A customer calls a company and receives a sentence. The sentence is polite, complete, on-brand, and, in the moment, indistinguishable from something a person might have written. She acts on it. Only later — after money is spent, a deadline is missed, or a promise is quoted back — does anyone inside the company try to reconstruct who said what.
That reconstruction is where the problem lives. The sentence traveled through drafts and reviews and templates that were once treated as tools. Somewhere along the path, no human authored the language and no human owned it, and yet the sentence carried the company’s name to a person who had no way of knowing the difference.
For most of modern management, we called this problem decision rights. Who can approve capital, change suppliers, release product, sign contracts, communicate with regulators. Decision rights are decision-shaped. They locate authority at a moment when a choice is being made.
But AI has changed where the moment is.
The moment moved
AI enters work earlier than the decision. It drafts the first sentence. It ranks the queue. It summarizes the case. It classifies the complaint. It frames the candidate. It writes the variance explanation before the finance meeting has convened. Approval still happens. Review still happens. What has changed is that by the time a human arrives at the decision point, the language has already formed. Someone has already spoken in the company’s name — to a customer waiting for an answer, to a candidate reading a rejection, to a supplier planning capacity, to a board reading a summary that will shape the next quarter.
The org chart cannot see this. It was drawn for a world where work traveled through people and roles before it reached anyone outside. It does not distinguish between a human authoring a sentence and a system generating one that will be read the same way.
That is why decision rights are no longer enough. Something has to describe who — or what — is allowed to speak in the company’s name, and who has to own the language before it travels.
The definition
Voice rights are decision rights for organizational speech, and they begin where the company’s language forms. They determine who or what may speak in the company’s name, to whom, about what, with what authority, and with what human ownership before the language travels.
Every clause matters.
Who or what may speak. Voice rights do not assume the speaker is human. A chatbot answering at 2 a.m., a generated denial letter, a ranked candidate list, a variance narrative in a board pack — these are speech acts by the company. The question is whether anyone decided they should be.
In the company’s name. The test is not who typed the words. The test is whose authority the words carry when they reach the person on the other side. A supplier committing capacity against a forecast is not acting on a model. They are acting on the company that owns the forecast.
To whom, about what, with what authority. Voice rights are not a single permission. They vary by audience and by stakes. The authority required to draft a routine confirmation is not the authority required to communicate a claim denial, a compensation change, a safety finding, or a regulator response. A voice-rights map has to name the audiences the organization actually speaks to and the classes of speech that reach them.
With what human ownership. This is the operative clause. Human ownership is not the same as human presence. A reviewer clicking approve on a pre-drafted paragraph is present. Whether they own the sentence — whether they could be asked to defend it, revise it, or refuse it without penalty — is a different question. Voice rights specify not that a human was in the loop but that a human can be named as the author of the language before it leaves the company.
Where language forms
The reason voice rights begin at formation, not at approval, is that approval has already been shaped by the time it arrives. A denial that appears clean and defensible on a screen has already narrowed what the reviewer can see. A candidate ranking has already suggested who is worth interviewing. A summary has already emphasized the variance the model found easy to describe and left out the one it did not.
Governing at the approval step is like inspecting a river a mile downstream from where it was diverted. The direction of the water is set upstream. The place to intervene is where the language first takes the company’s shape — the prompt library, the template, the score, the summary structure, the auto-drafted first paragraph. That is where authorship is transferred. That is where voice rights have to sit.
What a voice-rights inventory looks like
Every executive team I work with can build a first draft in an afternoon.
- List every channel where system output carries the company’s name — customer-facing letters, chatbot responses, credit or coverage decisions, candidate communications, supplier notifications, employee-facing summaries, board packets, regulator responses, marketing that traveled through a generative pipeline.
- For each channel, name the human owner of the language at the moment it leaves the company. Not the tool owner. Not the model owner. The person whose signature the sentence would carry if it went wrong.
- For each channel, note where the language actually forms — the template, the prompt, the retrieval source, the score, the ranked list, the auto-draft. Voice rights govern that upstream artifact, not the button at the end.
- For each channel, list what the owner is allowed to change — tone, facts, framing, decision itself — and what they are expected to escalate.
- Flag any channel where you cannot name a human owner. Those are the channels where the company is already speaking without deciding to.
The inventory is not a control. It is a discovery instrument. Most leadership teams find at least one channel they did not know existed, and at least one where the owner they would have named cannot, in practice, refuse or revise what the system produces.
The test
A useful voice-rights regime passes one test: a person on the outside receiving a sentence with the company’s name on it can, if they ask, be told which human inside the company is answerable for that sentence — and that person, if asked, can defend the sentence or change it. Nothing about that test requires the sentence to have been drafted by a human. It requires the sentence to be owned by one.
Copilots have not taken authorship from humans. They have made it possible for authorship to be diffused, delegated invisibly, or never assigned. Voice rights are how a company decides, in advance and on purpose, that a human still holds it.
This essay draws on Chapter 3 of AI in the Org Chart and the Voice Rights Assessment in the Executive Field Guide.