Everyone Copied Everyone Else's AI Strategy

I counted the press releases once. In the six months after ChatGPT launched, something like forty SaaS companies announced "AI-first" strategies on the same template: a blog post, a quote from the CEO about "transforming the industry," and a promise that AI would be embedded in every product by next quarter. The wording was so similar across companies that you could swap the logos and nobody would notice. DiMaggio and Powell (1983) have a name for this. They called it mimetic isomorphism, and they were not complimenting anyone.

The word isomorphism sounds technical but the idea is simple. Organizations in the same field tend to look alike over time, and not because they all independently converged on the optimal structure. They copy each other. DiMaggio and Powell identified three pressures that make organizations converge. Coercive pressure comes from outside: regulations, laws, accreditation standards. GDPR is coercive isomorphism. The EU told companies to change how they handle data, and companies changed, not because they independently decided privacy was good business, but because the alternative was fines up to four percent of global revenue. Mimetic pressure comes from uncertainty. When nobody knows what the right move is, the safest thing is to copy whoever seems successful. After ChatGPT went viral, every enterprise software company bolted a chatbot onto its product and called it AI strategy. That was not strategy. That was mimetic isomorphism with a press kit. Normative pressure comes from professionalization: shared educational backgrounds, industry certifications, professional standards. When every CISO gets the same certification body, reads the same frameworks, attends the same conferences, they start making the same recommendations. The normative pressure makes those recommendations feel like best practice rather than what they are: shared habit.

There is a trap in here that my professor flagged repeatedly. Coercive isomorphism must be external. An internal executive mandate to "adopt AI across the organization" is not coercive isomorphism, no matter how forceful the CEO sounds in the all-hands meeting. Coercive means government regulation, legal requirements, or accreditation standards pushing from outside the organization. AACSB accreditation pushes business schools toward certain structures. FedRAMP pushes cloud vendors toward certain security practices. A CEO screaming about digital transformation is just a CEO screaming.

Scott (1995) extended the institutional picture with three pillars. The regulative pillar covers what DiMaggio and Powell would call coercive and some normative pressure: laws, rules, sanctions. The logic here is expedience. You comply because the cost of noncompliance is too high. The normative pillar covers values, norms, and professional obligations. The logic is social appropriateness. You adopt a practice because professionals like you are expected to adopt it. The cultural-cognitive pillar is the quietest and the most powerful one. It operates through taken-for-granted schemas. Organizations adopt technologies not because they are required, and not even because they are considered best practice, but because alternatives are literally inconceivable within the organizational frame. Nobody at a major university debates whether the institution needs a website anymore. The question does not arise. That is the cultural-cognitive pillar doing its work.

When I put these frameworks next to what happened after ChatGPT, the pattern is hard to miss. The mimetic pressure was enormous. Nobody knew what generative AI meant for their industry, and under uncertainty, organizations copy perceived leaders. consulting firms published white papers about AI transformation, and their clients adopted the recommendations because the uncertainty was too high to sit still. That mimetic rush made every "AI strategy" deck look the same because the organizations were not independently reasoning toward AI. They were responding to the pressure to look legitimate in a field where everyone else had already announced their AI plans.

The coercive pressure followed. The EU AI Act, GDPR enforcement actions, and sector-specific regulations started pushing organizations toward certain AI governance structures. That is coerced convergence, and it is not necessarily bad. Some regulation protects people. But it means organizations adopt AI governance frameworks because they have to, not because they reasoned their way to good governance. The gap between external compliance and internal commitment is where bad implementations live.

The normative pressure is building. Industry associations are publishing AI ethics standards. Professional certifications for AI governance are emerging. Conferences have entire tracks on responsible AI. Over time, these professional norms will make certain AI practices feel like the only reasonable way to do things, and organizations will adopt them without asking whether they fit the context. That is when isomorphism stops being about learning and starts being about conformity.

Tornatzky and Fleischer (1990) gave us a framework for thinking about where these pressures land. The TOE framework says technology adoption is shaped by three contexts: the technological context (what is available and what the firm already has), the organizational context (firm size, management support, slack resources), and the environmental context (competition, regulations, vendor ecosystems). TOE is useful because it tells you where to look. But my professor was very clear about one thing: TOE is a bucket, not a theory. It identifies contextual factors but specifies no causal mechanisms, no directionality, and no testable propositions. You cannot explain adoption with TOE alone. You pair it with institutional theory for why, and with structuration or affordance theory for what happens after. TOE maps the terrain. Institutional theory explains why everyone on that terrain is walking the same direction.

I keep thinking about what this means for organizations claiming AI transformation. When I read the institutional theory literature alongside what I wrote about digital transformation not being the same as digitization, the diagnosis sharpens. An organization that adopts AI because competitors are adopting AI, or because a regulation demands certain AI disclosures, or because every CTO with the same certification is recommending the same vendor, is not transforming. It is converging. Transformation requires changing the business model and value proposition. Isomorphism produces organizational convergence without organizational change. The technology arrives. The practices stay the same. The result is what you would expect: expensive AI wrapped around processes that were never rethought.

The uncomfortable question is whether some of this convergence is rational under uncertainty. DiMaggio and Powell were not saying isomorphism is always irrational. They were saying it is driven by legitimacy-seeking, not efficiency-seeking, and those two logics produce different outcomes. An organization that adopts AI for legitimacy will look like it has an AI strategy. An organization that adopts AI for efficiency will have one. The problem is that legitimacy-seeking and efficiency-seeking are hard to distinguish from the outside, because mimetic adoption produces the visible artifacts of strategy, press releases, product pages, conference keynotes, without the underlying transformation that would make those artifacts meaningful.

The cultural-cognitive pillar makes this harder to notice, not easier. Once "we use AI" becomes the taken-for-granted baseline for legitimacy in an industry, the question shifts from "should we adopt AI?" to "which AI vendor?" The first question was the strategic one. The second question is a procurement decision disguised as strategy. Scott's cultural-cognitive pillar explains why nobody goes back to the first question once the baseline is set. It is not that organizations forget to ask. It is that the question no longer makes sense within the institutional frame.

I find institutional theory sharp for explaining why adoption happens and limited for explaining what happens next. It tells you why organizations converge. It does not tell you why some organizations use the technology well and others do not. That is where absorptive capacity and dynamic capabilities come in, and I think connecting institutional pressure to organizational ability to exploit what pressure pushed them toward is where the interesting research lives. Why the same AI tool thrives in one organization and dies in another is not a question institutional theory can answer alone. As I have written before when discussing why use is the wrong construct for agentic systems, adoption and effective use are different things.

Organizations will keep copying each other. That is what organizations do under uncertainty, and institutional theory predicts it cleanly. But if the copy does not come with the capacity to learn, adapt, and transform routines, the mimetic isomorphism produces convergence without competence. The companies that look like they have an AI strategy and the companies that actually have one are following the same mimetic script. Only some of them are writing their own lines after the script runs out.