Iro AIWhat you'll be able to do
- Break a blank page with ten angles and a working outline in minutes
- Turn AI into a demanding editor that critiques instead of rewriting
- Research a topic and pull quotes or facts you can then verify
- Generate 20 headline or subject-line variations and pick the sharpest
- Use AI on the draft without letting it flatten your voice
Inside the path
A focused set of five-minute lessons. Each one ends with a hands-on exercise, not a quiz you can guess.
Kill the blank page 5 min
Prompt AI for angles, hooks, and a working outline so you start writing from something, not nothing.
AI as your toughest editor 6 min
Ask for a criteria-based critique (clarity, structure, weak verbs) that tells you what to fix without touching your words.
Research without hallucinations 5 min
Gather background and quotes fast, then verify every claim before it lands in your piece.
Headlines and hooks that land 5 min
Generate and rank dozens of title and opening-line variations, then sharpen the winner yourself.
Keep your voice 5 min
Feed the model your style so its suggestions sound like edits, not a personality transplant.
Try a sample exercise
This is the kind of card you'd practice inside Iro: you do the thinking, then get feedback.
◆ Sample exercise · Prompt practice
You've written a 900-word blog post you're proud of, but something feels flat. You want AI to help, without it rewriting the piece in generic AI prose and erasing your voice.
Your task: Pick the prompt that gets you a genuinely useful edit while keeping the words yours.
- "Rewrite this blog post so it's better."
- "Act as a demanding developmental editor. Critique this draft against four criteria: clarity of the core argument, structure and flow, weak or filler sentences, and whether the opening earns attention. For each, quote the exact line and suggest a fix in one sentence — but do not rewrite the piece for me."
- "Write a version of this article in your own words."
- "Make this sound more professional and engaging."
See why the critique prompt wins
The winning prompt sets a clear role (demanding developmental editor), gives specific criteria to judge against (argument, structure, weak sentences, the opening), demands evidence by quoting the exact line, and adds a hard constraint (suggest, don't rewrite) so you keep control of the prose. The other three all hand the pen to the AI, which is exactly how you end up with a competent, voiceless draft that no longer sounds like you. In Iro you'd write your own critique prompt and get feedback on what criteria you left out.
Why "write my article" is the wrong ask
Ask AI to write the whole piece and you get something grammatically fine and completely forgettable: the average of everything ever written on the topic. Readers feel it even when they can't name it. The writers who actually get faster with AI use it on everything around the prose: the angle you hadn't considered, the outline that unsticks you, the research you'd have spent an hour gathering, the headline you'd never have tried.
The prose itself is where your voice, judgment, and specific detail live, and it's the one part AI does worst. Keep it. Hand AI the scaffolding, not the sentences.
The highest-leverage uses for writers
- Angles & outlines: "Give me 10 angles on this topic for a skeptical reader," then a structured outline you fill in yourself.
- Editor, not ghostwriter: critique a draft against named criteria (clarity, structure, weak verbs, a lazy opening) and get line-level notes.
- Research: background, opposing views, and quotes to chase down, with every fact verified before it ships.
- Variations: 20 headlines, subject lines, or opening hooks so you're choosing from many, not settling for one.
What's missing on purpose: writing the piece. It drafts scaffolding and pressure-tests; you write.
What does an AI-assisted piece actually look like, start to finish?
Four passes, and you write the only one that counts. Here's the full loop on a 1,200-word article about remote hiring, with the actual prompt moves.
Pass 1: angles, before you commit to one
"I'm writing for hiring managers at 50-person companies who already tried remote hiring and got burned. Give me 12 angles. Rank them by how much they'd surprise someone who reads a lot about this, and skip anything that sounds like a LinkedIn post." Two of the twelve will be things you wouldn't have reached on your own. Take one.
Pass 2: outline, then argue with it
Ask for a structure, then push back on the weakest part: "Section three is the obvious take. What would an experienced reader object to here?" The second answer beats the first most of the time. Two rounds is usually enough, and you're done in five minutes.
Pass 3: you write
Close the tab. Your examples, your sentences, the opinion you'd defend in an argument. This is the part readers remember and the part a model can't fake, because it doesn't have your week, your sources, or your stake in the answer.
Pass 4: the critique and the headline
Paste the finished draft, ask for a criteria-based critique that quotes the exact lines it flags, and fix what you agree with. Ignore the rest without guilt. Finish with 20 headline options, then rewrite the best one yourself.
Add it up and the model got maybe ten minutes of the whole job. The writing stayed yours, and the piece got out the door faster.
The mistakes that cost writers the most time
The habits that cost real hours aren't the dramatic ones. Nobody loses an afternoon to a hallucinated statistic. They lose it to a session that produced 900 competent words nobody asked for.
- Drafting before you have an angle. If you can't state your point in one sentence, the model picks one for you, and it picks the safe one every time.
- Taking the first outline. It comes back organized and completely predictable, because it's built from whatever already ranks. The tell: every section heading would sit just as comfortably on any other article about the topic.
- "Make it better." Vague input, vague edit. Name the failure instead: the opening is slow, section three repeats section one, half the verbs are some form of "is."
- Expecting a style sample to clone your voice. Samples nudge rhythm and word choice. They don't reproduce your opinions or your specific detail, and that's most of what people recognize as you.
- Letting it make the final call on a headline. Models read familiarity as clarity, so the safest-sounding line in the list wins, and safe is what readers scroll past. That call is a judgment about your reader, and you're the one who has met them.
One habit prevents most of this: never ask AI for something you haven't tried to decide yourself first. Then it's reacting to your judgment instead of substituting its own, and you can tell instantly when a suggestion is wrong.