What Does Show, Don't Tell Mean to an AI — and Can It Actually Do It?

If you've ever typed a prompt into an AI writing tool and watched it produce sentences like "She felt devastated" or "He was extremely angry," you already understand the core frustration behind ai show dont tell — the gap between what we ask artificial intelligence to do and what it actually delivers. Show, don't tell is one of the most fundamental principles in literary fiction, and yet it's also one of the hardest things to get an AI to execute consistently.

This isn't a small craft quibble. For indie fiction authors who care deeply about the texture of their prose, this problem cuts right to the heart of whether AI can be a genuine creative collaborator or just a fast, sloppy first-draft machine. The answer is more nuanced — and more hopeful — than most writing communities tend to acknowledge. Let's get into it.

What "Show, Don't Tell" Actually Means — and Why It's Hard to Define

Before we talk about what AI does with this principle, it's worth being precise about what the principle actually demands. "Show, don't tell" is one of those phrases that gets repeated so often it starts to lose its edges. Writers hear it in workshops, read it in craft books, see it scrawled on manuscript feedback — but the instruction is genuinely more complex than it first appears.

At its simplest, telling delivers information directly: "Maria was nervous." Showing renders that nervousness through observable, sensory detail: the way Maria's thumbnail finds the corner of her cuticle, the way she re-reads the same line of the email four times. The reader infers the emotion; the writer never names it.

But this is where most explanations stop, and they shouldn't. Because showing is not just about replacing emotion words with physical gestures. It's about trusting the reader to do interpretive work. It's about understanding that specificity creates credibility — that a particular detail convinces us in a way that a label never can. Cormac McCarthy doesn't tell you the road is bleak; he shows you a landscape so perfectly rendered in ash and silence that bleakness becomes the only conclusion available.

The Three Layers of Showing

It helps to think of showing as operating on three distinct levels:

• Sensory grounding: What does the character see, hear, smell, feel, taste? Physical reality anchors the reader.
• Behavioral specificity: What does the character do — not broadly, but precisely? Not "she paced" but "she walked to the window, turned, walked back, and stopped with her hand on the kettle she didn't actually want to boil."
• Implication and restraint: What is deliberately not said? The white space around the action is part of the showing. Hemingway built an entire iceberg theory on this idea.

When we ask AI to show rather than tell, we're asking it to operate on all three levels simultaneously — and to do so with the kind of specificity that feels earned rather than generated. That's a tall order.

Why AI Defaults to Telling: The Statistical Problem

Here's something that reframes the whole conversation: AI doesn't default to telling because it lacks imagination. It defaults to telling because telling is what most text looks like.

Large language models are trained on vast corpora of human writing. That writing includes literary fiction, yes — but it also includes fan fiction, web content, how-to articles, Wikipedia entries, and millions of pages of functional prose where emotional clarity matters more than aesthetic restraint. When you ask an AI to describe a character's grief, it reaches for the most statistically probable response. And statistically, most writers say "she felt grief" or "he was overcome with sadness" because most writers — especially developing ones — are still learning to show.

The AI isn't failing craft. It's reflecting the average of the writing it learned from. This is a crucial distinction because it means the problem isn't inherent to AI — it's a defaults problem. And defaults can be overridden.

The Compounding Problem: Vague Character Data

There's a second reason AI prose tends to tell rather than show, and it's one that indie authors have direct control over: insufficient character specificity. Showing requires particularity. You cannot write a specific physical gesture if you don't know what kind of person would make that gesture. You cannot render meaningful behavioral detail without knowing your character's history, nervous habits, and emotional triggers.

When an author prompts an AI with nothing more than "write a scene where Jenna finds out her brother has been lying to her," the AI has no raw material for specificity. It defaults to generic emotion tags because that's all the information it has. Feed the AI a full character portrait — Jenna's childhood, her relationship with touch, the specific way she goes cold when betrayed rather than hot — and the resulting prose has a fighting chance at real showing.

This is exactly why tools that maintain persistent character context matter. If you've looked into the why does ai keep changing your character's personality mid-story? problem, you'll recognize this pattern: without anchored character data, AI doesn't just forget who your character is — it flattens them into the nearest archetype, which produces prose that tells rather than shows every single time.

The AI Show Don't Tell Problem in Practice: Before and After

Let's make this concrete. Here's the kind of output you typically get when you ask an AI to write an emotional scene without sufficient context or craft instruction:

Before (AI default — telling):

Daniel felt a deep sense of loss as he stood in his father's study. He was very sad and overwhelmed by grief. Everything reminded him of his father and he couldn't stop thinking about all the things he'd never said. It was extremely painful to be in that room.

This is prose that tells us what to feel rather than giving us the experience of feeling it. Now here's the same scene prompted with specific character detail, behavioral instruction, and an explicit craft directive to avoid naming emotions directly:

After (AI with context and craft constraint):

Daniel picked up the letter opener — his father had used it to open bills, only bills, always bills, as though correspondence that mattered deserved to be torn open by hand. He turned it over. Set it down. Picked it up again. The desk lamp was still angled toward the chair where his father had sat, and he couldn't make himself adjust it. Outside, a car backed out of a driveway. Someone's life, continuing.

That second passage isn't perfect literary fiction — but it's doing real work. It's grounding us in physical specificity, using a repeated behavior to imply psychological state, and ending on an image rather than an emotion label. The AI produced it. The author directed it.

How to Actually Get AI to Show Rather Than Tell

This is where we move from diagnosis to craft. There are several concrete techniques that consistently improve AI output in this area.

1. Replace Emotion Words in Your Prompt

If your prompt contains words like "sad," "angry," "happy," or "anxious," you are almost guaranteeing that those words will appear in the output. Instead, describe the situation and the character's behavioral tendencies — never the emotion itself. Let the AI work from the outside in.

Instead of: "Write a scene where Clara feels betrayed by her best friend."

Try: "Write a scene where Clara has just discovered that her best friend told her ex-husband where she was living. Clara tends to become very precise and formal when she is hurt — she starts organizing things, speaking in complete sentences, refusing small courtesies like offers of tea. She does not cry in front of people."

2. Anchor Scenes in Sensory Specificity

Tell the AI exactly what environment the character inhabits and what they can perceive. Vague settings produce vague prose. Specific settings give the AI something to work with — and sensory detail is the foundation of all genuine showing.

3. Use Structural Constraints

Explicitly forbid certain moves. "Do not name any emotion directly" is a powerful constraint. So is "Do not use adverbs to describe how speech is delivered." These negative instructions shape output as meaningfully as positive ones.

4. Give the AI Your Character's Voice and History

The more the AI knows about your character's specific past — not their personality traits in general, but the events that shaped them — the more it can draw on particular, earned detail rather than generic gesture. A story codex that captures each character's history, speech patterns, and behavioral tendencies gives AI writing tools the raw material they need to produce prose that actually shows rather than tells.

5. Ask for Revision, Not Generation

One of the most effective approaches is to generate a raw scene and then prompt specifically for showing revisions. "Rewrite this passage so that every emotional state is communicated through physical action, dialogue subtext, or environmental detail — never through direct naming." Revision prompts often outperform generation prompts because they give the AI a draft to work from.

What AI Does Surprisingly Well — and Where It Still Falls Short

In fairness to the technology, there are areas where AI-assisted showing has genuine strengths. AI tools are often quite good at generating sensory detail when prompted carefully. They can produce atmospheric description that grounds a scene physically. They can iterate quickly on subtext — offering multiple versions of the same piece of dialogue at different registers of concealment and implication.

Where they continue to struggle is in what we might call earned restraint — the kind of showing that works precisely because of everything the author has built before it. When Kazuo Ishiguro's Stevens in The Remains of the Day refuses, again and again, to name what he feels for Miss Kenton, the restraint is devastating because we've lived in his repressed interiority for 250 pages. An AI producing that scene cold, without the weight of the novel behind it, would have no mechanism for that devastation. The showing depends on the architecture. AI can execute individual techniques; it cannot yet build the long structural showing that the best literary fiction depends on.

This is worth naming honestly — not to discourage you from using AI tools, but to clarify the collaboration correctly. AI is excellent at the sentence-level execution of craft techniques when given the right constraints. The larger architecture — the pacing of revelation, the cumulative weight of withheld information — remains the author's work.

If you're evaluating which tools genuinely support this kind of craft-aware collaboration, it's worth researching the best AI writing tools for fiction with an eye toward which ones allow you to maintain persistent character and story context across sessions, since that context is the foundation of any real showing capability.

The Role of Canon and Consistency in Showing

There's a structural issue that often undermines showing in AI-generated fiction that doesn't get discussed enough: inconsistency erodes specificity. If an AI forgets that your character has a prosthetic hand, it cannot write the particular physical gestures that belong to someone who navigates the world that way. If it forgets that your setting has no electricity, it cannot render the sensory texture of candlelight and shadow that would make a scene feel real.

Good canon enforcement for AI-generated scenes is not just a continuity issue — it's a craft issue. Showing depends on specificity, and specificity depends on the AI knowing and respecting the established facts of your world and your characters. Every inconsistency is a missed showing opportunity.

ProseEngine approaches this by keeping your story's canon, character profiles, and world rules accessible to the AI throughout the writing session — which means the physical and behavioral specifics that make showing possible are actually available when it matters. It's the difference between prompting a stranger and prompting someone who has read your whole manuscript.

For authors wondering about the practical and financial side of integrating these kinds of tools into their process, it's worth understanding what AI novel writing software costs at different feature levels, since the tools with persistent context features tend to sit in a different pricing tier than basic generation tools.