Cross-Border Digital Trade by Accident: What the Data Says About US MSMEs Going Global Without Meaning To

A local business that never chose to go global

Consider a two-chair salon, a solo attorney, or a plumbing company that works one metro. None of them file export paperwork. None of them ship abroad. Yet each of them now maintains a public, machine-readable presence that any person or system in any country can read: a website, a Google Business Profile, a set of reviews, and increasingly an entry that an answer engine can quote when someone asks it a question. In the language of digital trade, the storefront is already global even when the business is entirely local.

This is the accidental part. The firm made no decision to participate in cross-border digital trade, but the conditions of participation, a discoverable and comparable online identity, are met by default. The exposure runs both ways. A foreign buyer can find the business, and a foreign competitor can be surfaced next to it in the same answer or the same set of results. The question this piece examines is how much that matters, and answering it requires separating what the data establishes from what it does not.

What the export data actually shows about small firms

Start with the ground truth, because the accidental-exposure thesis only earns attention if the deliberate version is already well understood. United States small firms participate in cross-border trade in very large numbers and at a very small scale per firm, and that pattern is stable and well measured.

Why crossing a border was always a selection effect

The reason most small firms export little is not an accident of measurement. It is the central prediction of the trade theory that has organized this field for two decades. In his 2003 model of heterogeneous firms, Marc Melitz showed that only firms above a productivity threshold self-select into exporting, because the fixed costs of reaching a foreign market screen out the rest. Firms that cross borders are systematically different, more productive, from firms that do not.

That framework explains the export data cleanly. The small firms that do export are the ones productive enough to clear the threshold, and there are relatively few of them by value. It also sets up the interesting tension in this piece. Melitz describes a world where going global is a costly, deliberate act that most firms rationally decline. Digital discoverability does not repeal that logic for actual selling and fulfillment, which still carry real fixed costs, but it does quietly lower the threshold for one specific thing: being seen and compared across a border. Presence and transaction have come apart.

The paperwork got fixed faster than the capability

The policy layer has been moving in the direction of small-firm enablement for years. The World Trade Organization frames its e-commerce moratorium, which keeps electronic transmissions free of customs duties, and the Trade Facilitation Agreement provisions for e-signatures, paperless trading, and e-invoicing, as explicit supports for smaller businesses trying to trade across borders. On paper, the door is being opened.

The same institution is candid that the door being open is not the same as a small firm being able to walk through it. The WTO handbook for policymakers supporting micro, small, and medium enterprises names the binding constraints directly: smaller firms shoulder disproportionate burdens from non-digital customs procedures, lack familiarity with the mechanics of e-commerce, and face persistent transport and logistics gaps. The legal and technical layer is being fixed faster than the operational-capability layer. This is the shape of every wave of borderless commerce: the door opens before most firms can use it.

Read against that backdrop, accidental cross-border exposure sits in an odd position. The capability gap the WTO describes governs deliberate exporting, actually selling and shipping to a foreign customer. Discoverability sidesteps that gap entirely, because being findable requires no customs knowledge, no logistics, and no decision. A firm can be globally visible while remaining wholly unable, and uninterested, in globally transacting.

How discoverability made the exposure accidental

The mechanism that turns a local business into an accidentally cross-border one is the same mechanism reshaping local discovery generally: the shift from a private, click-through web to a synthesized, machine-read one. Two data points frame it.

First, being read by a machine is now decoupled from being clicked by a person. A rigorous Pew Research Center browsing-panel study of United States adults in March 2025 found that when a Google AI summary was present, users clicked a traditional result in about 8 percent of searches, versus 15 percent without a summary, and clicked a link inside the summary itself in only about 1 percent of visits; they also abandoned the session more often, 26 percent versus 16 percent. The engine read the sources and answered; the human often never visited any of them. A business can be part of the answer a foreign or distant user receives without ever registering a visit from them.

Second, the set of businesses a generative engine will name is far narrower than the set classic search will list, which sharpens who gets caught in a cross-border comparison and who is invisible to it. Industry monitoring, which we flag as not yet peer-reviewed, reports generative engines recommending on the order of 1 percent of local businesses in category queries against roughly a third for classic local search. We treat that specific ratio as contested and directionally suggestive rather than established, but the direction is consistent with the Pew evidence: the answer layer is a narrower gate, and being inside it or outside it is now part of how any buyer, near or far, encounters a business.

Reading the accidental thesis

It would be easy, and wrong, to inflate this into a claim that every corner shop is now an international exporter. The data does not support that. Here is the careful version.

What is established: the export figures, the productivity-threshold theory, the WTO capability gap, and the measured decoupling of machine-reading from human-clicking. What is emerging, and framed as such, is the interpretation that these combine into a meaningful new category of accidental cross-border presence for local firms. There is no audited dataset yet quantifying how often a purely local United States business is actually surfaced to, or compared by, a foreign buyer through an AI answer. That primary measurement does not exist in the public literature at the time of writing.

There is also a genuine counter-reading worth stating. Chris Anderson's Long Tail thesis holds that low distribution costs let niche and small participants win by reaching dispersed demand. One optimistic version of accidental exposure is simply the Long Tail arriving for local services: a distant buyer who genuinely wants exactly what a small firm offers can now find it. Whether that upside materializes, or whether the narrow answer gate mostly concentrates attention on a few already-visible firms, is an open empirical question, not a settled one.

What this means for a business that only serves its town

The practical takeaway is smaller and more useful than the headline. A local business does not need an export strategy. It needs to understand that its online identity is already readable across borders and already comparable against anyone, and that the surfaces doing the comparing, classic search, the local map pack, AI answers, and reputation, are now the surfaces that decide whether it is seen at all.

That reframes the work. The point is not to chase foreign customers. The point is that the same structured, machine-legible, well-reviewed presence that wins a nearby buyer is also what determines how the business appears to any distant one, and whether it appears in the narrow answer at all. You cannot manage an exposure you cannot see. The first step is measuring where the business actually stands across those surfaces, which is a diagnostic question, not an export one.