You've tried AI prompts. A few drafts came back usable, most sounded like a brochure, and none of it became a habit, because each session starts from a blank chat window.
An AI content engine fixes the repeatability problem. It's a small system: a folder of source material only you could supply, a board that tracks each piece by ID and status, a fixed prompt sequence, and a human review that decides what ships.
This guide, part 1 of a 5-part build, sets up the first version with a spreadsheet, a document editor and whichever AI assistant you already use (ChatGPT, Claude or Gemini all work). By the end of an afternoon you'll have 1 approved post, a board that shows how it got there, and 3 more briefs ready for next week.
Jump to the free First AI Content Engine Build Kit
How to build an AI content engine: set up the source folder and board
Building an AI content engine starts before any prompt. Gather the material only you can supply into 1 source folder, give each source a short ID, then set up a spreadsheet board where every piece of content has an ID, a buyer question, its sources and a status. The assistant drafts from that folder and nothing else.
Meet the example business
Ridgeline Ops is an illustrative business invented for this series, and its owner, Maya Okafor, is fictional too. A former distribution-center operations manager, she helps 20-to-80-person distribution companies fix order handoffs between sales and the warehouse.
Her buyers are operations directors and owners. Her goal: 1 LinkedIn post a week and 1 email a month that lead to discovery calls. Every figure in her notes is invented.
Build the source folder (45 minutes)
Maya's folder holds 3 kinds of source, each with an ID.
| Source | What it is | ID format |
|---|---|---|
| Client notes | 6 years of anonymized notes, 1 per finding | N-1, N-2 and so on |
| Interview | A 30-minute recorded conversation, transcribed with timestamps | Interview 6:10 to 8:45 |
| Past posts | 5 LinkedIn posts she wrote herself | P-1 to P-5 |
Yours might hold call notes or a recorded talk; each item still gets an ID. Strip client names and anything that identifies a company before a note goes in. The assistant only sees what's in the folder, which is the point: a fact that isn't there can't slip into a draft without you noticing.
Set up the board (30 minutes)
Make a spreadsheet with 1 row per piece and these columns, shown here with Maya's finished RL-1 row.
| Column | RL-1 |
|---|---|
| Content ID (stable, never reused) | RL-1 |
| Buyer question, in the buyer's words | Why do rush orders get promised before the warehouse knows about them? |
| Format | LinkedIn post |
| Sources | N-3; Interview 6:10 to 8:45 |
| Brief link | RL-1 brief |
| Status | Approved |
| Draft version (rises whenever the words change) | 2 |
| Draft link | RL-1 draft |
| Approved version (the draft version you signed off) | 2 |
| Review notes | Cut an unsourced statistic; replaced the closing line |
| Next step | Reply "handoff" for the checklist |
| Publish date, then live link | Tuesday, November 10 |
| Edit minutes (opening the draft to approval) | 35 |
Each status has an entry rule. A piece that can't meet it stays put, which catches weak material before it eats an hour of editing.
| Status | Rule for entering it |
|---|---|
| Idea | A buyer question exists |
| Ready | The brief is complete and every claim it allows has a source ID |
| Drafting | Prompts are running and the draft has a version number |
| Review | A draft exists and is waiting for your edit |
| Approved | You've read the listed draft version and signed it off |
| Published | It's live and the link is on the board |
Move 1 buyer question through brief, draft and review
Moving a buyer question through the engine takes 4 steps: write a source brief that limits what the draft may claim, run a fixed prompt sequence, review the draft line by line against its sources, and approve a numbered version. RL-1 took Maya about 90 minutes the first time, and 35 of them went on review.
Step 1: Write the source brief
Status: Ready. The brief is short and strict: buyer, question, answer, the evidence the draft may use, and the next step.
SOURCE BRIEF: RL-1
Buyer: operations director or owner, distribution company, 20 to 80 staff
Question: Why do rush orders get promised before the warehouse knows about them?
Format: LinkedIn post, 100 to 200 words
Main answer: The promise happens on the sales call and the order reaches
the warehouse hours later. Fix the moment of promise before the paperwork.
Evidence the draft may use:
N-3: at a 38-person fastener distributor, 11 of 26 late orders in March
were rush orders promised before the warehouse saw them
N-3 (fix): rush orders over $2,000 now get a 5-minute call to the shift
lead before the rep confirms a date
Interview 6:10 to 8:45: "A rep promises Thursday on a 10:40 call. The
order gets keyed after lunch. By the time the pick list prints,
Thursday's truck is full."
Opinion: the handoff design causes this; reps follow the process they're given
Next step: Reply "handoff" and I'll send you the 1-page handoff checklist
I use with clients.
Leave out: client names, any number not listed above, promised results
Step 2: Run the prompt sequence
Status: Drafting. Open a fresh chat and run these 3 prompts in order, pasting the brief and the 2 past posts closest in tone (P-1 and P-3 for Maya) into the first. Keep the prompts in a document so the sequence stays identical every week.
I run [business name], [1 line on what you do and for whom]. Below
are a source brief and 2 of my past LinkedIn posts for tone.
Before you write anything, reply with:
1. The main answer in 1 sentence
2. The evidence you plan to use, by source ID
3. Anything in the brief that's unclear or missing
Wait for my reply before drafting.
[paste the source brief]
[paste 2 past posts]
Draft the LinkedIn post from the brief.
- 100 to 200 words, paragraphs of 1 to 3 sentences
- Open with a moment the buyer would recognize from their own week
- Use only the evidence in the brief, and tag each fact with its source ID
in brackets
- Close with the next step from the brief, word for word
Check your draft against the brief. List:
- each factual claim with its source ID, or NO SOURCE
- any sentence another consultant could have written word for word
- anything you added that the brief didn't ask for
Leave the post as it is. I'll do the edit myself.
Prompt 1 catches a misread brief before it becomes 200 words of confident text; fix its summary or answer its questions, then send prompt 2. Save that draft in a new document, link it on the board and set Draft version to 1. Prompt 3's claim list speeds up the review.
Step 3: Read the example draft
Here's an example draft of the kind prompt 2 can return, with prompt 3's flags added in brackets. It was written for this article as an illustration; no tool or model produced it.
Late orders are one of the biggest problems distribution companies face. [NO SOURCE]
When sales and the warehouse aren't on the same page, communication breaks down and customers feel it. Up to 40% of late shipments start with a handoff problem. [NO SOURCE]
At one client, 11 of 26 late orders in a single month were rush orders that sales promised before the warehouse saw them. [N-3]
The fix is better communication. Sales needs to loop in the warehouse earlier, and the warehouse needs to know what's coming.
If this sounds familiar, send me a DM and let's talk.
Prompt 3 flagged both unsourced claims. Maya moved RL-1 to Review.
Step 4: Review, revise and approve
Status: Review. Edit in the draft document, line by line against the brief, and give each change a reason and a source. Once the words change, it's version 2.
| Example draft line | Problem | Approved line | Source |
|---|---|---|---|
| Late orders are one of the biggest problems distribution companies face. | Generic; any consultant could post it | A rep promises Thursday delivery on a 10:40 call. | Interview 6:10 |
| Up to 40% of late shipments start with a handoff problem. | Invented statistic | Cut | None |
| At one client, 11 of 26 late orders in a single month... | Supported, but vague about who and when | At a 38-person fastener distributor I worked with, 11 of 26 late orders in March... | N-3 |
| The fix is better communication. | No action, and it puts the blame on people | Rush orders over $2,000 now get a 5-minute call to the shift lead... | N-3 (fix) |
| If this sounds familiar, send me a DM and let's talk. | Wrong next step | Reply "handoff" and I'll send you the 1-page handoff checklist... | Brief |
The approved post, version 2:
A rep promises Thursday delivery on a 10:40 call.
The order gets keyed after lunch. By the time the pick list prints, Thursday's truck is full.
At a 38-person fastener distributor I worked with, 11 of 26 late orders in March were rush orders promised before the warehouse saw them.
The reps were doing their jobs. The process asked them to promise first and check later, so we moved the check: rush orders over $2,000 now get a 5-minute call to the shift lead before the rep confirms a date.
If your rush orders keep arriving late, start with the minute the promise gets made. The paperwork comes second.
Reply "handoff" and I'll send you the 1-page handoff checklist I use with clients.
The RL-1 row moved like this that afternoon.
| Time | Status | What happened |
|---|---|---|
| 1:05 p.m. | Ready | Brief complete, 3 evidence lines listed |
| 1:20 p.m. | Drafting | Prompts 1 to 3 run; draft version 1 saved |
| 1:50 p.m. | Review | 2 claims flagged NO SOURCE |
| 2:25 p.m. | Approved | Version 2 signed off; 35 edit minutes logged |
The version number is the approval. If Maya changes a word after approving version 2, the draft becomes version 3 and goes back to Review. The rule looks fussy now and saves a mess in part 4, when software starts moving approved work for you.
The First AI Content Engine Build Kit has this board, brief, prompt set and post, plus blank versions.
Connect approved content with publishing and a useful next step
Approved content needs 2 more things before the engine is finished: a publishing slot and a next step the reader can take. Schedule the approved version exactly as written, record the live link on the board, and make sure the next step points toward the service you sell. Then write the next 3 briefs while the sources are fresh.
Publish the approved version
Paste the approved version into LinkedIn's scheduler word for word. LinkedIn's help page says you can schedule a post between 10 minutes and 3 months ahead, so Maya schedules RL-1 for Tuesday morning. Once it's live, she adds the link and sets the status to Published.
Give every piece 1 next step
Each piece asks for 1 action that moves a buyer closer to a conversation about your service. Maya's ladder has 3 rungs: reply with a keyword for a 1-page resource (RL-1 uses "handoff"), download a longer guide from the monthly email, or book a 20-minute handoff review.
Posts mostly use the first rung, because she can follow a reply up by hand. Log which content ID produced each reply; part 5 turns those counts into a review.
Which buyer questions deserve a slot at all is a strategy call, covered in the AuthorityOS guide to personal brand content strategy.
Prepare the next 3 briefs
While the source folder is open, bring 3 more rows to Ready. Each gets a full brief in the Step 1 format; the table shows the lines that matter most.
| ID | Buyer question | Sources | Main answer | Next step |
|---|---|---|---|---|
| RL-2 | What should a sales-to-warehouse order form contain? | N-12; P-3 | Most forms carry 15 or more fields; the warehouse uses about 7 | Reply "form" for the template |
| RL-3 | Why do order changes after 2 p.m. cause so many mis-picks? | N-7; Interview 14:20 | Changes land after the pick wave; set a cutoff and 1 owner | Reply "cutoff" for the memo |
| RL-4 | How can I tell whether handoff problems cost us customers? | N-15; Interview 21:30 | Check which accounts stopped reordering within 90 days of a bad shipment | Book a 20-minute handoff review |
RL-4 has a weakness worth spotting now: N-15 describes the method but has no numbers, so a draft will be tempted to invent some, as part 5 shows.
Part 2, getting AI to write in your voice, turns the source folder into a voice and source pack. Part 3, turning 1 interview into 5 content pieces, gets more out of Maya's recording.
Block 3 and a half hours this week: 30 minutes for the board, 45 for the source folder, 90 for your first buyer question and 45 for the next 3 briefs. Next week's post then starts from a Ready row, which is the whole reason to build the engine.