Generating Variations

Overview#

Sometimes the hardest part of producing something isn’t the execution — it’s choosing a direction. You need a tagline, a pitch, an email subject line, a section heading, a way to frame a recommendation. You have a rough idea of what you want to say but not the best way to say it. So you write one version, stare at it, wonder if there’s something better, and either stick with it out of inertia or start second-guessing.

This playbook uses AI to generate the options you’d never get to on your own. Instead of agonizing over one version, you produce ten or twenty variations quickly, then use your judgment to pick the strongest or combine elements from several. AI is good at generating volume and exploring angles. You’re good at recognizing what works.

The trap is generating variations without knowing what you’re looking for. Twenty options with no evaluation criteria just gives you a harder decision. Pair this with a rubric and it gets much sharper.

When to Use#

You need a tagline, headline, or subject line and the first version you wrote is fine but not great.
You’re framing a pitch or recommendation and you’re not sure which angle will land with your audience.
You have messaging that needs to work across contexts and you want to explore different tones, lengths, or framings.
You’re stuck between two approaches and want more options before committing.
You’re naming something — a project, a product, an initiative, a document. Naming is hard and volume helps.

The Play#

Define what you’re varying and what’s fixed#

Before generating anything, get clear on constraints. What stays the same across all variations? What’s open?

If you’re generating tagline variations, the core message might be fixed but tone, length, and framing are open. If you’re generating pitch variations, the audience and the ask might be fixed but the hook and supporting argument are open.

Being explicit about constraints keeps the variations useful rather than scattered. It also forces you to articulate what you care about before you start, which sometimes clarifies the direction on its own.

Give the AI rich context and ask for volume#

Don’t just say “give me ten taglines for X.” Give it the context to produce variations that are actually relevant. What is this for? Who’s the audience? What tone? What’s been tried before? What are you reacting against?

Then ask for volume. Ten to twenty is a good starting range. You’re not looking for ten perfect options — you’re looking for a spread that covers different angles, tones, and approaches. Some will be bad. The point is to see possibilities you wouldn’t have reached yourself.

If the first batch is too similar, push harder. “These are all variations on the same angle — give me ten more that take a completely different approach.” Or get specific: “Five under ten words, five that lead with a question, five that are deliberately provocative.”

Evaluate with a rubric, not your gut#

This is where it connects to rubric-driven assessment. With twenty variations in front of you, gut feel alone isn’t reliable. You’ll gravitate toward the one closest to what you already had in mind, which defeats the purpose of generating options.

Build a quick rubric before you evaluate. Doesn’t need to be elaborate. For a tagline, the dimensions might be: clarity (does someone unfamiliar with the context get it?), memorability (would someone remember this an hour later?), tone fit (does it match the voice we want?), specificity (does it say something concrete or could it apply to anything?).

Score each variation against the rubric. The highest-scoring ones might not be the ones you liked on first read — that gap is worth paying attention to. It tells you something about the difference between what appeals to you and what serves the purpose.

Combine and refine#

The best final version often isn’t any single variation — it’s a hybrid. One might have the right tone but run too long. Another might open strong but land weak. A third might frame the idea in a way you hadn’t considered but use language that doesn’t fit.

Pull the best elements from across variations and combine them, then refine. You’re moving from exploration to craft. The AI generated raw material; you’re shaping it.

If you want, run the refined version through a secondary AI review to catch any AI patterns that crept into the language.

Save what you didn’t use#

Some of the variations you don’t pick now might work later. A tagline that’s wrong for the website might land in a presentation. A pitch framing that misses for this audience might be right for a different one. Keep the full list somewhere accessible — it’s a bank of ideas you’ve already done the work to generate.

Rubric Driven Assessment — Score variations against a rubric rather than picking on gut feel. This is what makes the process rigorous rather than just brainstorming.
AI Driven Writing — If you’re stuck on framing early in the writing process, generate variations of your thesis or key messages before committing to an outline.
Audience Translation — Generate variations targeted at different audiences from the same core message, then evaluate which framing works best for each.
Secondary AI Review — Clean up the final version after combining and refining from multiple variations.

Examples#

Examples coming soon.