The first answer AI gives you is the average answer. Iteration is how you turn it into your answer.

Most people use AI like a vending machine: type a request, take what comes out, move on. The result is competent, polished and forgettable. It sounds like everything else, because in a real sense it is everything else, averaged together.

The research backs this up. A 2026 study found that people who simply anchored on AI drafts produced more uniform writing. Those who were pushed toward originality did something different: they edited, revised and re-prompted with more specific preferences, and their final work came out noticeably more diverse, close to what people wrote without AI at all (arXiv, 2026).

The tool was the same. The process was what changed. This article is about that process.

Why first outputs are generic

AI language models work by predicting the most likely next word, based on enormous amounts of text. Ask a broad question and you get the most probable response: the safe middle of everything it has seen. That's not a bug. Without more to go on, the middle is its best guess.

Three things make the pull toward the average stronger:

  • Missing context. The AI doesn't know your audience, your voice, your industry or your opinion. It fills those gaps with defaults.
  • Tuned to please. Models are trained to give answers people rate well, and broadly acceptable answers rate well. Researchers found that writing with a feedback-tuned model made people's essays more alike than writing with an older base model or without AI (Padmakumar & He, 2023).
  • Everyone asks the same way. When millions of people send similar prompts, they get similar outputs. A well-known 2024 study by Doshi and Hauser found AI ideas made individual stories better but made the group's stories more alike (Culture Futurist).

So the first draft isn't the finish line. It's a starting position: the generic baseline you now get to push away from.

The iteration loop

Good AI work runs in rounds, not in one shot. Each round has four steps, and you repeat them until the output sounds like you and says something only you would say.

  1. Draft. Get a first version on the table, from the AI, from you, or both. Don't judge it yet; it's raw material.
  2. Critique. Read it as a demanding editor. What's vague? What's cliché? What could appear on any other site? Mark the exact lines.
  3. Direct. Give specific feedback: what to cut, what to keep, what to change and why. Add the context it was missing: an example, a real number, your opinion.
  4. Refine. Get the revised version and compare it to the last. Keep what improved, then start the loop again on what's still generic.

Most work gets noticeably better in two or three rounds. The key is step 2: the loop only works if you're the one deciding what "better" means.

Iteration moves that work

You don't need a perfect prompt. You need good follow-ups. Here are six moves that reliably pull output away from boilerplate.

1. Point at the exact problem. "Make it better" gives the AI nothing to work with. Quote the weak line and say what's wrong with it.

  • "The second paragraph sounds like a press release. Rewrite it the way you'd explain it to a friend over coffee."

2. Inject what only you know. Your stories, data, clients and opinions are the one thing the AI can't average out.

  • "Add this example from my own work: [your story]. Use it to anchor the main point."

3. Ask for range, then choose. One answer is the middle. Several answers show you the edges.

  • "Give me five very different angles on this, including one that's contrarian. Don't play it safe."

4. Ban the clichés. Name the phrases and patterns you're tired of seeing.

  • "No 'in today's fast-paced world,' no 'game-changer,' no lists of three adjectives. Rewrite."

5. Show, don't describe. A sample of your writing teaches voice faster than any adjective.

  • "Here are two paragraphs I wrote. Match this voice: sentence length, humor, directness."

6. Make it critique itself. Turn the AI into its own harsh editor before you read the next draft.

  • "What's the most generic part of this draft? Fix it, then tell me what you changed."

Mistakes that keep you stuck in boilerplate

  • Accepting the first draft. It's the most generic version you'll ever get. Shipping it means publishing the average.
  • Starting over instead of refining. Hitting "regenerate" over and over just rolls the dice again. Specific feedback on one draft beats five fresh attempts.
  • Vague feedback. "More engaging" or "more professional" pushes the AI toward a different generic, not toward you.
  • Letting the AI hold the pen the whole time. Rewrite key sentences yourself. Your edits are the signal that shapes the next round.
  • Stopping when it's "fine." Fine is the enemy. The question isn't "Is this correct?" but "Could anyone else have written this?"

Treat AI like a collaborator, not a vending machine

A vending machine gives everyone the same thing. A collaborator gets better the more you work together. The difference isn't the tool. It's whether you stay in the conversation after the first answer.

Iteration also does something for you: every round of critique forces you to decide what you actually think, what you value and what "good" means for your work. The output gets more original because you get more clear.

So next time the first draft lands, don't ask "Is this good enough?" Ask "What would make this unmistakably mine?" Then go one more round.

Sources

Ideas mine, English polished with AI. How I use AI →