Click here to get this post in PDF
Picture a shopper in Madrid landing on your product page. She’s interested. She’s got her card out. Then she hits a sentence that doesn’t quite make sense, a product name that reads slightly off, a shipping note that seems to contradict itself two lines later. She doesn’t email you about it. She doesn’t leave a review. She just closes the tab and buys from the competitor whose site sounds like it was written by a person who understands her.
That moment happens more than most small business owners realise, and it happens for a very specific, very fixable reason.
The number that should worry every SME selling abroad
According to CSA Research’s “Can’t Read, Won’t Buy” study, one of the most cited pieces of research on this subject, 76% of online shoppers prefer to buy products with information available in their native language, and 40% say they will never buy from a website that isn’t. That’s not a soft preference. That’s nearly half your addressable international market walking away before you ever get the chance to make your case.
For a small business expanding into a new market for the first time, this is easy to underestimate. You assume a product page that reads “well enough” is good enough. It isn’t. Shoppers aren’t grading your translation on a curve. They’re deciding, in about three seconds, whether your business feels trustworthy enough to hand their card details to. A clumsy sentence doesn’t read as “small business doing its best.” It reads as “this might not be a real company.”
Why “good enough” AI translation keeps failing SMEs
Here’s where most small businesses get tripped up. AI translation tools have gotten remarkably fluent, so it’s tempting to assume fluent means accurate. It doesn’t. A single AI model can produce a sentence that sounds completely natural and still be wrong, because it’s making one confident guess with no way to check itself. There’s no second opinion built in.
This isn’t just a small-business problem. It’s one of the most common (and expensive) mistakes small businesses make when expanding internationally: treating translation as a box to tick rather than a trust signal to protect. Even large, well-resourced companies get burned by it. HSBC’s global “Assume Nothing” tagline was mistranslated in several markets as “Do Nothing,” the exact opposite of the intended message, and the bank spent $10 million on a rebrand to fix it. That’s not even the most expensive example on record: one mistranslated word once cost $71 million in a medical malpractice case, a reminder that translation errors aren’t just embarrassing, they carry real financial risk. If a bank with a full marketing department can miss a mistranslation like that, a five-person eCommerce team relying on a single free translation tool is not immune.
The problem isn’t that AI translation is unreliable. It’s that trusting one model’s single output, with nothing to check it against, is a bit like asking one person to proofread their own work. Most mistakes slip through precisely because there’s no second opinion in the process.
What a second opinion actually looks like
This is the part that’s changed recently, and it’s worth small business owners knowing about it before they either overpay for full human translation on every page or underpay by shipping raw AI output live.
Rather than trusting a single AI model’s guess, some tools now run the same text through multiple AI systems at once and select the version those systems actually agree on. MachineTranslation.com’s SMART system, for example, compares the output of up to 24 AI models for every segment of text and selects the translation that the majority of them agree on, rather than betting on any single model’s interpretation. That agreement functions as a built-in accuracy check: if 20 out of 24 models land on the same phrasing, that’s a meaningfully stronger signal than one model’s confident-sounding guess. The company reports this consensus approach cuts translation error risk by roughly 90% compared to relying on a single AI model, which is the gap between a product page that reads naturally and one that quietly costs you the sale.
It’s a small shift in approach, but it directly targets the exact failure mode described above: fluent-sounding, individually wrong.
Making it practical for a small team
None of this requires an SME to hire a localization manager or budget for professional translation on every page. A workable approach for a lean team looks like this:
- Run customer-facing pages (product listings, checkout copy, key marketing pages) through a consensus-based tool rather than a single free translator, since these are the pages doing the actual selling.
- Keep large catalog files together. Translating a product catalog or spreadsheet in one upload, with formatting preserved, saves the reformatting work that eats up hours for small teams.
- Reserve human review for anything with legal or financial weight, such as terms of service or contracts, where a wrong word carries more risk than a wrong product description.
- Treat translation quality as a conversion metric, not a one-off task. If bounce rates or cart abandonment look worse on localized pages than on the English original, that’s usually a translation quality signal, not a pricing or product problem.
This isn’t just about the words either. Shopify’s own guidance on store localization makes a point worth repeating here: the details that seem minor, spelling conventions, date formats, currency symbols, are exactly what makes a page feel like it was built for the shopper rather than translated to them. Get the sentence right and skip those details, and the page still feels foreign. A consensus-based translation approach helps with the sentence-level accuracy; the rest is worth checking manually before a page goes live in a new market.
The bottom line for small business owners
Going international is one of the few genuine growth levers left for a small business that’s maxed out its home market. But the 40% figure is a real ceiling, and it’s one that’s entirely within a small business’s control to lower. The fix isn’t spending more. It’s making sure whatever AI tool is translating your product pages has a second opinion built in before that page ever goes live. Testing that against your own product copy costs nothing and takes a few minutes, which is considerably less than the cost of losing a customer who never told you why they left.
You may also like: Why Shoppers Abandon Carts (and what you can do about it)
Image source: elements.envato.com
