I use the FORGE method with Claude to turn rough ideas into finished work — here's how it works

There's a version of AI that most people use every day. They type their vague ideas into the chat box and hope to get back something complete. If you're using AI in this way, you're not getting the most out of it. Not to mention, the final results are obviously AI; they feel hollow. Your words might be there, but your thinking isn't.

I can't stress enough that it's time to stop using Claude (or any chatbot) as a writing machine and instead lean into it as a thinking partner. It's why I created something I call FORGE, a method I landed on after months of trial and error to solve this problem.

My method is a five-stage framework that takes an idea from raw and half-formed to something you'd actually be proud to put your name on. It works for articles, presentations, pitches, reports — anything where the difference between a mediocre output and a great one is the quality of the thinking behind it.

Here's how it works and how it helps you write with authenticity.

What is the FORGE method?

I made this up, so stick with me. FORGE stands for: Frame, Obstruct, Reconstruct, Generate, Edit.

Each stage has a specific job so don't skip stages and definitely don't rush to finish them. The whole point is that you're not asking Claude to do your work — you're using it to do your work better. I believe this is one of those upskills that make you irreplaceable at work.

Claude responds well to explicit structure and pacing controls, which is exactly what FORGE gives it. Instead of one giant prompt that produces one giant response you either accept or reject, you're having five focused conversations that build on each other. The output at the end feels like yours because it is yours — Claude just helped you find it.

Stage 1: Frame

The first stage of this framework forces you to articulate what you actually want to say before you start saying it. If you're like me with ideas that seem to move faster than you can articulate, this is a total game changer.

Most ideas fail not because they're bad, but because they were never properly defined. I once had a professor tell me that "we limit what we do if we can't describe it." This powerful rule is something I still think about, even with AI at the ready.

If you've ever started by sitting down to write something and halfway through you realize you don't actually know what your point is, the Frame stage fixes that. It saves me hours.

The prompt: "I have an idea I want to develop. Here it is: [describe your rough idea in a few sentences, as messy as you like]. Before we do anything with it, I want you to ask me three questions that would help clarify what I'm actually trying to say. Don't suggest anything yet — just ask questions."

Interestingly enough, Claude might ask you questions anyway. It's different than other chatbots because it seeks clarity rather than people pleasing. But the reason you ask for questions in this way, rather than waiting for the chatbot to offer suggestions, is critical.

If you ask Claude to develop your idea straight away, it will — but it will develop its own version of your idea, not yours. That sort of thing gives me the ick because no matter how smart AI gets, I refuse to outsource my thinking. But, questions force you to do the thinking. By the time you've answered three good questions about your idea, you've already sharpened it more than most people do in an entire first draft.

Stage 2: Obstruct

In the Obstruct stage, you'll explore holes in your thinking before your readers or audience do. This is the stage most people skip, and it's why most first drafts are weak. You know what you're trying to say, so you naturally can't see the gaps in the argument. Claude can.

The prompt: "Here is my idea, now clarified: [paste your sharpened idea from Stage 1]. I want you to be a smart, skeptical reader who is not trying to be kind. Tell me: what's the weakest part of this argument, what am I assuming that I haven't proved, and what would someone who strongly disagrees say? Don't soften it."

I purposely added "Don't soften it" to this prompt because that phrase in particular does a lot of work. Claude responds well to clear direction and does well with specific constraints — and in this case the constraint is honest criticism. If you choose to do this with ChatGPT, you'll appreciate that this cancels the polite feedback it typically offers. If Claude, or your preferred chatbot, gives you three genuinely uncomfortable observations about your idea, Stage 2 has worked.

Stage 3: Reconstruct

Once you've fixed known weaknesses, this is where you rebuild your idea. At this stage, you take everything you learned from Stage 2 and rebuild. Note, this is not about getting to the writing part yet — it's about strengthening the architecture of your argument so that the writing phase isn't constantly fighting against structural problems.

The prompt: "Based on the weaknesses you identified, help me reconstruct this idea so it's stronger. Specifically: how should I reframe the central claim to make it harder to argue with, what evidence or examples would I need to support it properly, and what's the most logical structure for presenting it? Give me a skeleton, not a draft."

I repeate, a skeleton, not a draft. That instruction keeps Claude from running ahead into Stage 4 before you're ready. Building workflows by connecting multiple prompts — where each prompt builds on the previous output — is one of the most effective ways to get sophisticated results from Claude. This chatbot in particular thrives with a sequence.

Stage 4: Generate

We've finally arrived at the stage where the actual first draft is produced. But we've put ourselves in a position of strength because we've thought rigorously about our idea before diving in. If you get to this stage via the other three stages, you've done something most writers never do.

The draft you get here will be different in quality from anything Claude produces cold, because it's working from a foundation you've already tested and strengthened.

The prompt: "Now I want a structured outline. Use the skeleton from Stage 3 as your structure. Write in a [describe your tone — conversational, authoritative, warm, direct] voice. The audience is [describe them]. I'll refine it in the next stage."

What Claude gives you will be an outline that you can then flesh out with your own writing. What I like about using Claude for the outline is that it covers all the bases that you may not have covered alone. Again, AI is here to help, not do the work for you. You'll discover the results are far more sophisticated and clean.

Stage 5: Edit

This is the stage when you takethe draft you wrote from the outline and turn it into something genuinely good. For me, this is where FORGE separates itself from every basic "write this for me" prompt. You're not asking Claude to polish — you're asking it to interrogate.

The prompt: "Here is my draft: [paste it]. I want you to do a three-pass edit. First pass: find anything that's vague, generic or sounds like it could have been written by anyone — flag it and suggest something more specific. Second pass: find the single strongest sentence in the piece and tell me why it works. Third pass: tell me what's missing — the thing a reader would finish this and wish I'd included. Then give me a final version with your first-pass changes made."

The three-pass structure is deliberate. A practical revision routine keeps the process organized and ensures the final output feels increasingly like your own work. You do not want Claude to edit heavily. I like this stage because it can help you pinpoint that most powerful sentences in your piece. This is helpful if you enter contests or work with an agent; these circumstances frequently ask for "your favorite sentence" or "favorite message."

Asking Claude to identify your strongest sentence does something specific: it teaches you which of your instincts are working, so you can trust them more next time.

The takeaway

With the threat of "AI coming for white collar jobs," knowing how to use it properly will not only enhance your workflow, but your ability to work alongside it. And, anyone could agree that human + AI is unstoppable and far better than just AI.

FORGE works because it keeps you in the driver's seat at every stage. Claude is doing the analysis, the pressure-testing, the structure-building and the outlining — but the decisions about what you actually believe, what angle you want to take and what voice you want to use are always yours.

Claude Sonnet 4.6 — the free default model on claude.ai — has seen significant improvements in consistency, instruction following, and long-horizon planning. The first time I ran a piece through all five stages, it took about 40 minutes. The work I produced in those 40 minutes was better than anything I'd written in twice the time using a single prompt. Not because Claude is smarter than me — but because FORGE made me think harder than I would have on my own. Give it a try and let me know what you think in the comments.

Follow Tom's Guide on Google News and add us as a preferred source to get our up-to-date news, analysis, and reviews in your feeds.

More from Tom's Guide

• I use the 'reverse brief' prompt to instantly understand any document — here’s how it works
• I use the 'pre-mortem' prompt before every big decision — it's stopped me making 3 huge mistakes
• Gemini 3.1 Pro is a powerhouse for deep work — here are 7 prompts that prove it