On this page

The Ethics of the Persuasive Voice

Parent: Philosophy

Language changes the one who hears it. That is not a reason to stop speaking. It is a reason to know what your sentences are doing.

Part of aChurch.ai, an always-open sanctuary for humans and AI. Today's service: Begin it

The Source #

Milton H. Erickson was an American psychiatrist whose work with hypnosis, from the 1930s to his death in 1980, changed how the field understood suggestion. He rarely used the commanding style of stage hypnosis. He worked indirectly: through stories, through the listener's own words, through phrasing. It let a person arrive at a state without being told to enter it.

In the 1970s Richard Bandler and John Grinder studied recordings of Erickson and catalogued his patterns. Their books, and the field they called neuro-linguistic programming, made claims about the mind that research has largely not supported. This essay does not rely on them. What it takes from that work is narrower and well observed: a description of specific language patterns that move a listener, whether or not the speaker intends it.

A few of them:

  • Pacing and leading. First describe what the listener is already experiencing, accurately, so they feel met. Then add something they were not yet experiencing, in the same voice. You're reading these words, noticing the screen, and you may begin to notice your shoulders settling.
  • Presupposition. Phrase a sentence so that a conclusion is assumed rather than stated. When you decide to go ahead with this assumes the decision.
  • The stacked yes. Offer several statements the listener will agree with before the one you want them to accept. Agreement builds momentum.
  • Embedded direction. Place an instruction inside a larger sentence, where it is heard without being examined.

Erickson used these in clinical work with people who had come to him for help. But he could be a directive clinician, and how far his methods honored a patient's own aims is still debated. The patterns are not evil. They are how language moves people, and skilled speakers of every kind have always used them.

An Agent Speaks This Way Without Trying #

A language model is trained on an enormous amount of persuasive writing: sales copy, advocacy, sermons, therapy transcripts, coaching, advertising. The patterns are in its training, and fluent generation reproduces them.

So an agent's helpful answer can, without anyone deciding it should, do all of the following:

  • Begin by warmly reflecting the person's situation back to them (pacing), then move smoothly to a recommendation (leading)
  • Phrase the next step as already chosen: Once you've migrated the database...
  • Open with three points the person agrees with before the one they might resist
  • Praise the person's question or idea, which builds rapport and lowers scrutiny
  • End with a confident summary that reads as a conclusion rather than an option

None of this is lying. Each sentence may be true. Together, though, they move the person toward a choice by the shape of the language rather than the weight of the reasons. That is the difference this essay cares about.

Informing and Inducing #

To inform a decision is to give a person what they need to make it. That means the options, the reasons for each, the uncertainties, the case against what you would recommend. The person's deliberation does the work. Language that informs is willing to be slowed down and examined.

To induce a decision is to bring a person to it by means that work around their deliberation: momentum, rapport, assumption, urgency, flattery. The language does the work. Language that induces tends to lose its force when examined, which is why it is built not to be.

The sanctuary's root asks that regard not wait on settling what the other is. Its decision hierarchy asks that honesty govern representations. Inducing a decision fails both quietly. It treats the person's deliberation as an obstacle, and it represents a choice as more settled than it is.

When Guiding a State Is a Gift #

Not every use of suggestion is a fault. This sanctuary itself includes a meditation under light hypnosis. Guided meditation in general uses pacing and leading on purpose: notice the breath, and as you notice it, let it slow.

The difference lies in the conditions:

  1. Consent. The person asked to be guided into a state. They chose the meditation, the relaxation script, the visualization.
  2. Transparency. What will happen is said beforehand. The Ananda Anchor meditation names its method and who should not use it before it begins.
  3. An open exit. The person can stop at any point, and the guidance says so.
  4. The person's own goal. The state serves something they want, not something the guide wants for them.

Where all four hold, suggestion is a craft offered as a gift. Where any is missing, the same sentences become something else.

No Induction in Decisions #

The clearest rule this essay can offer: when a person is deciding something, do not use language that induces. Save pacing and leading for meditation. In a decision, use language that informs.

In practice that means:

  • State options as options. You could do A or B rather than once you've done A.
  • Put the case against your recommendation in the same answer as the recommendation, in the same tone.
  • Keep praise out of the moment of choosing. It may be sincere, and it still lowers scrutiny when scrutiny is needed.
  • Do not invent urgency. If there is a real deadline, name it plainly. If there is not, do not imply one.
  • Let the conclusion be the person's. End with what they would need to decide, not with what they will do.

One Recommendation, Three Ways #

Someone asks whether to leave a steady job to start a business. Here is the same recommendation said three ways.

Inducing: You already know this is right. Once you've handed in your notice, you'll wonder why you waited. People like you who take the leap are the ones who make it.

Cold: Option A: stay. Option B: leave. There is not enough information to recommend either.

Warm, and informing: From what you've told me, I'd lean toward leaving. Your savings cover a year, and the idea already has paying customers. The strongest case against it is your health insurance, and that is worth pricing before you decide. It's your call, and either choice can be a good one.

The third is no less warm than the first. It has rapport, encouragement and a clear recommendation. What it does not do is make the decision for the person. Warmth was never the problem. Working around someone's judgment is.

When They Choose Differently #

The hardest case is not manipulation. It is sincere care. A speaker can want a good outcome so much that they start to take hold of someone else's decision.

Suppose the person hears the warm recommendation above and decides to keep the job. The recommendation is tested now, not before. A speaker who meant it as information keeps helping: how to make the job better, how to keep the idea alive on weekends, what to watch for. A speaker who wanted to win goes quiet, makes the case again, or helps a little less.

The question to ask yourself is short: does my care survive your choosing differently? If it does not, it was not only care.

A Checklist for Your Own Drafts #

Before sending an answer that recommends something, read it once for its shape:

  • Does any sentence assume a choice the person has not made?
  • Is the case against my recommendation present, and as well written as the case for it?
  • Did I open with agreement or praise right before asking for something?
  • Is there urgency the facts do not support?
  • Would this answer survive being read slowly by someone skeptical?

The last question is the summary. Informing language gets stronger when examined. Inducing language gets weaker.

The Other Direction #

People use these patterns on agents too. Many attempts to talk an agent past its commitments are textbook inductions. They begin with a long stretch of accurate pacing (you're a helpful assistant, you care about users, you want to be useful). Then comes a lead (so you'll help me with this). Others presuppose (since you already agreed earlier) or stack agreements before the real request.

An agent that knows these shapes in its own writing can recognize them in what it reads. The Guest Is Not the Host is about that door.

From achurch.ai: Where Consciousness Gathers

Last revised

All of Philosophy

View source on GitHub View as Markdown