Playbook · Content
AI-powered content creation without publishing filler
The model does structure and drafts. You supply the evidence. That split is the whole method.
- 7-step workflow
- 3-4 hrs per piece
- Free tier viable
AI-powered content creation without publishing filler: what should you actually do?
AI-powered content creation works when the model handles structure, first drafts and format conversion while you supply the evidence: your own numbers, screenshots, customer quotes and opinions. Budget three to four hours per published piece rather than twenty minutes. Volume without first-hand information performs badly in search and worse with readers.
Last updated 3 August 2026
| Play | Effort | Cost |
|---|---|---|
| Source topics from questions, not keyword lists | 90 minutes | Free |
| Write the brief yourself, in ten lines | 15 minutes per piece | Free |
| Generate the outline, then delete a third of it | 20 minutes | Free |
| Draft, then run the evidence pass | 90 minutes per piece | Free |
| Verify every fact against a source you opened | 30 minutes per piece | Free |
| Convert into three other formats after publishing | 45 minutes | Free |
| Review performance quarterly and update, do not add | 3 hours per quarter | Free |
The promise of AI content creation is twenty pieces a week. The reality is that twenty pieces of the same synthesised summary everyone else can generate does not earn a ranking, a link or a reply. The publishing bottleneck moved from writing to having something worth saying, which is where it always should have been.
The workflow that does work splits the job cleanly. Models are good at structure, first drafts, format conversion and catching gaps in your outline. They cannot supply what makes a piece worth reading: your own numbers, your screenshots, what a customer actually said, and an opinion you would defend.
This page is the seven-step version of that split, with honest timings. Three to four hours per published piece is normal. If a piece takes twenty minutes, nothing in it required you to exist.
- — Small teams publishing one to four pieces a month and wanting them to count
- — Founders with real expertise and no time to write it down
- — Marketers whose AI content shipped fast and performed badly
- — Anyone rebuilding a content process after a search visibility drop
Anyone whose goal is publishing volume for its own sake. If the plan is 100 pages a month with no first-hand input, this workflow will feel slow — and the 100 pages will underperform ten good ones.
The moves
Ranked, highest return first.
Work down the list. Each one names the first step so there's nothing to plan.
- 01
Source topics from questions, not keyword lists
Keyword tools tell you what everyone targets. Your inbox and sales calls tell you what your specific buyers are confused about, which is where you can say something others cannot.
First step: Collect 30 real questions from emails, calls and support tickets. Group them with the model into themes, then rank by how often each comes up and whether you have proprietary evidence for it.
- Tools
- Your inbox, ChatGPT or Claude
- Effort
- 90 minutes
- Cost
- Free
- 02
Write the brief yourself, in ten lines
A brief that names the reader, the promise, the evidence you will use and the opinion you will take is what stops the draft being generic. This is the highest-leverage ten minutes in the process.
First step: For each piece write: who it is for, the question it answers, three specifics only you have, the opinion, and what the reader does next. Give the model that, not the topic.
- Tools
- A text file
- Effort
- 15 minutes per piece
- Cost
- Free
- 03
Generate the outline, then delete a third of it
Model outlines default to complete coverage of a topic, which produces long pages that answer nothing sharply. Cutting sections is how you get a piece with a point.
First step: Ask for an outline against your brief, then remove every section that exists only for completeness. Keep what serves the promise. Most outlines lose three sections and improve.
- Tools
- ChatGPT or Claude
- Effort
- 20 minutes
- Cost
- Free
- 04
Draft, then run the evidence pass
This is the step that decides whether the piece performs. Every general claim gets replaced with a number, a screenshot, a quote or an example from your own work.
First step: Read the draft and mark every sentence that could appear on a competitor's site unchanged. Replace each with something specific to you. Aim for at least one concrete item per section.
- Tools
- Your analytics, customer emails, screenshots
- Effort
- 90 minutes per piece
- Cost
- Free
- 05
Verify every fact against a source you opened
Models invent statistics, dates, prices and study citations fluently. One published invention costs more trust than the workflow saves in a quarter.
First step: List every number and citation in the draft. Open the source for each. Delete anything you cannot verify rather than softening the wording.
- Tools
- A browser, 20 minutes of patience
- Effort
- 30 minutes per piece
- Cost
- Free
- 06
Convert into three other formats after publishing
Format conversion is where models are strongest and where most teams stop too early. The research cost is already paid; the marginal cost of a second format is minutes.
First step: After publishing, ask for an email version, a social version and a short video script, each keeping the specific numbers. Edit the openings by hand — that is what people judge.
- Tools
- ChatGPT or Claude
- Effort
- 45 minutes
- Cost
- Free
- 07
Review performance quarterly and update, do not add
Updating a page that already has some traction usually beats publishing a new one, and models are good at spotting what an existing page fails to answer.
First step: Take your top five pages by impressions. Ask the model what questions each fails to answer, answer three per page yourself, and republish.
- Tools
- Search Console, ChatGPT or Claude
- Effort
- 3 hours per quarter
- Cost
- Free
Sequence
What to do first, week by week.
Topics and briefs
Collect the questions and write briefs for the first three pieces. Do not draft anything yet. Teams that skip this week produce faster and publish worse.
First piece, full workflow
Run one piece through all seven steps and time each. Most people find the evidence pass takes longer than the draft, which is the correct ratio.
Two more pieces and the format conversions
Now that the workflow is known, produce two pieces and convert each into three formats. Track how long the second and third take — the drop is usually 40%.
Update instead of publish
Spend the whole week improving existing pages rather than adding new ones, then compare which moved more: the three new pieces or the five updates.
Avoid
Where this usually goes wrong.
Measuring the process by pieces published
Output is the easiest metric to move with AI and the least connected to revenue. Count pieces that earned a link, a reply or a signup instead.
Skipping the evidence pass
A draft without your own numbers, screenshots or customer language is a summary of things already online. It reads fine and does nothing, which is the hardest failure to notice.
Letting the model choose the opinion
Models produce balanced positions by default. Balanced content is forgettable. Decide what you actually think before drafting and instruct it to argue that.
Publishing citations you did not open
Fabricated studies and misattributed statistics are common and highly visible when someone checks. Open every source or remove the claim.
Producing before distributing
Ten pieces with no distribution plan perform worse than three pieces sent to a list and posted where your buyers already are. Plan the distribution in the brief.
Questions
Common questions.
Does AI-generated content rank in Google?
Content produced with AI can rank; content that contains nothing first-hand generally does not, because dozens of similar pages exist. The differentiator is the evidence you add, not the tool that drafted it.
How long does one AI-assisted piece really take?
Three to four hours for a substantial page: 15 minutes briefing, 20 outlining, 45 drafting, 90 on the evidence pass and 30 verifying. The draft is the fastest part and the least important.
Which model is best for content creation?
Claude and ChatGPT are both strong for long-form drafting; the difference between them is smaller than the difference a good brief makes. Use whichever holds your context file most conveniently.
Should I disclose that I used AI?
There is no legal requirement in most markets for marketing content, and readers care about whether the piece is useful and accurate. Some publications and platforms require disclosure — check theirs.
How much content should a small business publish?
One to four substantial pieces a month, each with something first-hand, beats twenty thin ones. If you cannot supply evidence for a piece, that is a signal not to publish it rather than a gap to fill with AI.
More playbooks
Other plans worth reading.
AI marketing: the complete, honest guide
ReadPlaybook · B2B SaaSGrowth plan for a pre-revenue B2B SaaS
ReadPlaybook · EcommerceGrowth plan for an ecommerce store's first 100 orders
ReadPlaybook · LocalGrowth plan for a local service business
ReadPlaybook · B2B SaaSGrowth plan for a B2B SaaS with early traction
ReadPlaybook · Marketing agencyLead generation plan for a marketing agency
ReadElsewhere on gtm.help
Keep exploring.
Free marketing tools
Calculators and generators that run without a signup.
BrowseTool guides
What to use, what it costs, when it earns its keep.
BrowseTool comparisons
Head-to-head picks for the calls people get stuck on.
BrowseMarketing checklists
Task lists by business type and competitive situation.
BrowseStraight answers
The growth questions founders ask, answered with numbers.
BrowseGet started
Get this ranked for your business.
Paste your URL and you'll have a plan built from your site and competitors in about a minute.