Your Supplier's Perfect English Is a Subscription: Vetting Business Partners in 2026
Ten years ago, a supplier's English told you something real. If a factory manager in Shenzhen or a freight forwarder in Rotterdam wrote back within the hour in clean, idiomatic, businesslike English, that fluency was expensive to fake. It meant a trained employee, a dedicated correspondence desk, or a founder who had personally invested years in the language of your market. Flawless grammar and a fast, polished reply were a proxy for organization: someone competent was minding the account.
That proxy is gone. In 2026, a supplier with no English speaker on staff, no functioning quality system, and a shell company's worth of paperwork can produce a flawless, warm, perfectly punctuated reply in any language, in seconds, indefinitely. Perfect written English is now a subscription, not a skill — and every credit team, procurement officer, and accounts-receivable manager still scoring counterparties on "how professional did their emails sound" is measuring noise. This piece lays out which due-diligence heuristics have quietly died, which ones still work, and a practical sequence for vetting a new international business partner before you extend terms, ship product, or sign a contract.
Why correspondence quality stopped being a reliable signal
The logic of the old heuristic was simple: producing fluent, professional written communication used to require either a native or highly trained speaker, or a slow, visibly imperfect process of drafting and translating. Either way, the quality of the English (or French, or Mandarin, or Portuguese) in a supplier's reply correlated with something real about the organization behind it — headcount, training investment, seriousness about export markets. A rushed, badly translated reply was itself informative: it told you this counterparty was small, under-resourced, or new to international trade.
That correlation has broken because the cost of producing fluent correspondence has fallen close to zero and the speed has fallen close to instant. Anthropic's Economic Index, which studies how AI usage maps onto real work tasks, finds that a meaningful share of everyday AI output is exactly this kind of writing — "email or message" drafting accounts for 4.61% of the artifact types Claude conversations produce globally, and "document or report" output for 14.91% more. Separately, the Index's occupational task data shows that "compose business correspondence for supervisors, managers, and professionals" is a measurable, non-trivial share of matched work tasks worldwide (0.76%), and that this share is elevated well above that baseline in some markets — in the UAE specifically, correspondence-drafting tasks run to 1.35% of matched activity. That is not a claim about any individual company's behavior; it is evidence that correspondence generation is a common, mainstream, and geographically uneven use of AI tools right now, not a rare or hypothetical one.
Layered on top of that is a second Index finding: globally, just over half of measured AI usage (48.62%) reflects the task being substantially handed off to the AI system rather than the human staying actively in the loop as a collaborator (51.38% augmentation). Applied to correspondence, that split matters for due diligence: a rising share of the "professional" emails, proposals, and website copy a buyer reviews during vetting may have been produced with the human doing comparatively little of the actual composing. Fluency, in other words, no longer implies a fluent — or even present — human behind it. It also says nothing about whether the counterparty is legitimate, solvent, or capable of fulfilling an order; it only explains why the writing itself has stopped being diagnostic.
Dead signals: what no longer tells you anything
None of the four heuristics below are "bad" exactly — a genuinely strong supplier can still show all of them. The problem is that a genuinely weak, under-capitalized, or fraudulent one can now show all of them too, at effectively zero marginal cost. That symmetry is what kills a signal's usefulness in due diligence: it only works if the thing being faked is expensive to fake.
The common thread is that each of these used to be costly to produce and is now cheap. A due-diligence process that still weights them heavily is effectively scoring counterparties on how recently they opened a chat window, not on how creditworthy, capable, or real they are.
Live signals: what still carries real information
The signals below survive because they are either anchored to a public record that is expensive or illegal to falsify, or because they test something a generative tool cannot manufacture on demand: a verifiable history, a physical footprint, or the ability to improvise a specific, unscripted answer under follow-up.
Notice what these have in common: they all point outward, to a record or a party independent of the counterparty's own communications. A registry entry cannot be reworded by better prose. A bank will not vouch for a company that does not hold an account. A named reference customer either picks up the phone or does not. None of these can be manufactured by making an email sound more professional.
Documentation willingness as its own signal
One live signal deserves separate treatment because it is behavioral rather than documentary: how a counterparty reacts when you ask for verification. A legitimate, established supplier is generally unbothered by a request for a certificate of incorporation, a bank reference letter, export licenses, or a recent audited financial statement — producing these is routine for them because they already exist and are already used elsewhere. A counterparty built primarily on persuasive correspondence tends to respond to the same request with delay, deflection, a request to "trust the relationship first," or documents that are themselves suspiciously polished but unverifiable. The friction (or lack of it) in getting real paperwork is more diagnostic than the paperwork's appearance once you have it.
This is also where fluent AI-assisted writing can work against a bad-faith counterparty rather than for them: a scripted, generated reply to a specific documentation request often reads as fluent but non-responsive — warm in tone, vague on the actual ask. Learning to notice that gap between tone and substance is now a core due-diligence skill in its own right.
A practical due-diligence sequence before extending terms
None of these four steps depend on how the supplier writes. They depend on records that exist independently of the relationship, and on behavior — willingness to document, consistency under specific questioning, speed of independently-verifiable corroboration — that a drafting tool cannot supply on a counterparty's behalf. A credit team that runs all four before extending terms is testing the business, not the prose.
The Anthropic Economic Index figures cited above describe AI usage matched to workplace tasks in a single-period snapshot — they show what share of measured conversations involve correspondence-style drafting or full task automation, not a trend line, and not the behavior of any specific supplier, counterparty, or individual. The occupational and country figures reflect tasks commonly performed in a given category, not a claim that any named company or person used AI to write a particular message.
Practically, this data does not change what credit teams are legally required to verify before extending terms — it explains why one long-standing shortcut (judging competence from writing quality) has stopped working, and why the verification steps above now carry more of the weight that fluent correspondence used to carry.
Frequently Asked Questions
Does this mean AI-assisted correspondence is a red flag on its own?
No. Most legitimate businesses now also use AI tools to draft routine correspondence, and that alone tells you nothing bad about them. The point is narrower: fluent, fast, polished writing has stopped being evidence of legitimacy either way, so it should carry close to zero weight in your assessment rather than being treated as reassuring.
What is the single fastest check to run on a new international supplier?
An independent registry lookup. Take the company registration number the counterparty provides and search the relevant national or state company registry yourself, rather than clicking a link they send. It takes minutes and immediately confirms whether the entity, its filed directors, and its registered address actually exist as described.
How do we verify physical presence for a supplier we can't visit in person?
Request a live, unscripted video walkthrough of the facility rather than pre-recorded footage, cross-check the address against satellite or street-view imagery, and where the relationship value justifies it, commission a local inspection agent or trade attaché to confirm the site in person.
Why does asking the same question twice, weeks apart, help?
Specific operational details — a shift pattern, a certification number, a named contact at a bank or reference client — are easy to state once but hard to reproduce consistently without an underlying reality behind them. Comparing answers given in separate, unscripted conversations surfaces inconsistencies that a single polished reply would never reveal.
Should we stop trusting formal, professional-sounding proposals altogether?
Not stop trusting them — stop scoring them. Treat tone and polish as neutral, and shift the weight of your decision onto registry verification, payment history, references you contact independently, and willingness to produce real documentation quickly.
What if a supplier resists providing bank references or financial statements?
Treat resistance or delay as meaningful data rather than a minor inconvenience. Established, financially healthy counterparties generally have this documentation on hand already and produce it quickly because they use it with other partners too; persistent deflection is one of the more reliable warning signs available.
Sources and References
Verifying a counterparty before money or goods move is exactly the discipline that prevents a recovery case later. Contact Cosmopolite for a free case assessment. No recovery, no fee.
