The Digital Divide Never Closed, It Moved: A Four-Level Reading of MSME Technology Access

The divide moved down a level, it did not close

When people say "the digital divide is closing," they usually mean the first and most visible layer: the share of firms and households that have any internet connection at all. On that measure there has been real progress. But connection is the entry gate, not the finish line, and treating it as the finish line is precisely the error that a generation of digital-inclusion policy made.

Jan van Dijk's work on the information society reframed the divide as a sequence of successive thresholds rather than a single line between the connected and the unconnected. Each threshold a firm clears reveals the next one behind it. A business can own a connection and still lack the skills to use it, or own the skills and still fail to convert usage into economic outcomes. The divide does not vanish when the visible layer is solved; it relocates to a layer that is quieter, more expensive to address, and easier for headline statistics to miss.

That relocation is the through-line of this article. The evidence base, from the OECD's cross-country SME surveys to the United States broadband record, shows MSMEs clustered not at the motivation layer but at the material, skills, and usage layers beneath it. And the arrival of generative answer engines has not reset that pattern. It has extended it.

Van Dijk's four levels of access

Van Dijk's framework, developed in The Deepening Divide and successive work, decomposes "access" into four levels that must be cleared in order. The value of the model is diagnostic: it tells you that a firm stuck at one level cannot be helped by an intervention aimed at another.

Where MSMEs actually get stuck: the material level

If the motivation level were the problem, wider awareness campaigns would close the gap. The measured barriers point elsewhere. The OECD's Digital for SMEs (D4SME) work, drawing on firm surveys across France, Germany, Italy, Japan, Korea, Spain, and the United States, finds that small firms trail medium and large firms on the foundations of material access. Only 45 percent of small firms have access to high-speed broadband, against 65 percent of medium-sized firms, according to the OECD's SME digitalization research.

The United States broadband record shows the same shape at the extreme. In the Federal Communications Commission's 2024 Section 706 report, roughly 24 million Americans, about 7 percent of the population, lacked access to fixed broadband at the 100/20 Mbps benchmark, and that share rose to nearly 28 percent of rural Americans and above 23 percent of people on Tribal lands. A separate nationally representative survey by Amazon and the United States Chamber of Commerce Technology Engagement Center found that about 20 percent of rural small businesses were not using broadband at all, with roughly 5 percent still on dial-up.

These are not marginal cases. They are the floor beneath every subsequent conversation about digital marketing, e-commerce, or artificial intelligence. A firm that cannot reliably get online cannot be reliably found online, and the material level for a meaningful minority of small firms, especially rural ones, is still unresolved.

The skills and usage levels: a capacity gap, not a willingness gap

Above the material floor sit the two levels that most often go unmeasured. The OECD's survey work names the primary barriers to SME digital adoption as low awareness, insufficient internal resources, skill deficiencies, and financial limitations. Read against Van Dijk's model, three of those four barriers are skills-and-usage problems, and all four are matters of capacity rather than attitude.

This distinction matters because it changes what a useful intervention looks like. If small firms were unwilling, persuasion would help. Because they are capacity-constrained, short of money, staff, time, and specialist skill, the fix is not encouragement but delivered capability. A tool handed to a firm without the resources to operate it clears no threshold at all; it simply moves the firm from "not connected" to "connected but not benefiting," which is a lateral move within the divide, not an exit from it.

MSMEs are not laggards in disposition. They are under-resourced at the levels of access that no headline connectivity statistic captures. That is the capacity gap, and it is the gap that every layer added since has landed on top of.

The gap widens as the tools get harder

A comfortable assumption holds that technology gets easier over time, so the divide should narrow on its own. The OECD data contradicts this for the tools that matter most. The digitalization gap between small firms and large enterprises has widened even as basic uptake has risen, and the gap is largest for the most sophisticated tools, enterprise resource planning, customer relationship management, supply-chain integration, big-data analytics, and cloud purchasing.

The cloud figures make the pattern concrete. Cloud adoption among small firms reached 41 percent in 2021, a three-point year-on-year gain, according to the OECD. Yet over the same period the gap with large firms grew from 31 to 33 points. Both statements are true at once: small firms adopted more, and fell further behind. Progress at the bottom was outpaced by acceleration at the top.

This is the usage level behaving exactly as Van Dijk's model predicts. When a new capability rewards resources, skill, and integration, the firms already ahead on those dimensions extract more from it, and the divide at the usage level widens rather than converges. It is the reason to be skeptical of any claim that a genuinely powerful new technology will, by itself, level the field for the smallest firms. The record of the last decade points the other way.

The fifth level: machine legibility

The four-level model was built for a web that humans read. Generative answer engines change the reader. When a buyer asks ChatGPT, Perplexity, Google's AI Overviews, or Gemini for a recommendation, a machine reads the web, decides which sources to trust, and synthesizes a single answer that names a few businesses inside it. Being present on the web is no longer sufficient; a business must be legible to the systems that now do the reading and the choosing.

Call this a fifth level of access: machine legibility, the degree to which a firm's identity, offering, and credibility are structured, consistent, and corroborated in the forms that answer engines can parse and cite. It is a genuinely new requirement, and it sits on top of the original four rather than replacing them. A firm that has not cleared the material level cannot reach it; a firm strong on the human-facing web can still fail it.

The emerging discipline aimed at this level already has a name in the literature. The 2024 peer-reviewed paper that introduced "Generative Engine Optimization" measured which content levers change whether a source is cited inside a generated answer, and found that adding cited statistics, quotations, and authoritative sources raised a source's visibility in the systems it tested. The object being optimized is no longer a rank in a list; it is a citation inside an answer, and the levers that earn it are properties of how legible and corroborated a business is, not merely how well it ranks.

What this adds up to

Two conclusions follow, and both resist the extremes. The first is that the digital divide for MSMEs is best understood as a stack of unmet levels, not a single line, and that most firms are stuck below the visible surface, at material access, skills, and beneficial usage. Closing the connectivity layer did not close the divide; it revealed the layers that connectivity statistics never measured.

The second is that the machine-legibility level is a continuation of that pattern rather than a rupture. It is new in substance but familiar in structure: a capability that rewards resources, consistency, and corroboration, landing on firms that were already under-resourced at the levels beneath it. The appropriate response is neither to dismiss it as hype nor to treat it as an emergency that erases everything before it. It is to measure where a given business actually stands across every level that now decides whether it is found and chosen, and to work the levels in order.

That measurement discipline is the practical value of the four-level lens. It refuses the false comfort of "you're online, so you're fine," and it refuses the false alarm of "AI has changed everything." It asks the harder, more useful question: which level is this specific firm stuck at, and what would actually move it.