn’t packaging; it’s a condition of the sale.
But there’s a twist in that same research: 65% of consumers say they prefer content in their own language even when its quality is poor. That statistic is why “the AI version is fine” feels true to clients; a mediocre version in-language really does beat no version at all. Tolerating copy and being persuaded by it, however, are different events. Consumer research has shown for years — notably in the Journal of Consumer Research by Puntoni, De Langhe, and Van Osselaer — that messages in a person’s native language carry measurably greater emotional force. Emotion is the mechanism persuasion runs on, and it is the first thing generic machine output strips away.
Then add the capability gap: multilingual benchmarking, including Appen’s 2025 work, keeps finding that large-model output quality drops outside English, and drops furthest in precisely the growth markets clients are using AI to reach cheaply.
To be fair to the machines: for routine content, AI-assisted translation often tests respectably, and the honest version of this argument concedes that. The failure mode is specific: creative, persuasive copy crossing a language and a culture. That is a small slice of a client’s content by volume, and a very large slice of its revenue consequence.
The invoice that never arrives
This underperformance persists because it is never attributed to language. When the German landing page converts at 1.1% against the UK’s 2.4%, nobody in the client’s Monday meeting says “the translation is flat.” They say the market is competitive, the pricing is off, the season is soft. The cost of “fine” never appears on an invoice; it appears as a conversion rate, in a foreign market, blamed on something else.
Which means your clients aren’t lying when they tell you they’ve seen no problems. They genuinely haven’t. Your job, and your opening, is to make the invisible visible.
Four ways to prove it, using the client’s own data
- Change the metric before you argue. A client comparing your quote to a machine’s is having a procurement conversation, and procurement conversations are lost on price. Move it: what is this market worth to them next year, and what does half a point of conversion cost? Cost per word versus revenue per market.
- Ask for one number they already have. Conversion rate by market. If the machine-translated markets persistently underperform the source market, and they very often do, the client’s own analytics start making your case before you’ve claimed anything.
- Propose a split test on a single asset. One landing page or one paid campaign: machine version against transcreated version, same spend, four to six weeks. It’s a small line item that produces decisive evidence, and merely offering it signals you know how the test ends.
- Build the “best line” exhibit. Take the client’s highest-performing source-language line, the one they spent months testing, run it through the machine, and show them on one page what happened to the device that made it work: the pun that collapsed, the rhythm that died, the cultural hook that vanished. It lands because it’s their line, not a stock example.
Sell the tier, not the fear
None of this requires arguing against AI, and I’d caution against trying: your clients use it daily, and for a substantial share of their content it genuinely is fine. The argument that wins is about matching method to commercial consequence. Support macros and spec sheets can flow through the machine, with automated checks verifying the output at the machine’s own pace. I’ll declare an interest here: that verification layer is exactly what we build at LanguageCheck.ai, so weigh my enthusiasm accordingly. The copy that carries a brand into a new market, the words that have to make a stranger feel something, deserves human craft. And the case for it isn’t sentimental. It’s arithmetic.
This is the first article in a short series on selling in the age of AI. The next piece will look at what scaled AI content is doing to language itself, and why distinctiveness is becoming the most defensible thing a language service company sells.
About the author
Anthony Neal Macri is a marketer and fractional CMO who spent years as a buyer of language services, launching products across international markets. He is now CMO at LanguageCheck.ai, an AI-powered translation quality assurance platform. He writes about marketing, visibility and AI at anthonynealmacri.com, and you can connect with him on LinkedIn.
