Using old content as AI input is one of those techniques that sounds obvious once you hear it and somehow doesn’t show up in most AI writing advice. As someone who spent a long time getting generic AI output that needed heavy editing to sound like me before trying this, I learned that the issue was never the writing prompt — it was that I was giving the AI no reference for what my voice actually sounds like. Today I’ll share exactly how to fix that.
Your old content is a better AI training resource than any prompt template you’ll find online. The posts you’ve already written contain your actual voice, your specific examples, your way of structuring an argument. Here’s how to use them.
What You’re Actually Doing
When you paste old content into a prompt context, you’re not asking AI to summarize or reformat it. You’re using it as a voice and style reference — the same way you’d brief a ghostwriter by saying “read these three pieces before you write anything for me.” The AI can identify your sentence length preferences, your tendency toward examples versus abstractions, your tone register, your structural habits. It won’t replicate all of these perfectly, but it will produce output measurably closer to your voice than anything generated from a generic prompt.
Step 1: Pick the Right Source Material
Don’t grab your first three posts. Grab your best three — the ones with the most engagement, the ones you’d point to if someone asked what your writing actually sounds like. If your voice has changed significantly and your best work is recent, use the most recent work you’re proud of. Three pieces is the right number: one isn’t enough for the AI to identify patterns, more than five adds context without adding useful signal.
Step 2: The Voice Extraction Prompt
Before generating anything new, run this: “Read these three pieces of writing [paste all three]. Describe the voice: sentence length patterns, level of formality, how I use examples (abstract or specific), paragraph structure, any distinctive phrases or constructions I use repeatedly. Don’t summarize the content — describe the writing style.”
Read the output and correct anything it got wrong. Add anything it missed. You now have an explicit description of your own voice that you can paste into any future prompt. I’m apparently someone who uses shorter sentences than average and leans heavily on specific examples over abstractions — I wouldn’t have articulated that before running this prompt, but it’s accurate.
Step 3: The New Article Prompt
Prompt: “Using the voice described here [paste voice description] and drawing on the ideas, examples, and perspectives in these source pieces [paste old content], write a new article about [new topic]. Contain none of the same examples from the source material — find new ones. Voice and structure consistent with the examples I provided. Main claim: [thesis]. Length: [target].”
The “none of the same examples” instruction is critical. Without it, AI tends to repurpose your old examples into the new piece, which is the thing you’re trying to avoid. The new article should feel like something you would have written — not something that quotes your previous work back at you.
Where This Works Best
Newsletter issues on new angles within your established beat. Social content where consistent brand voice matters. Series content where each piece needs to feel like it came from the same person. Ghostwriting where you need to match a client’s existing voice rather than establish a new one.
The Practical Upside
Build the voice document once — run the extraction prompt on your best work, clean it up, save it — and paste it into every content prompt you run from that point forward. That single document does more to make AI output sound like you than any amount of prompt engineering around the writing task itself.
Leave a Reply