The one-person marketing team: an AI workflow from brief to post

You own a small business, or you're the only marketer in one. Every Monday you mean to post three times this week, write the newsletter and reply to comments. By Thursday you've posted once, in a hurry, and the newsletter is still a blank page.

AI tools were supposed to fix this, and for many people they've only added another tab. You open ChatGPT, Claude or Gemini, type "write me an Instagram post about our sale", get something bland, fix it by hand, and wonder whether it saved any time at all.

What helps more than a better prompt is a workflow: a fixed set of steps, where AI does the parts it's good at and you sign off at the points that matter. This article lays out the one I'd set up for a one-person team.

Workflows, agents and why the difference matters

Two terms first, because they get mixed up.

An AI agent is an AI system that can take actions on its own: search the web, read your files, draft a document, then decide what to do next. An AI workflow is a set of steps you define in advance, where AI handles some of the steps.

Anthropic, the company that makes Claude, wrote a widely shared guide to this called Building effective agents (December 2024). It describes workflows as systems where AI and tools follow "predefined code paths", and agents as systems where the AI will "dynamically direct their own processes and tool usage". Its main advice is to start with "the simplest solution possible, and only increasing complexity when needed."

That advice suits a one-person marketing team well. You don't need a fleet of autonomous agents. You need a dependable chain of steps, each with a clear input and output, and a few places where you check the work.

The same guide describes a pattern called prompt chaining: breaking a task into steps where each step uses the previous one's output, with checks (it calls them "gates") in between. One of its examples is "Generating Marketing copy, then translating it into a different language." That's the backbone of the workflow below.

The workflow, step by step

Here's the chain. Each step has an input, the AI's job, and what you check before moving on.

Your time goes into the checks, not the typing. If the setup is right, those checks take minutes.

Step 1: the brief

Everything downstream depends on this, so it's worth getting right. A good brief answers: who it's for, what you want them to do, the one message, the facts that must be in it, and what to avoid.

A prompt to copy:

I run [business] for [customers]. This week I want [goal, e.g. more bookings for weekday slots]. Turn this into a one-page marketing brief with: audience, the single message, three supporting facts I've given you (don't add any), the call to action, channels, and things to avoid. Ask me questions if anything is missing instead of guessing.

That last line matters. Asking the AI to ask rather than guess cuts down on made-up details later.

Step 2: research, with sources

Ask the AI to find what customers ask about your topic, what competitors posted recently and any facts worth using, each with a link. Then actually open the links. AI tools can still state things confidently that turn out to be wrong, and a wrong fact in your post is your mistake, not the tool's.

Step 3 and 4: ideas, then drafts

Have the AI propose several angles before it writes anything. Choosing between five ideas takes you a minute; rewriting one bad draft takes twenty.

When it drafts, give it your voice notes: a short document with words you use and avoid, how you address customers, a couple of past posts you liked. Save this once and paste it into every drafting session (or store it in the tool's project or custom instructions feature if it has one). This is the single biggest improvement most people can make to AI drafts.

Step 5: a second AI as reviewer

The Anthropic guide calls this the evaluator-optimizer pattern: "one LLM call generates a response while another provides evaluation and feedback in a loop." (LLM means large language model, the kind of AI behind these chat tools.)

In practice, open a fresh chat and paste:

You're reviewing marketing drafts for [business]. Check each one against these rules: [your voice notes, banned words, legal limits]. Flag any fact or number that has no source. Flag anything that sounds generic. Don't rewrite; list problems only.

A fresh chat matters, because a reviewer that wrote the draft tends to like it.

Step 6: publish, with you holding the button

Let the AI schedule posts, but never let anything publish without your approval. The guide itself says good agents "pause for human feedback at checkpoints". In marketing, publishing is the checkpoint that can't be undone.

Step 7: learn from the results

Once a week, paste your post results (reach, saves, clicks, enquiries) into the AI and ask what the best performers have in common. Treat its answer as a hypothesis for next week's brief, not a conclusion. With a handful of posts, patterns can be chance.

A worked example: one week for a yoga studio

Here's how one cycle might look for a small yoga studio (a made-up example to show the flow).

• Goal: fill the 7 am weekday classes, which are half empty.
• Brief (AI, 2 min; you, 3 min): Audience is office workers nearby. Message: "Start the day before your inbox does." Facts: class times, the price of a 10-class pass, free first class. Avoid: health claims.
• Research (AI, 5 min; you, 10 min opening links): People nearby search for "morning yoga near me" and "yoga for back pain from sitting". You drop the back-pain angle because it edges into health claims.
• Ideas (AI proposes five, you pick three): a short video of the studio at sunrise, a carousel "What a 7 am class actually feels like", a post on the free first class.
• Drafts and review (AI, 10 min; you, 15 min): the reviewer flags "transform your life" as generic. You cut it.
• Publish: scheduled for Monday, Wednesday and Friday at 6:30 am. You approve each.
• Learn: next Monday, you paste the results and ask which post brought enquiries.

That's a week of posts for well under two hours of your time, most of it spent deciding, not typing. (The timings are estimates for the example, not measured results.)

Rules that keep you out of trouble

• No invented facts. Tell the AI it may only use facts you gave it or that it can link to. Check the links.
• Quality over volume. Google's guidance on AI-generated content (February 2023) says its systems reward "original, high-quality content" however it's produced, and that using AI "with the primary purpose of manipulating ranking in search results is a violation of our spam policies". Publishing a hundred thin posts won't help you.
• Label AI imagery where platforms expect it. Meta's approach to labelling AI-generated content adds an "AI info" label when it detects industry-standard AI signals or when people disclose that content is AI-generated. If a photo of your product is AI-made, don't pass it off as a real photo.
• Keep your own voice. If every post reads like it could come from any business, your customers will scroll past. Your notes, opinions and real stories are the parts AI can't supply.