How to Turn Voice Memos Into Blog Posts (That Still Sound Like You)

Most people explain an idea better out loud than on a blank page. Ask a friend what they think about remote work and you’ll get two lively minutes with an opinion, an example and a joke. Ask them to write 800 words on it and you’ll get a stiff paragraph and a closed laptop.

So speaking a draft makes sense. The trouble starts afterwards. Raw transcripts are messy: half-sentences, “um, so basically”, the same point made three times. And a common complaint about AI voice-to-text tools is the opposite problem: the cleanup is so aggressive that the result reads smoothly but doesn’t sound like you anymore. Your examples are gone, your phrasing has become generic, and you’re left editing someone else’s essay.

The fix is a method that sits between those two failures. Four steps.

Step 1: Talk, don’t write

Speaking a draft is a skill, but a small one. These habits make the biggest difference:

• Talk to one person. Picture a specific reader, a colleague or a friend who asked you the question, and explain it to them. You’ll naturally drop the jargon and the throat-clearing.
• Tell stories and give examples. “Last month a client asked me…” is the stuff that makes a post yours, and it’s exactly what you’d skip when typing. Say it.
• Don’t stop to fix. If a sentence comes out wrong, just say it again, better. Cleaning up a repeated sentence later is easy; recovering a thought you lost while editing is not.
• Say the structure out loud. “New section: why this matters.” “Okay, the second reason is…” These spoken signposts make the transcript much easier to restructure later, whether you do it or an AI does.
• Record several short takes, over a few days. One 30-minute monologue tends to wander. Five five-minute takes, each on one part of the topic, recorded when the thought is fresh (on a walk, after a meeting, while the coffee brews) give you better material and more of it. A second take on Thursday often contains the sharpest version of what you fumbled on Monday.

With SpeakPen you record on your phone (iPhone or Android) or in the browser. Each take becomes its own note with a title, a summary and the full transcript, so a week of takes is easy to scan rather than a pile of audio files you’ll never listen back to.

Step 2: Keep the raw transcript as the source of truth

This is the step most workflows skip, and it’s why voice gets lost.

Whatever you do with AI later, the original transcript is the record of what you actually said, in the order you said it, with your words. Treat it like the master tape: you can make as many edits as you like, but you never record over it. If a rewrite flattens a line you liked, you go back to the transcript and take it.

SpeakPen keeps the full transcript alongside the title and summary on every note, in the language you spoke. Rewriting is optional, and the transcript stays in the note either way.

Step 3: Shape it, with the right tool for the size of the job

One note: the quick first pass

If you recorded a single, reasonably complete take, the built-in rewrite as blog post in SpeakPen is a fast first pass. You get a structured draft in seconds, and because the transcript is still sitting in the note, you can compare the two and pull back any phrasing the rewrite smoothed away.

The same note can also become a LinkedIn post or a tweet. That’s often the right move for a thought that doesn’t need 1,000 words: say it once, post the short version today, and save the long version for later if it keeps coming up.

Several takes: work with an agent

A real article usually comes from three to six takes recorded over a week. That’s a job for an AI agent that can read all of them at once. SpeakPen has a read-only MCP server, so Claude, ChatGPT, Claude Code, Cursor and other MCP clients can search and read your notes (setup is on the Connect page).

The important part is the prompt. Generic “turn this into a blog post” instructions are what produce the generic voice. Tell the agent explicitly what it is not allowed to do:

Search my SpeakPen notes for everything I recorded about [topic]
in the last [2 weeks]. Read the relevant notes in full.

Turn them into a blog post draft of about [1,000] words. Rules:
- Use my phrasing. Keep my sentences wherever they work; only
  smooth a sentence if it's genuinely unreadable as spoken.
- Keep every example, story and number I mention. Do not replace
  them with generic ones.
- Cut only filler, false starts and repetition. When I said the
  same point in several takes, keep the clearest version.
- You may reorder and restructure freely. Use my spoken signposts
  ("new section", "the second reason") as hints.
- Do not add arguments, facts or examples I didn't say. If a
  transition or gap needs new text, write it in [square brackets]
  so I can see exactly what you added.
- After the draft, list: points I made that you left out, and
  places where my takes contradict each other.

The square brackets are the key. They turn the AI’s contribution from invisible to reviewable: everything outside the brackets is something you said, and everything inside is a suggestion you can accept, rewrite or delete.

Step 4: The edit pass is yours

The agent’s draft is a draft. The last pass belongs to you, and the best tool for it is your own voice again: read the draft aloud.

Reading aloud catches exactly the problems this workflow is prone to. A sentence that you’d never say out loud is a sentence the AI wrote, so rewrite it until it sounds like you. A part where you stumble is a part that’s unclear. A paragraph you’re bored reading is one you can cut.

While you’re at it:

• Resolve every [bracketed] addition: keep, rewrite or delete.
• Check the “left out” list. Sometimes the best line is there.
• Go back to the transcript for any spot that feels off. The original is still there in the note.

Where do the topics come from?

Once you’ve done this a couple of times, you’ll notice something: your voice notes are already full of posts you haven’t written. The rant after a frustrating meeting, the explanation you gave a new hire, the idea that you’ve mentioned three weeks in a row.

Ask your agent to dig through the last month of notes for you:

Look through my notes from the last month. Which ideas have I come
back to more than once, or explained at length? Suggest 5 blog post
topics I clearly have opinions on, and list the notes each one
would draw from.

Topics you’ve already talked about are topics you can actually write, because the material exists.

Why this works

Capturing is easy. Using what you captured is the hard part, and it’s why so many note collections turn into a graveyard nobody visits. Turning voice memos into posts is one of the most direct ways to close that loop: the notes stop being a backlog and start being drafts.

If you want the bigger picture of how agents fit in, read From voice note to finished draft. And for the capture side, Capture first, think later covers how to record notes an agent can work with.

Start today: pick one topic you’ve explained out loud to someone this month. Record a five-minute take now, explaining it to that person. Record one more tomorrow. On the third day, run the prompt above and read the draft aloud.