How to Develop a Character with AI: Step-by-Step Workflow (2026)

1) Quick Answer: How to Develop a Character with AI

The 2026 workflow for developing a character with AI has seven steps: lock the five load-bearing fields (want, need, contradiction, voice, red lines), generate the want and need with AI, invent the defining contradiction yourself, draft the voice with AI stress-testing, set the red lines, pressure-test against the plot, and run the character in a sample scene. Each step has an AI prompt pattern that works and a structural decision the writer (not the AI) must make. The full first-draft character takes about a day; the pressure-test takes another day.

Characters are not biographies. They are contradictions in motion. AI can scaffold the biographical fields (name, age, role, backstory) faster than you can type them. What AI cannot do is invent the specific contradiction that makes the character a character. The judgment is yours. The time savings are AI's.

I have built character profiles for documentary subjects (real people who already exist) and consulted with novelists shifting from biographical character work to contradiction-first work. The pattern that holds: AI accelerates the five load-bearing fields, leaving the contradiction-invention work for the writer.

For the definitional pillar, see The 12 Best Tools for Character Development in 2026 (We Tested Them All). For the related bible workflow, see How to Build a Story Bible with AI.

2) Why Character Development with AI is Different in 2026

The 2010 character development workflow was a Word document with twenty biographical fields: name, age, hair color, favorite food, defining childhood memory. Writers filled them all in, then wondered why the character felt flat. The fields were filler; none of them decided what the character does on page 47.

The 2026 workflow strips the fields back to five that actually matter, and uses AI to scaffold them. Want is the external goal. Need is the internal goal. Contradiction is the defining tension. Voice is how the character talks. Red lines are what the character would never do. These five decide everything the character does.

The shift in 2026 is that AI handles the first-draft generation of all five fields except contradiction. It is competent at biographical filler (it has read every character description on the internet) and at proposing voice patterns and red lines, which it derives from the want and need. The contradiction is where the character becomes a character, and the contradiction is where AI fails. AI tends to propose generic contradictions ("brave but afraid", "kind but selfish") that read as tropes. The writer's job is to invent the specific contradiction that makes this character not a trope. The AI does 70% of the work in 30% of the time; the 30% the writer does is the part that matters.

3) Prerequisites: What You Need Before You Start

Three things to gather before opening the canvas or AI tool.

1. A premise or logline. What is the story about? One sentence: protagonist, goal, obstacle. The character development workflow needs at least a premise; without one, the AI will generate characters that fit no story.

2. The character's role in the plot. Is this the protagonist, antagonist, ally, mentor, foil? Role shapes what the character needs to be. Protagonists need want-and-need with a contradiction strong enough to carry the story. Antagonists need a want that opposes the protagonist's want with credible logic. Allies and mentors need voice and red lines more than full arcs.

3. A canvas-based workspace with AI. Tools like Storyflow are built for this pattern: the AI reads everything on your active board (plot beats, theme notes, the other characters) plus up to one Tactic and three documents you @-mention, so a character develops in the context of the story around it rather than in an empty chat window. Storyflow's Story Blueprints library (200+ templates on Plus, Pro, and Max, from $7.99/mo on the annual plan) includes character profile templates pre-structured around the five load-bearing fields; the free plan covers the canvas and basic AI, so you can build the five fields by hand first. Try Storyflow. Where it is the wrong choice: if you only need to develop a single character in isolation with no surrounding story, ChatGPT alone is sufficient.

With those three in place, the character development workflow takes about a day for a first draft and another day for pressure-testing.

4) Step 1: Lock the Five Load-Bearing Fields

Before generating any character content, lock the framework: want, need, contradiction, voice, red lines. These are the only fields that matter for what the character does in the story.

Biographical fields (name, age, hair color, favorite food, parents' professions) are filler unless they directly bear on the want, need, or contradiction. The mistake new writers make is filling in 20 biographical fields before locking the five load-bearing ones.

Want. What the character wants externally. Concrete, observable, with a clear success condition. "Rescue her daughter." "Cook meth to provide for his family." "Win the case before her firm folds." Not "find herself" or "learn to love" (those are themes).

Need. What the character needs internally. The growth the story will force on them. "Learn to trust her own judgment." "Accept that he is not the protagonist of his own life." "Forgive the mentor who failed her." The need is often hidden from the character; the story reveals it.

Contradiction. The defining tension that makes the character specifically interesting. "She is the most empathetic detective in her precinct, and she cannot read social cues at all." "He is a man of God who has never genuinely prayed." The contradiction is what AI cannot invent.

Voice. How the character talks. Specific verbal patterns, vocabulary choices, rhythm. The reader should be able to identify the character from a single sentence of dialogue.

Red lines. What the character would never do, no matter the cost. The line they will not cross. The thing that, if they did it, would mean they are no longer themselves.

Open a card on the canvas for each of the five fields, and leave them empty. The next steps fill them.

5) Step 2: Generate the Want and the Need with AI

AI is competent at scaffolding the want and need from a premise. Use a prompt pattern that names the role and the genre.

AI prompt pattern for want and need:

"I am developing the protagonist of a [genre] story. The premise is [logline]. The character is the [role]. Generate 5 possible external wants (concrete, observable, with a clear success condition) and 5 possible internal needs (the growth the story will force on the character). For each, name why this want or need fits the premise."

The AI will return ten options. Most will be competent but not exciting; one or two will surprise you. Pick the want that creates the strongest story. Pick the need that makes the want hard to achieve. The two should be in productive tension: the character wants something, but to get it they need to grow in a direction they resist.

For the antagonist, repeat the prompt but ask for wants that oppose the protagonist's want. The antagonist's want should make narrative sense from inside their own logic, not just be evil for the protagonist to fight.

Where AI fails on this step: it proposes wants that fit the genre but not your specific story. A thriller AI defaults to "find the killer". The writer's job is to find the want unique to this premise ("find the killer who killed her own brother, who was also the killer's brother") that makes the story specifically interesting.

6) Step 3: Invent the Defining Contradiction Yourself

This is the step AI cannot do. The contradiction is what makes the character a character.

Working contradictions are specific and load-bearing. They are not "brave but afraid" (every character is). They are "she is brave only when no one is watching her, and the story requires her to be brave in front of an audience". Specific. Constraining. Story-generating.

To invent the contradiction, ask:

1. What does the character want? (From Step 2.)
2. What does the character believe about themselves? Usually a story-friendly self-image: I am the responsible one, I am the rebel, I am the protector.
3. What is the gap between the self-image and the reality? The gap is the contradiction.

Examples of working contradictions:

• A federal marshal who believes she is the protector of her family, and who has been lying to them for ten years about what she actually does at work.
• A chemistry teacher who believes he is the good father providing for his family, and who is becoming the criminal who endangers it.

The contradiction must be load-bearing. If it can be resolved in one scene with no consequences, it is not load-bearing.

Use AI to stress-test the contradiction. Prompt: "Here is the contradiction I have for my character: [contradiction]. What three story situations would force this contradiction to surface? What three story situations would let the character avoid the contradiction?" The AI's answers reveal whether the contradiction is producing story or hiding from it.

7) Step 4: Draft the Voice with AI Stress-Testing

The voice is how the character talks. AI is competent at drafting voice patterns from the want, need, and contradiction.

AI prompt pattern for voice:

"Here are the five fields for my character: want [X], need [Y], contradiction [Z], with their role as [role] in a [genre] story. Generate 10 sample lines of dialogue in this character's voice. The lines should reflect the want, need, and contradiction. Include three angry lines, three vulnerable lines, three neutral lines, and one comic line."

The AI will produce ten lines. Some will be generic; some will surface a specific verbal mannerism. Look for the mannerism the AI happens to discover. A repeated word, a specific syntactic habit, a tic. Once you find it, hardcode it as part of the voice: "she uses 'I suppose' as a hedge", "he says 'right' three times in a row when he is uncertain", "she breaks her sentences in the middle when she is lying".

For dialogue stress-testing, chat with the character in three scenarios: a high-stakes confrontation, a quiet emotional moment, and a comic exchange. The voice that survives all three is load-bearing.

Where AI fails on this step: AI-generated voice tends to sound like other characters' voices from the AI's training data. The writer's specific verbal choices break that pattern. The mannerism is the move.

8) Step 5: Set the Red Lines

Red lines are what the character would never do. The list is short (usually three to five items) but load-bearing. Whenever you are tempted to have the character break one, you either revise the red line or revise the scene.

AI prompt pattern for red lines:

"Given this character's want, need, contradiction, and voice, what are 5 specific actions the character would never take, no matter the cost? Frame them as concrete behaviors, not abstractions."

The AI will propose five. Some will be obvious ("would never kill an innocent person"). Some will be specific and useful ("would never lie to her sister"). The writer's job is to pick the three to five that produce the most story, and to add the show-specific red lines the AI cannot generate.

The most important red line is usually the one the story will eventually force the character to cross. "She would never lie to her sister" creates a story when the plot forces her to lie to her sister. The red line is not the thing the character does not do; it is the line the story will test.

9) Step 6: Pressure-Test the Character Against the Plot

Before the character is ready, pressure-test the character against the plot.

Test 1: Want-plot fit. Does the protagonist's want actually drive them through the plot's events? If the plot would happen with or without this character's want, the want is not load-bearing.

Test 2: Need-arc fit. Does the protagonist's need require them to change in a way that resolves the plot? If the character can stay the same and the plot still resolves, the need is decorative.

Test 3: Contradiction-story fit. Does the contradiction produce at least three story situations? If the contradiction never forces the character into a tough choice, the contradiction is not load-bearing.

Test 4: Voice consistency. Read three sample lines of dialogue. Do they sound like the same person? If not, the voice has not landed yet.

Test 5: Red line consequence. Does at least one red line get crossed (or nearly crossed) during the story? If no red line is ever tested, the red lines are not load-bearing.

AI prompt pattern for pressure-testing:

"Here are the five fields for my character: [paste them]. Here is the plot outline: [paste it]. Run the five tests above. For each test, name whether the character passes, fails, or partially passes. Where the character fails or partially passes, suggest what to revise."

The AI will surface problems faster than the writer can. Revise based on the AI's surfacing. The revision is the writer's; the surfacing is the AI's.

10) Step 7: Run the Character in a Sample Scene

The final test is to write a scene with the character in it. The scene reveals whether the character is ready or whether more development is needed.

Pick a high-stakes scene from the middle of the plot. Write 500 to 1,000 words with the character in it, then read it. Three things to look for:

Voice. Do the character's lines sound like the same person? If they drift, the voice needs more work.

Consistency with the contradiction. Does the character's behavior in the scene reflect their contradiction? If the character acts in a way that resolves the contradiction too easily, the contradiction needs to be tightened.

Surprise. Does the character do anything in the scene that surprises the writer? Strong characters surprise their writers. If the scene reads exactly as the writer planned, the character is still flat.

Use AI to critique the scene. Prompt: "Here is the character: [paste the five fields]. Here is the scene: [paste it]. Does the character's behavior fit the five fields? Where does the scene drift? Where does the character surprise you?" The AI's critique catches drift the writer missed.

Once the scene works, the character is ready. It will keep evolving as the story runs, but the five fields are locked enough to support the manuscript.

Worked Example: Developing Sofia from Scratch

Here is one character taken through all five fields, start to finish, with the AI prompts and the decisions the AI could not make.

The premise. A hospice nurse spends her career helping strangers say the things they never said before they die. Then she is assigned to care for the estranged father who walked out when she was nine. Role: protagonist. Genre: grounded drama. That is the entire brief before the AI touches anything.

Step 1, lock the fields. Five empty cards on the canvas: want, need, contradiction, voice, red lines. No biography yet. No hair color, no birthday, no favorite song.

Step 2, want and need with AI. The prompt asked for five external wants and five internal needs from the premise. The AI returned ten. Most were filler ("wants closure", "needs to heal"). One want landed: get through the six-week assignment and transfer out before he dies, so no one at the agency ever learns he is her father. Concrete, observable, hard success condition. The need the writer paired it with: stop mistaking control for care. She wants to leave; she needs to stay present. The two fight each other, which is the point. The AI generated ten options in thirty seconds; choosing the two that fight each other took a human who knew the story. The judgment is yours. The time savings are AI's.

Step 3, invent the contradiction (the writer's job). Asked, the AI proposed "compassionate but closed off", a trope. The writer threw it out and built the real one from the self-image gap: Sofia is the most trusted end-of-life nurse in the county, known for helping families finish unfinished business, and she has not said one true sentence to her own father in nineteen years. She finishes everyone's business but her own. Specific, constraining, scene-generating. AI stress-test prompt: "What three situations force this contradiction to surface?" The answers (a family asks her how to forgive; her father tries to apologize; a colleague praises her gift for closure) confirmed the contradiction produces story.

Step 4, voice with AI stress-testing. The prompt asked for ten lines across angry, vulnerable, neutral, and comic registers. Nine were generic. The tenth surfaced a keeper: Sofia says "we" when she means "you" ("we're going to feel some pressure now") to hold clinical distance, and the tic cracks the moment a question gets personal and she goes silent mid-sentence. The writer hardcoded both. The AI found the tic; the writer decided it was load-bearing.

Step 5, red lines. AI proposed five; two survived selection. She would never lie to a patient about their prognosis, and she would never let a patient die alone. The second is the one the plot is built to test.

Step 6, pressure-test. Run the five tests against the outline. Want-plot fit passes: her want to leave drives every avoidance beat. Contradiction-story fit passes: three forced situations already exist. Red line consequence: the outline drives toward her father dying alone unless she breaks her own transfer plan and stays. The red line and the want collide in the climax, exactly where they should.

Step 7, sample scene. Eight hundred words of the midpoint: her father, lucid for an hour, asks her to help him write a letter to "a daughter I lost". She does not tell him who she is. She switches into "we". Then he asks a question that lands too close and she stops mid-sentence. The surprise the writer did not plan: Sofia finishes the letter anyway, in her own words, and signs his name. The character did something the outline never specified, which is the signal the five fields are holding. Sofia is ready to carry a manuscript.

Where Storyflow Is the Wrong Tool for Character Work

Every honest workflow names where its tools stop. Storyflow is strong at developing and organizing the character with AI context on a canvas, and there are three cases where it is the wrong choice.

It is a development canvas, not a prose or screenplay editor. Storyflow is where you draft the five fields, map the arc, and keep the profile beside the plot. It is not where you write formatted screenplay pages or a polished manuscript. For the actual writing, pair it with Final Draft or Highland for screenplays, or Scrivener or a word processor for prose.

It is cloud-only. There is no local-first or offline mode. Writers under NDA, in regulated contexts, or who simply prefer their character bible on their own disk should keep sensitive work in a local tool.

It is newer, and the AI reads the current board, not your entire series. The Story Blueprints library is broad, but character-specific template depth for narrow genre conventions is thinner than in decades-old craft software, so some frameworks you will build yourself. Because the AI reads your active board plus up to three @-mentioned documents rather than a whole multi-book series at once, series continuity means segmenting by board rather than asking one question across ten books.

12) The Bottom Line

Developing a character with AI in 2026 is a seven-step workflow: lock the five load-bearing fields, generate want and need with AI, invent the contradiction yourself, draft the voice with AI stress-testing, set the red lines, pressure-test against the plot, and run the character in a sample scene. Each step has an AI prompt pattern that works and a structural decision only the writer can make.

The judgment is yours. The time savings are AI's. AI does 70% of the work in 30% of the time. The 30% the writer does is the part that matters: the contradiction that makes the character not a trope.

The strongest 2026 workflow uses Storyflow's canvas for character profile scaffolding with story context, AI for option generation at each step, and writer revision for the actual decisions. Try Storyflow: build the five fields on the free plan, then add the Story Blueprints library on Plus ($7.99/mo on the annual plan) when you want the character templates.

The most useful exercise this week: develop a single supporting character from a story you are working on, using the seven-step workflow. It does not need to be a protagonist. The workflow teaches you what the load-bearing fields look like.

13) Author

Justkay Documentary Filmmaker and Founder of Storyflow

Justkay has built character profiles for documentary subjects (real people who already exist) and consulted with novelists shifting from biographical character work to contradiction-first work. The Five Load-Bearing Fields framework came from watching writers fill in 20 biographical fields before any of them decided what the character does in a scene. The workflow above reflects what working writers actually do in 2026.