THE IDEA

Start with what you made

A chat was not in the edit. It did not film the take, hear the false start, or notice when the room finally went quiet.

Do not publish unread or ask for another creator’s voice. Your material goes in. A draft comes back. You decide what survives.

WHAT THE LABS FOUND

Editing beats magic

Paste real material. Define the edit. Inspect the answer. Revise again.

BEFORE YOU PASTE

Name the source and the risk

Mara is about to paste a documentary transcript. One speaker shared a private family detail after filming. Another person’s unpublished pitch is in the same file. Mara stops.

Say this before you paste:

• This is mine to use: an outline, transcript, or footage I control.
• This must stay out: private contracts, addresses, passwords, embargoes, and work belonging to someone else.
• These facts are fixed: names, quotes, numbers, dates, links, products, and what appears on screen.
• This is the job: draft, shorten, arrange, compare, or mark cuts.
• This is my final check: I read, hear, or watch it before release.

Use first names or labels when identity is not needed. Follow any platform, client, sponsor, or collaborator rules. Do not ask the model to invent those rules.

The failure here is exposure, not awkward prose. A clever output cannot undo a careless paste.

VOICE PASS

The draft sounds rented

Quinn’s essay has the right facts and the wrong temperature. Every paragraph lands with a polished lesson. She never writes like that.

Then do this. Read the alternatives beside your sample. Keep only what you would say without the screen.

Blind-paste risk: It may replace one generic voice with another.

ROUGH DRAFT

From six beats to a first pass

Bea has six beats for a lunch-packing video. Beat four says only “Wednesday leak.” She needs a draft, not a fabricated kitchen disaster.

Then do this. Rewrite the opening. Fill each gap from memory or leave it out.

Blind-paste risk: A plausible bridge can turn a thin note into a false memory.

TITLE PASS

The title promises too much

Nico fought a scheduling tool on Tuesday. His generated title says it “destroyed his business.” It did not.

Then do this. Read the title beside the ending. If the work cannot pay off the promise, weaken it.

Blind-paste risk: The most clickable option may be the least accurate.

SCRIPT PASS

The script lost your mouth

Imani’s audio outline is solid. The draft has long sentences and radio-host transitions. She trips over them during rehearsal.

Then do this. Read one page standing up. Cut every braced bridge that sounds like an announcement.

Blind-paste risk: Elegant text may be difficult to breathe and record.

CUT LIST

The middle sags

Anya has a 14-minute transcript and no timestamps. She explains the setup three times. She wants an edit map, not invented timecodes.

Then do this. Watch the footage with the list. Restore any breath, reaction, or setup the transcript missed.

Blind-paste risk: A clean text cut can create a visible jump or emotional break.

VISUAL PROMISE

The still cannot prove it

Cole has a still of a cracked planter. A suggested thumbnail says “PERFECT FIX.” The finished repair is not in the frame.

Then do this. Put the words over the actual still. If the picture cannot carry the claim, change the words.

Blind-paste risk: The brief may promise visual evidence that does not exist.

NEWSLETTER

The draft has a fake week

Jules has three published posts and two notebook scraps. A first draft mentions a launch he only considered.

Then do this. Open every link. Delete anything reporting work you did not finish.

Blind-paste risk: Tentative notes can become public promises.

CLIP FINDER

One talk, three clips

Sofia has a 22-minute interview. One sentence sounds sharp alone, but the guest corrects it in the next breath.

Then do this. Watch ten seconds before and after each clip. Reject any excerpt whose meaning changes.

Blind-paste risk: A sharp sentence may become misleading when its correction is removed.

SERIES PLAN

Use the shelf, not a fantasy calendar

Marisol has nine mending videos, three leftover clips, and one filming afternoon. She does not need thirty imaginary ideas.

Then do this. Choose the option with the fewest unsupported assumptions. Check that the footage exists.

Blind-paste risk: A neat plan may rely on resources or demand you never supplied.

FINAL PASS

Find claims before they find you

Dev’s review says the battery lasted all day. His notes say six hours. He needs a verification list, not smoother prose.

Then do this. Open the original source for every quote and number. Weaken, label, or remove what you cannot verify.

Blind-paste risk: The checklist can repeat the draft’s mistake.

STUCK / WEEK PLAN

One production hole at a time

The first reply is too smooth. Do not change tools yet. Name the failure.

Small week: Monday, repair one paragraph. Tuesday, draft one outline. Wednesday, test titles against the ending. Thursday, watch a cut list against footage. Friday, verify claims and visual promises.

Stop when the immediate hole is fixed. Ten unreviewed versions are not progress.

PUBLISH CHECK

Your name is on the work

Rae’s post is scheduled for 8:00. One sentence says “thousands asked.” Her source note says three people emailed. She catches it with four minutes left.

The model cannot remember the room, grant permission, or take responsibility for a claim under your name. Publish only the version you edited.

KEEP / CUT

Save the instructions that earned their place

Leon has eleven chats called “final.” He cannot remember which instruction fixed the script and which created fake timestamps.

Keep:

• one voice-comparison prompt with a real sample;
• one outline-to-rough-draft prompt;
• one transcript cut-list prompt;
• one visual-promise check;
• one final claims checklist.

Cut:

• giant universal prompts;
• requests to “make it viral” or imitate someone;
• old chats containing private material;
• generated facts you never verified;
• twenty alternatives when you need one editable pass.

Name prompts by the job: CUT LIST — NO TIMESTAMPS INVENTED is clearer than CONTENT MASTER. A useful prompt makes the next human decision easier.

EDITOR'S NOTE

The draft is not the published work

The cut list is not the cut. The thumbnail brief is not the frame. The title option is not the promise you must make.

Use these boxes on material you own or have permission to use. Keep private work out unless the relevant rules and tools permit the upload.

Verify claims at the original source. Never manufacture audience statistics, quotations, performance results, sponsorship terms, discounts, reviews, or social proof.

Do not ask for another creator’s voice. Supply your own sample, then make the final judgment yourself. A model can compare language patterns. It cannot know with certainty what sounds like you.

If an edit makes the work smoother but less accurate, restore the accurate version. If a shorter clip changes the speaker’s meaning, keep the context. If a stronger thumbnail promises a frame you do not have, weaken the promise.

Your review is the part that makes the work yours.

With care,
Everyday AI

SOURCES

The four claims behind the lab page

OpenAI / Chatterji et al., “How People Are Using ChatGPT,” September 15, 2025. Writing is about 40 percent of work-related messages; about two-thirds of writing messages request modification of supplied text.
https://openai.com/index/how-people-are-using-chatgpt/

Anthropic, “Be clear and direct.” Clear instructions, roles, long material first, and queries last; up to 30 percent improvement in tests, especially with complex multi-document inputs.
https://docs.anthropic.com/en/docs/build-with-claude/prompt-engineering/be-clear-and-direct

Anthropic, “The AI Fluency Index,” 2026. Iteration is associated with more fluency behaviors. The finding is correlational.
https://academy.claude.com/tutorials/the-ai-fluency-index

Google Cloud, “Prompt design strategies.” Objective, persona, constraints, context, and response format.
https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/prompts/prompt-design-strategies

These sources do not show that AI improves creator quality, audience growth, engagement, or publishing results.