Who owns the prompts a consultant writes for you?

Prompt ownership is the contractual question of who holds, at the end of a consulting or outsourcing engagement, the prompts an outside consultant wrote to solve a client's use case: the client who paid for the work, or the consultant who wrote it and would like to reuse it, in adapted form, with another client. The question concerns the input, the prompt text as potential know-how, not the output an AI generates: that is the subject of generative content copyright due diligence, a distinct problem. A prompt built for an enterprise case, with calibrated examples and multi-step reasoning that reduces error on a specific task, can carry independent economic value: it is exactly the kind of asset Article 98 of Italy's Industrial Property Code protects as a trade secret when it is secret, has value because it is secret, and its holder protects it adequately. But it is not automatically the client's asset, just as consultant-written code is not automatically the client's without a clause.

See also: AgentOps · The observability and lifecycle management discipline for AI agents in production: monitoring, debugging, versioning of prompts and tools.

Why the problem sits exactly in consulting

In the Forward Deployed Engineer delivery model, an outside engineer works inside the client's process and writes the prompts that make an agent work on that specific case: real data, industry terminology, process exceptions. That work looks more like custom software than a generic deliverable, and IT outsourcing contract practice has had a clause on who owns code written by a consultant for decades, precisely because ownership is not obvious without an explicit clause. With prompts the problem is younger, but the structure is identical: whoever commissioned the work assumes it is theirs because they paid for it, whoever wrote it assumes they can reuse their own expertise, and both positions remain reasonable until the contract decides.

What a clause should specify

An operational clause should specify at least three things: whether prompts produced during the engagement are work for hire owned by the client or remain at the consultant's disposal; whether the consultant may reuse, in adapted form, the prompt's structure (not the literal text, which often contains client-specific data or terminology) on other engagements; and what protective measures the client maintains if it claims the prompt as a trade secret, since Article 98 requires adequate measures to keep it secret and a prompt left in a shared repository with no access control does not meet that bar. The 2023 case in which several Samsung engineers pasted internal source code into ChatGPT, later fearing the leak of sensitive data to an outside service, shows the same principle from the opposite side: whoever fails to protect the input it shares with an AI vendor loses control over that information, consulting engagement or not.

An enterprise example

An outside consultant spends six months with a manufacturing company's customer service team building an agent that classifies tickets, and writes a system prompt that embeds the company's fault taxonomy and examples drawn from its historical tickets. At the end of the engagement, the consultant proposes the same setup, with the taxonomy swapped out, to a competitor in the same sector. Without a clause that forbids or explicitly permits this, neither side is obviously wrong: the client funded the work and considers the result part of its competitive edge, the consultant built methodological expertise it considers its own. A clause written before the engagement would have avoided the dispute.

Why it matters for decision makers

For anyone commissioning AI work from a consultant, prompt ownership deserves the same contractual attention code ownership has had for years: write it before the engagement starts, not infer it afterward. Under the Forward Deployed Engineer model, where the prompt is born in direct contact with the client's data, the question is even more concrete: the absence of a clause protects nobody. A prompt tracked and versioned with the discipline of AgentOps at least makes visible who wrote what and when, the minimum condition for discussing ownership at all, before even deciding it.

This entry is informational and does not constitute legal advice: qualifying a prompt as a trade secret and drafting an ownership clause should be handled by your own counsel, case by case.