Generate Content With AI: A Repeatable Workflow With a Named Human at Every Stage

The draft was fine. Nobody owned what came before and after it.

Say you’re the content marketer at a 15-person software company. On Tuesday the product lead sends you a three-line note about a new integration and asks for a blog post, a LinkedIn post and a newsletter blurb by Thursday. You paste the note into an assistant, ask for all three, and twenty minutes later you have three clean drafts.

On Thursday morning the head of sales messages you. The blog post says the integration “connects with over 40 tools.” It connects with twelve. The assistant filled the gap in a three-line note with a number that sounded right, you skimmed past it because the paragraph read well, and the product lead assumed you’d checked. Nobody was careless, exactly.

Nobody was named.

That’s the failure this guide is built around. The drafting went fine. What went wrong happened before the prompt, in a thin brief with no facts to draw on, and after it, with no one whose job was to check a claim against anything. So the workflow below puts a locked brief and a source pack in front of the draft, a named human at each of five stages, and one run sheet that travels with the piece.

It makes sense to put the effort around the draft rather than into it, because drafting is the step AI has already shortened most. In an MIT experiment with 444 college-educated professionals doing writing tasks such as press releases, short reports and delicate emails, using ChatGPT cut the time taken by about 37% and raised the quality grades.^[1] That was a controlled test of single tasks with the brief already set, so it tells you nothing about whether the facts in the output were right. It does suggest where your hours stop being scarce.

The draft is cheap now. The brief, the checking and the sign-off are where a content team’s time goes or leaks.

Five stages, one record per piece

The same five stages work for a blog post, a LinkedIn post, an ad set, a video script or a newsletter. What changes by format is the checklist at stage three, not the order. Each stage gets one named human, and a run sheet records who.

Here is the run sheet filled in for the integration post from the opening, as it would have looked if the workflow had existed that Tuesday. It’s one row per stage and can live in a spreadsheet or a project card.

One name per stage sounds fussy until a piece goes wrong. When two people share a stage, each tends to assume the other did the checking, which is a pattern we’d call common rather than something research has measured. When all five names are yours, writing them down still tells you which hat you were wearing when you let the claim through.

Lock the brief and the source pack before you prompt

A brief is what you’d hand a freelancer. A source pack is the one thing an assistant doesn’t have, which is the facts. If you leave it out, the model writes from what it has seen about the topic in general and fills every gap with something plausible, as the “over 40 tools” line did. Locked means changes to either go through one person, so the drafter and the checker are looking at the same material.

This gives the checker something to compare against, which “does this seem accurate” doesn’t. A pack usually holds:

• Documents you own: product pages, release notes, approved messaging, past campaign results.
• Numbers someone has confirmed, with the name of who confirmed them.
• Quotes and names, only where the person has agreed to be used.
• Two or three past pieces that sound the way you want this one to sound.
• Required language: disclosures, legal lines, claims your industry has to word a certain way.

If you want facts from outside the company, such as an industry figure, find them yourself in the original source and put the link in the pack. Don’t ask the assistant to bring them back, because a statistic it returns can’t be checked against anything you gave it. Once the brief is a template, save it somewhere the whole team can reach; our guide to building an AI prompt library for your team covers how to keep that kind of shared file current.

Draft once, then check claims and voice as two jobs

Now the prompt. It does three things the one-liner doesn’t: it restricts the model to the pack, it tells the model what to do when a fact is missing, and it makes the draft arrive with a table you can check.

Then check, in two passes, and in this order.

Claims come first because that’s where the damage is, and the reason is how these tools fail: the output looks as finished when it’s wrong as when it’s right. In a Harvard Business School and BCG field experiment with 758 consultants, those given a task chosen to sit outside what the model handled well were 19 percentage points less likely to reach a correct solution than consultants working without AI.^[2] It’s a working paper on consulting cases, not marketing copy, so treat it as a warning about method rather than a measurement of your drafts. Our reading is that polish gives you no evidence of accuracy, which is why the checker works from the claims table against the pack and ignores how the prose sounds.

Voice is the second pass, and it stays separate because rewriting for voice can quietly turn a sourced sentence into an unsourced one. Once the claims are signed off, run a short prompt that’s only allowed to change tone.

If your voice notes are thin, our guide to training AI on your brand voice shows how to build them. For a full scoring method on finished pieces, how to evaluate AI-generated marketing content goes further than the five-question check below. For format craft, the guides to a blog post or a video script handle one format each. This one covers the process around them.

Who edits, who signs, who publishes

Content teams get into trouble when “someone will look at it” stands in for a named person with a specific thing to look at. The table below makes the split explicit for the formats most marketing teams publish. Start with it as written and adjust the approvers to your size.

Before the signer signs, they score the piece.

It takes a couple of minutes, and it’s the same five questions for every format.

Questions two and five aren’t arbitrary. Google’s guidance on AI-generated content says it is critical to “manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing,” and that using generative AI to produce many pages without adding value for users may violate its spam policy on scaled content abuse.^[3] That’s a search platform’s guidance, not a law, but it names the same two failures the check guards against: nobody verified it, and nothing was added.

A gap of a few hours, or a different person, lets you see what the piece says rather than what you meant it to say.

What changes when you're solo and when you review a team's drafts

Everything above assumes someone checks someone else’s work.

A solo marketer doesn’t have that, and a reviewer on a larger team has a different problem.

The solo marketer writes the brief and reads the draft, so every gap in the brief is invisible. If you assumed the integration worked with a certain tool and didn’t write it down, the draft will agree with you and you’ll approve it, because it matches your assumption. The fixes are the claims table, which forces each assumption onto the page, and the cold read: a night between drafting and signing, then check the table against the pack before you read the prose.

A brand manager, comms lead or legal reviewer on a team usually gets a draft and nothing else. Without the brief and the pack they can’t tell what was an instruction and what the model invented, so they end up proofreading, which in our view is the quickest thing to do and gives the least protection. The fix is that the run sheet travels with every piece, and the reviewer is entitled to send back anything that arrives without a source pack and a claims table. That’s a norm worth agreeing with the team before the first piece goes through, not mid-review.

The weight of the check should match what a mistake would cost. A one-line caption that names nothing and promises nothing can run lighter than a landing page that states a price, and the bar is worth writing down so the next person doesn’t set it somewhere else.

How to tell whether the workflow is holding

Pieces going out on time tell you the process is running. They don’t tell you whether anyone is still checking. The measurement lens we use at Future Factors has three steps, use, persistence and impact, and here is how they look for a content workflow.

It also helps to keep the parts of change apart. Showing a team the prompts and the claims table is training, and it builds capability. The run sheet and the named owners are workflow design, because they give that capability a recurring place to land. The monthly sheet check and the cold-read habit are reinforcement, which turns it into behavior. Adoption is what you see if all three happened: the team still building packs in month four without being asked.

Teams can start this internally with the tables above, and plenty will. Hands-on training shortens the learning curve in specific ways: it lets a content marketer, a brand reviewer and a signer practice on one real piece together, it gets the whole team working to one review standard rather than one person who’s good at it, and it covers the judgment calls, like how heavy the check should be for a regulated claim, that a guide can only point to.

Here’s a first step for this week:

1. Pick one format you publish every week, and one piece coming up.
2. Fill in the brief and gather the source pack before you open an assistant.
3. Run the draft prompt and read the claims table before the prose.
4. Give the claims to someone else, or to yourself tomorrow morning.
5. Fill in a run sheet and keep it with the published piece.

Do that once, and you’ll know which stage your team’s real weak point is. That’s the one to work on next.

The rest can wait for the second piece.

Hina Mian, Co-Founder of Future Factors AI

Hina is a marketing strategist with over a decade of hands-on campaign experience across B2B and consumer brands. She writes about using AI to run leaner, sharper marketing without losing the human touch. Future Factors helps professionals and teams build practical AI capability through role-based training, workflow design, and hands-on adoption.

More about Hina →