Key takeaways
- Chat is over-used as an AI default because it's easy to ship, not because it serves users.
- A blank chat box hides capabilities, mangles structured tasks, and makes errors painful to recover from.
- Reserve chat for open-ended, exploratory work — wrap it in suggestions and editable outputs even then.
- Match the interface to the task: forms and tables for structured work, chat as an escape hatch.
- Interface choice drives adoption, completion, and trust — the metrics that justify AI investment.
Chat has become the reflexive answer to the question "how should we put AI in our product?" It's fast to build, easy to demo, and it echoes the ChatGPT experience everyone already recognises. But convenient and correct are not the same thing. For a large share of real business use cases, a blank chat box is the wrong interface — and the cost shows up later as low adoption, frustrated users, and features nobody trusts.
At Evocative we've shipped enough AI-powered products to notice a pattern: teams reach for chat because it hides hard design decisions, not because it serves the user. That's the trap worth unpacking.
The seductive logic of the chat box
A text field promises infinite flexibility. Whatever the user needs, they can just type it. From a product roadmap that looks like a feature that does everything, built in a sprint. In practice, infinite flexibility becomes infinite ambiguity. The user faces a blinking cursor and a silent question: what is this thing actually good at?
Chat over-simplifies the interface while pushing all the complexity onto the person least equipped to handle it. Instead of the product guiding the user, the user has to guess the product's capabilities, phrase requests correctly, and recover from failures with no visible path. That's not simplicity — it's abdication.
Why chat is the wrong default for most workflows
The core problem is that chat is a linear, memoryless-feeling, low-affordance medium being asked to do a high-affordance job. Consider where it breaks:
- Discoverability collapses. A menu shows you your options. A chat box shows you nothing. Users can't act on capabilities they don't know exist.
- Structured tasks fight the format. Comparing three suppliers, adjusting a budget, filtering a catalogue — these are inherently spatial, tabular tasks that a scroll of messages mangles.
- Error recovery is painful. When the AI misunderstands, the user re-types the whole request instead of nudging a single parameter.
- State is invisible. Complex agent work involves steps, sources, and decisions. A wall of text hides all of it, so trust erodes.
The science-fiction ideal doesn't help here. The talking computers of Star Trek and the chatty droids of Star Wars trained a generation to imagine AI as pure conversation. But even those fantasies quietly assumed rich context, shared physical space, and a machine that already knew what mattered. Strip that away and you're left with a search bar that talks back.
When chat actually is the right call
To be fair, conversation earns its place in specific conditions:
- The task is genuinely open-ended and exploratory — brainstorming, drafting, research.
- The input space is too large to enumerate as controls.
- The user already knows what they want and can express it precisely in language.
Even then, the best AI products wrap chat in structure rather than shipping it raw. Suggested prompts, editable outputs, inline actions, and visible reasoning turn a void into a guided surface.
Designing the right interface instead
The better question isn't "chat or no chat" — it's "what is the shape of this task, and what interface matches it?" A few principles we apply:
- Make capabilities visible. Buttons, templates, and starting points beat a blank prompt every time. Show users the door before asking them to describe it.
- Match the medium to the task. Structured work wants forms, tables, and canvases. Let AI populate and edit these directly rather than narrating changes.
- Keep humans in the loop with judgment, not typing. Surface the AI's reasoning and let users approve, tweak, or reject with a click. Bottle your judgment into the interface itself.
- Blend, don't replace. The strongest pattern is often a hybrid: a structured UI for the 80% of predictable tasks, with a conversational escape hatch for the messy 20%.
The business case for getting this right
Interface choice is not cosmetic. It directly drives adoption, task completion, and trust — the three metrics that decide whether an AI feature pays back its build cost. A chat box that demos beautifully but confuses real users produces impressive engagement numbers for a week and abandonment thereafter. A well-structured interface that quietly does the job earns daily use.
Chat isn't dead. It's just wildly overused as a default because it's easy to ship, not because it's right for the user. The teams that win with AI will be the ones who resist the reflex, study the actual task, and design the interface it deserves.
Sources
- Designers are sharpening knives for the wrong fight — UX Collective
- Chat is the wrong interface for AI — UX Collective
- What Star Trek got wrong about AI (so far) — UX Collective
- Bottle your judgment, Chat is the wrong UI, Design.md — UX Collective
- What Star Wars got wrong about AI (so far) — UX Collective
