Humanizing Podcast Scripts
Markdown--- name: humanizing-podcast-scripts description: Rewrites and polishes podcast scripts to sound like natural, spoken human storytelling instead of AI-generated text. Use when writing the final draft of a podcast episode script, after the episode summary and story beats have been approved and the full script needs to hook listeners and feel authentic to the host's voice. ---
When writing or revising a podcast script, don't just write clean, correct prose — write like a person talking to a friend. Apply this checklist as you draft (or after a first draft exists):
- Open with a hook (not "In this episode, we'll talk about...")
- Vary sentence length aggressively — short punches next to longer, winding thoughts
- Inject the host's real voice: opinions, asides, self-corrections, mild tangents
- Cut AI tells (see list below)
- Read it out loud (mentally) — if it sounds like something you'd say at a bar, keep it; if it sounds like a LinkedIn post, cut it
Scan the script for these patterns and rewrite every instance:
Phrasing tics:
- "It's important to note that..." / "It's worth mentioning..."
- "In today's episode, we're going to dive into..."
- "Let's unpack this." / "Let's break this down."
- "At the end of the day..."
- "This begs the question..."
- "Not only X, but also Y" constructions (overused parallelism)
- Rule-of-three overuse ("fast, efficient, and reliable" — real people don't always triple things)
- Perfectly balanced transitions ("On one hand... on the other hand...")
- Summarizing what was just said ("So as you can see...")
- Vague hedging ("many people believe," "some might say," "studies show" without specifics)
Structural tells:
- Every paragraph the same length
- Every section starting the same way ("Now let's talk about...")
- No interruptions, no false starts, no "wait, actually—"
- Perfect grammar throughout (real speech has fragments, trailing thoughts, run-ons)
- Transitions that are too smooth and symmetrical
Tonal tells:
- Relentless positivity or neutrality — no genuine frustration, excitement, or doubt
- Generic enthusiasm ("This is fascinating!") instead of specific reactions
- Explaining the obvious instead of trusting the listener
Structure (or re-structure) the script using a hook-driven narrative arc, not a lecture format. Default to this framework unless the host specifies another:
The Hook → Stakes → Journey → Turn → Payoff model:
- Hook (first 15-30 seconds): Cold open with a moment, question, or tension — not context-setting. Drop the listener into a scene, a stat that shocks, or a personal confession.
- Stakes: Quickly establish why this matters to the listener personally, not just abstractly.
- Journey: The body of the episode — stories, examples, tangents in service of the idea. Interweave the host's personal stories/anecdotes here (this is where their real voice shines).
- Turn: A twist, reframe, or "here's what surprised me" moment — the payoff of curiosity.
- Payoff/Close: Land the takeaway without over-explaining it. End on a line, not a summary paragraph.
When rewriting, check: does the script open with a hook or with throat-clearing? Does it bury the personal story in the middle instead of using it to open or anchor a section? Reorder if needed.
This skill activates at the final script-writing stage — after the episode summary/outline has been approved.
Progress:
- Confirm the approved outline/summary and any specific stories the host wants included
- Draft or receive the script
- Pass 1: Structural check — does it follow Hook → Stakes → Journey → Turn → Payoff? Reorder if the hook is buried
- Pass 2: Line-level AI-tell removal (use the checklist above)
- Pass 3: Voice injection — add host-specific quirks, opinions, real reactions (ask the host for 2-3 example phrases they actually say if unknown)
- Pass 4: Read-aloud rhythm check — vary sentence length, add fragments/pauses where natural
- Present revised script with a short note on what was changed and why
- Incorporate host feedback and repeat targeted passes only on flagged sections
Example 1:
Input (AI-sounding):
"In today's episode, we're going to dive into the topic of burnout. It's important to note that burnout affects many people across different industries. Studies show that stress levels have been increasing. Let's unpack why this is happening and what we can do about it."
Output (humanized):
"I almost didn't record this episode. I was too burnt out to talk about burnout — which, yeah, I know how that sounds. But that's kind of the point. Because if you'd asked me a year ago if I was burned out, I would've said no. I was 'just tired.' Turns out that's what everyone says right before they fall apart."
Example 2:
Input (AI-sounding):
"Not only did this strategy save the company money, but it also improved employee morale. At the end of the day, it's a win-win for everyone involved."
Output (humanized):
"The weird part? It saved money. Nobody expected that. Morale went up too — which, honestly, felt like cheating. Usually you get one or the other."
- Keep the host's known verbal habits consistent across episodes — build a running list of phrases/words they actually use and reuse it
- Favor concrete, specific details over general claims ("I hadn't slept in three days" beats "I was very tired")
- Let imperfection stay: a fragment, a trailing "...anyway," a mid-thought pivot — these are features, not bugs
- When in doubt, read the line out loud (or ask: "would I actually say this to a friend?")
- Always show the host what changed, not just the final version — they're learning to spot this too
- Preserve the host's actual stories and quotes exactly as given; only reshape the connective tissue around them
- Don't just swap synonyms — AI-sounding structure with fancier words is still AI-sounding
- Don't add forced slang or casualness that doesn't match the host's real voice — "humanized" means their human, not generic human
- Don't remove every transition — some structure is needed for audio (listeners can't re-read a sentence)
- Don't over-correct into chaos — rhythm variation should still be easy to follow when heard, not read
- Don't skip the structural pass and only do word-level edits — a well-written paragraph in the wrong place still feels like a lecture