How to use AI for content creation

You can get ChatGPT or Claude to write a post in a minute, but if you just ask, it sounds like everyone else's, and nobody comments. Here is how to use AI for content creation so it sounds like you: give it your words and your facts once, feed it something true from your own week every day, and check every piece before it goes out. Since May 2026, LinkedIn limits the reach of posts that read like AI with nothing of yours in them, so the generic way now costs you readers too.

Do these three things

About 2 hours a day, every day, for 4 weeks. Judge it after 28 pages and 112 posts, not after the first 10.

Show the three actions in fullHide the full list

1. 1Today: write a one-page voice and facts file: who you sell to, the words you use, 20 true facts and the phrases you never use. Save it in a free ChatGPT or Claude project, so every draft starts from your words, not the AI's.
2. 2Every day: give AI 10 minutes of raw material only you have, such as a customer call or a question a buyer asked. Turn it into 1 page that answers a search and 2 posts on each of your 2 channels. Add one true detail to each piece and check every fact before it goes out.
3. 3Every Sunday: count the comments, visits and enquiries each piece brought, using UTM links. Keep the kinds of post that worked, drop the weakest, and add your best lines to the file.

Get help doing these

ChatGPT or Claude reads this guide and works through it with you.

Do this with ChatGPT(opens in a new tab)Do this with Claude(opens in a new tab)

1. Write a voice and facts file and save it in a ChatGPT or Claude project

Write one page that tells the AI who you are for, how you talk and what is true about your business, and save it as a project's instructions in ChatGPT or Claude. AI sounds generic when it knows nothing about you. This page is how it learns, once, instead of in every prompt.

Fill in the template below. Take the first line from your ideal customer profile and your value proposition if you have them. For "Words we use", copy phrases your buyers said to you in calls, emails and reviews, word for word. For the facts, write only things you can prove: prices, numbers from your product or your work, what customers told you, how long things take. AI will use these instead of making numbers up.

Then open ChatGPT or Claude and create a project. Both have projects on the free plan: Claude allows up to 5, ChatGPT allows up to 5 files in each. Paste the file into the project's instructions and upload anything else useful, like your homepage text and 3 posts you wrote yourself that people liked. Start on the free plan. If you hit the free limit every day, pay $20 a month for one of them, not both.

Who we are for: [kind of buyer], [size], who [the moment the problem hurts].
What we sell: [product or service], [price], [what the buyer gets].How we write:
Short sentences. Plain words. "You", not "users".
Words our buyers use: [5–10 phrases, copied from calls, emails, reviews].
We always: [say a number where we have one / end with one question / name the real tool].
True facts we can use (never invent others):
[a number from our product or work]
[what a customer told us, with their permission]
[a price, a time, a result we can show]
... up to 20.
Never write: "in today's fast-paced world", "let's dive in", "game-changer", "unlock", "elevate", [words you hate].
Never invent: numbers, customer quotes, reviews, names, results. If a fact is missing, write [FACT NEEDED].

• Tool: a Google Doc, and ChatGPT or Claude on the free plan.
• Time: 1 hour.
• You will have: a project in ChatGPT or Claude that every draft starts from.
• It worked if: you ask for the same post twice, once inside the project and once outside it, and the project version uses at least 2 of your facts and none of your "never write" words. If you cannot tell the two apart, add 10 more facts and 5 more buyer phrases.
• Common mistake: a one-line prompt like "write a LinkedIn post about onboarding". With nothing of yours in it, AI gives you the average post on the internet.

If you run an agency: make one project per client, never one for all of them, and keep each client's voice separate. Ask the client to correct their file before you write anything, and agree who approves each piece. Make a separate project for your own agency's marketing too.

2. Collect 10 minutes of raw material every day that only you have

Every day, spend 10 minutes saving something true from your own work: a question a buyer asked, a number, a mistake you fixed, a photo. This is what makes your content different. Google's own guide says a first-hand account gives "a unique perspective", while a summary "simply restates information already available elsewhere".

Keep a Google Sheet called "Idea bank" with three columns: the date, what happened, and the one line that matters. Before you start step 3, put 20 items in it. Go back through your last month: calls, support emails, customer interviews, comments, and the questions people asked in online communities. After that, add at least 3 new items every day.

The fastest way is to talk. After a call, open a Google Doc, click Tools, then Voice typing, and say what the buyer asked and what you told them, for 2 minutes. It is free in Chrome, Edge and Safari. Paste the text into the sheet. If you run a SaaS, add every support question, every bug a user found and every change you shipped. Feedback from your first users is the best material you will ever have.

If you run a B2B company: the best items are the questions buyers ask on discovery calls and in quote requests: lead times, minimum orders, prices, what went wrong with their last supplier. Write each one down the same day.

If you run a D2C store: take 3 photos of your product in real use every day, and save every customer message and review. Ask before you quote anyone by name. Customer photos and videos are user-generated content, and they beat anything AI can draw.

• Tool: Google Sheets and Google Docs voice typing. Both free.
• Time: 1 hour once to fill the first 20, then 10 minutes a day.
• You will have: an idea bank with 20 items before day 1, and at least 3 new items every day.
• It worked if: the bank never runs out, and every draft you make in step 3 starts from one item in it.
• Common mistake: asking AI "give me 30 content ideas for my business". You get the same list every competitor gets. Google calls this commodity content, and its example is "7 Tips for First-Time Homebuyers".

3. Let AI draft 1 page and 2 posts a day on each of your 2 channels

Every day, take one item from your idea bank and ask your project for 5 drafts: 1 page for your website and 2 posts for each of your 2 channels. With the file and the item in front of it, AI drafts all 5 in about 20 minutes, which is what makes this number possible for one person.

First, pick your 2 channels: the 2 places where you can name the most buyers who post or comment. If you are not sure, use the channel guide for your business: SaaS, B2B, ecommerce or agency. Most founders selling to businesses start with LinkedIn; most stores start with Instagram and short videos. Pages go on your own site's blog, each one answering a search people really make.

Two posts a day on each channel sounds like a lot. Two posts a week is what most people do, and after a month they have 8 posts and no idea what works. At 2 a day you have 56 posts per channel in 4 weeks, which is enough to see which kind gets buyers talking. AI drafts; you add the one true detail in step 4. Spread the posts over the day, and plan the week ahead in a content calendar once you are past week 1.

Paste this into your project, with the day's item:

Today's item from my idea bank: [paste the item: what happened, and the one line that matters].Using my voice and facts file, write:
One web page that answers this search: "[the question a buyer types]". Answer it in the first two lines. Use only facts from my file or from today's item. Mark anything you are not sure of as [FACT NEEDED].
Two posts for [channel A], each under [150] words, each starting with a different line from today's item.
Two posts for [channel B], in the format that channel uses: [a short video script for 9:16 / a text post / a carousel outline].
No hashtags unless I ask. End each post with one question a buyer would want to answer.

One item can feed both channels in different shapes. A call becomes a page, a story post, a short tips post and a video script. When you want to turn one strong piece into a week of posts, repurpose it the same way.

• Tool: your ChatGPT or Claude project.
• Time: about 20 minutes a day for all 5 drafts.
• You will have: 5 drafts every day, each built on something real.
• It worked if: you have 5 drafts within 30 minutes of opening the project, and every one names something from today's item.
• Common mistake: publishing what AI wrote without editing, at high volume. Google's spam policy calls "using generative AI tools … to generate many pages without adding value for users" scaled content abuse. One page a day that answers a real question with something only you know is not that. The other mistake is posting where your readers are other founders, not buyers. One founder posted more than 50 times on Indie Hackers in 2026 and made $0, because the readers were builders, not buyers.

4. Edit every draft so it sounds like you, and check every fact

Before anything goes out, run each draft through the 6-line check below and add one true detail that only you could know. A post takes about 10 minutes to edit and a page about 40. This is the part AI cannot do for you, and it is why people stop scrolling.

One true detail only I could know: a number, a name (with permission), what happened.
Every fact checked against my file or a source I opened today. Every [FACT NEEDED] filled or cut.
No phrase from my "never write" list, and no sentence I would not say out loud.
Read out loud once. Anything I stumble on, I rewrite.
It ends with one question or one next step, not three.
Nothing invented: no made-up quote, review, customer, result or statistic.

Line 2 matters because AI states wrong facts in a confident voice. Google's guidance, updated 1 October 2026, says "It is critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing." Line 6 is also the law in the US: since October 2024, the FTC's rule on reviews bans fake reviews and testimonials, including AI-generated ones, and in December 2025 it sent its first warning letters. Real testimonials come from real customers.

Then publish. Pages go on your blog; ask Google to index each one the day it goes live. Posts go out at the times you planned. Put a UTM link on anything that points to your site, so step 6 can count it.

Reply to every comment yourself, in your own words. LinkedIn said on 20 May 2026: "It's ok to use AI to help you write, but your posts and comments need to represent your voice and your perspectives." It now limits the reach of posts that seem AI-made with no clear view, and is restricting AI-written comments. Since 30 July 2026, members can also mark a post "Seems like AI slop". On Reddit, many subreddits remove AI-written posts, so read each one's rules and write your answers yourself, the way the Reddit guide shows.

• Tool: your project (ask it "point out every claim in this draft that needs a source"), and the 6-line check.
• Time: about 40 minutes for the page and 10 minutes for each post: about 1 hour 20 minutes a day.
• You will have: 1 page and 4 posts live every day, each with something true of yours in it.
• It worked if: every piece passes all 6 lines, and by the end of week 2, comments say specific things ("we had this exact problem") instead of only "great post".
• Common mistake: letting AI fill a gap with a number that sounds right. If you cannot find the source in 2 minutes, cut the sentence.

If you run a D2C store: let AI draft your product descriptions and captions, but every size, material, ingredient and delivery time comes from you. Google's guidance says AI-made product data in a store must be labelled as AI-generated, so write those facts yourself.

5. Make images and short videos with AI, and label anything realistic

Use AI to write video scripts, on-screen text and simple images, then record the videos yourself on your phone. Label any video or audio made with AI that could pass for real. On every major platform, a script written with AI needs no label, but a realistic AI person, scene or voice does.

If one of your channels is video, your 2 posts a day there are short videos. Ask your project for a 30-second script from the day's item, with the first line written on screen. Record it yourself in portrait (9:16), add captions, and post. That takes about 20 minutes a video. The short-form video guide covers filming, and the YouTube guide covers longer videos.

Check every image and video against the platform rules before it goes out:

• YouTube: when you upload, find "AI use" in the Attributes section and choose Yes if the video makes a real person seem to say or do something they did not, changes footage of a real place or event, or shows a realistic scene that did not happen. Scripts, thumbnails and cloning your own voice for a voice-over need no label.
• Instagram, Facebook and Threads: Meta requires its AI label on posts with a photorealistic video or realistic-sounding audio that was made or changed with AI, and may penalise accounts that skip it. Turn on the AI label when you post. Meta also adds "AI info" itself when it finds AI markers in a file.
• TikTok: its 2026 rules require a clear label when AI or editing is used to show realistic people or scenes.

Never use AI to make a customer, a review screenshot, a "before and after" or a product photo that looks real. It is the fake review problem from step 4 in picture form. Use AI for diagrams, simple illustrations and ad creative backgrounds, and real photos for anything a buyer will trust.

If you run a D2C store: use real photos for every product image. If you ever use an AI-made image in Google's free Shopping listings, Google requires it to keep the file's AI metadata, so do not strip it. The simplest rule is the one in the Google Shopping guide: photograph the real product.

• Tool: your phone, your project for scripts, and each platform's own upload screen.
• Time: about 20 minutes a video; 1 minute to check the label rule.
• You will have: videos and images live every day, each labelled where the rules ask.
• It worked if: 0 posts removed or flagged, and every realistic AI clip carries its label.
• Common mistake: an AI-made "happy customer" holding your product. If a viewer could think it is real, it needs a label, and it still is not a real customer.

6. Every Sunday, count what each piece brought and update your file

Every Sunday, write down what the week's pages and posts brought: comments from buyers, visits to your site, and enquiries, signups or orders. Keep making the kinds of post that brought people, drop the weakest kind, and add the lines people answered to your voice file.

Add "Where did you hear about us?" to your signup, order or contact form. This is self-reported attribution, and at this size it tells you more than any dashboard. Then fill in one row each Sunday:

Week | Pages published | Posts per channel | Buyer comments | Visits from posts and pages | Enquiries, signups or orders | Where each one came from
Best kind of post this week: [story / tips / question / video]. Weakest: [ ].
Lines people answered: [copy them into the voice file]

Count only comments from people who could buy, not from other founders or friends. Pages are slower: check in Google Search Console how many have any impressions, and judge them at month 3, not week 4. A page that gets impressions but few clicks can be made to rank higher later.

After 112 posts (4 weeks at 2 a day on each channel), decide. If a channel brought 0 buyer comments and 0 visits, move its posts to the other channel or to a new one. If one kind of post brought most of the enquiries, make half of each day's posts that kind. Your aim is the channel that brought each deal, not likes. Once a month, step back for a marketing review.

• Tool: Google Sheets, Search Console, and each platform's free analytics.
• Time: 30 minutes each Sunday.
• You will have: one row a week, and a voice file that gets better every week.
• It worked if: you can name the piece that brought each enquiry, signup or order. If after 28 pages and 112 posts nothing has brought a single visit, go back to step 3 and check your channels are where your buyers are.
• Common mistake: counting likes and followers, or judging after 10 posts. Ten posts tell you nothing; 112 do.

Where Scout7 helps

Scout7 does parts of steps 1, 3, 4 and 5.

Step 1. When you sign up with your website, Scout7 reads it and builds a brand profile: positioning, audience, voice and guardrails, held as facts you can edit. Every post, article and video it makes reads from that profile, so you correct the facts once instead of in each prompt.

Step 3. Content Co-Pilot proposes topics or takes yours, then asks research-backed shaping questions before it drafts. Your answers are where your first-hand detail goes in. It writes the article with images, sources, FAQs and structured data, and can turn it into a LinkedIn carousel or a video. Image Studio makes image posts with captions in your brand voice.

Step 4. Scout7 schedules posts to LinkedIn, X, Instagram, Facebook, Threads and Reddit from one calendar. Nothing is published until you approve it, so the 6-line check still happens before every piece goes out.

Step 5. Video Studio writes the script, splits it into scenes you direct, films each one with the cast and voice you choose, in portrait or landscape, and adds captions. You approve the final cut. It works inside Claude and ChatGPT too, so you can run it from the assistant you already use.

Your AI content checklist for the next 4 weeks

0 of 12 done

Work through this list with an AI assistant

It drafts the work and keeps track of what is done.

Do this with ChatGPT(opens in a new tab)Do this with Claude(opens in a new tab)

Step 1 · Write a voice and facts file and save it in a project

• Write your one-page voice and facts file.
• Save the file in a ChatGPT or Claude project and test it with the same request inside and outside.

Step 2 · Collect 10 minutes of raw material every day

• Start an idea bank with 20 items from your last month of calls, emails and questions.
• Every daySpend 10 minutes saving a question, a number, a voice note or a photo from today.

Step 3 · Let AI draft 1 page and 2 posts a day on each of 2 channels

• Pick your 2 channels and where your pages will go.
• Every dayAsk your project for 1 page and 2 posts per channel from one idea bank item.

Step 4 · Edit every draft so it sounds like you, and check every fact

• Every dayEdit each draft with the 6-line check, then publish it.
• Every dayReply to every comment yourself, in your own words.

Step 5 · Make images and short videos with AI, and label anything realistic

• Every dayCheck every image and video against the label rule before it goes out.

Step 6 · Every Sunday, count what each piece brought

• Every SundayFill in one row of the Sunday sheet.
• Every SundayAdd the lines people answered to your voice file and drop the weakest kind of post.
• After 112 postsKeep or move each channel, and make the best kind of post half of each day's posts.

Copy listDownloadReset