What Do Grice's Conversational Maxims Have to Do With Design?

Before conversation design was a thing, there was a philosopher trying to explain why people understand each other at all: “Why does talking work?”.

In 1975, Paul Grice proposed that our conversations work because we assume the other person is trying to be cooperative. From that single assumption, he derived four maxims that govern almost every exchange we have:

The four maxims

• Quantity. Say as much as is needed, and no more. Give too little and the listener has to guess. Give too much and you bury the point, or worse, imply that the extra detail matters when it doesn't.
   
• Quality. Say what you believe to be true, and don't claim things you can't back up. This one seems obvious until you notice how often a product, or a person, hedges everything into vagueness to avoid being wrong, which breaks trust just as badly as being wrong does.
   
• Relation. Be relevant. This is the maxim that gets violated constantly in bad UX copy and bad chatbot design alike: a response that's technically true and technically complete, but answers a question nobody asked.
   
• Manner. Be clear. Avoid ambiguity, avoid unnecessary complexity, and get to the point in an order that makes sense. Grice included this one almost as a catch-all for everything that makes language hard to parse even when it's honest, sufficient, and relevant.

Not just linguistics

What strikes me about Grice's maxims is that they were never meant as rules to follow. They're a description of what listeners already assume is happening. When someone reads a piece of interface copy, or a chatbot response, or an error message, they're not evaluating it against some abstract standard of good writing. They're checking it, often unconsciously, against these four expectations. The moment one of them breaks, the person feels it, even if they couldn't name why.

That reframes a lot of the research and writing problems I run into. A confusing form field usually isn't a manner violation in the grammatical sense, it's a quantity violation, too little context for the decision being asked of the user. 

A chatbot that answers with a wall of text to be sure it covered everything the user could mean isn't being careful, it's violating relation by burying the actual answer under everything adjacent to it.

Where this gets interesting with AI

Language models are, in a sense, trained on an enormous amount of cooperative conversation, which means they've absorbed a rough approximation of these maxims by default. But approximation isn't adherence. I've seen plenty of AI-generated responses that are technically accurate and still violate quantity by over-explaining, or violate relation by answering a nearby question instead of the one asked.

This is where I think conversation design earns its place as a discipline rather than a subset of copywriting. Designing a good AI response isn't just about making sure the model has the right information. It's about deciding, deliberately, how much to say, how confidently to say it, how tightly to tie it to what was asked, and how to order it so it becomes crystal clear. 

Those are Grice's four maxims, just applied to a system instead of a person – we want AI to converse like a person, not be a person... but that is a different topic about the Uncanny Valley.

A useful diagnostic, not a rulebook

I don't treat the maxims as a checklist to enforce mechanically. Grice himself pointed out that people violate these maxims constantly, and on purpose, to create meaning: sarcasm, irony, understatement, all of it depends on a deliberate maxim violation that the listener is expected to notice and interpret. The goal isn't zero violations. The goal is knowing when a violation is intentional and meaningful, versus when it's just sloppy design that leaves the user working harder than they should have to.

When a piece of writing, a flow, or an AI response feels off and I can't immediately say why, running it against these four questions almost always tells me where the problem exists.