It’s Sunday night and your content calendar for the week is staring at you. Monday morning needs a blog post. Tuesday needs a newsletter. There are three LinkedIn posts you said you’d write, an Instagram caption you owe the photographer, a sales email sequence that has been “almost done” for two weeks, and a competitor who keeps showing up in front of your customers. None of it is actually written yet. You’d love to delegate this whole thing, but the budget for a marketing hire is the budget you don’t have.

I’ve sat in that exact chair, and almost every small business owner I work with at Markham Square has too. You don’t have a marketing department. You don’t have a copywriter on retainer. You probably are the marketing department, fitting it in around running everything else. And you keep hearing that AI is supposed to fix this for you, but every guide you read either pitches you a tool or hands you a one-line prompt and calls it a day.

What you actually need is a clear picture of where AI fits into the marketing work you’re already doing, and the workflows to make it stick. The good news is that this is way more doable than the noise around it makes it sound. You don’t need a new marketing strategy. You need to plug AI into the one you already have. We’ve been running this at our shop, and what used to take us 20 hours a week of content production now takes 1 to 2.

This guide walks you through how we did it: not by pasting better prompts into ChatGPT, but by codifying our marketing workflows into reusable AI skills that run the whole job from brief to publish. The post you’re reading was produced by that same workflow.

Stop chasing prompts. Start building skills.

Quick framework before we go further. We map AI adoption to four levels: thought partner (you’re in the chatbox, exploring ideas and getting one-off help), assistant (you’ve given the AI a skill to run a specific task end to end), true teammate (skills are chained into workflows, often on a schedule, with you reviewing the output), and system (the workflow runs your business with minimal supervision).

Most “AI for marketing” advice keeps you stuck at thought partner: open ChatGPT, paste in a clever prompt, copy the output back out, edit it, ship it. That works fine for a one-off. It does not scale. A week later you’re back to typing the same setup into the same chat box, getting subtly different results, and wondering why your time savings keep evaporating.

The assistant level is where the actual leverage lives. Here, you stop typing prompts and start building skills: documented, repeatable workflows that capture every step of how you do a task, from the inputs you need to the format you want the output in, and that you invoke once to run the whole thing end to end.

A skill is not a tool you buy. It is a markdown file (or its equivalent in your AI platform) that tells the AI exactly how to do a job: what to ask you for, what steps to run, what reference material to lean on, what format to return. Once it exists, you can run it tomorrow, next week, and six months from now and get the same quality every time. New person on your team? They run the same skill. The skill is the institutional memory.

This is the move that took us from 20 hours a week of marketing content to 1 or 2. Not better prompts. Codified workflows.

How to find a workflow worth codifying

Skills only pay off when the workflow under them is repetitive. The first task is finding the right candidate.

Look at your week and rank your marketing tasks on three things:

Frequency: how often do you do this? Weekly beats monthly.
Time per occurrence: how long does each pass take? An hour beats fifteen minutes.
Pattern density: how similar is each pass to the last? Drafting a sales follow-up email is high pattern density. Writing a quarterly investor update is low.

The winners are almost always the same handful: blog post drafts, social repurposing, sales follow-ups, newsletter sends, competitor scans. They run every week, they eat real time, and the structure barely changes from one to the next.

Pick the highest-scoring one and start there. Resist the urge to codify everything at once. One skill, well-built, beats five half-finished ones.

What a skill actually looks like

Before we walk through the marketing skills we run, here’s what is inside one of them so you can see the shape of the thing.

We have a skill called long-form writing that runs every blog post, newsletter, and pillar guide we publish. You give it a brief; it gives you back a finished, edited, voice-matched draft ready for review. Under the hood, it’s a single markdown file with these sections:

The intake step: what the skill asks you for before it does anything (audience, topic, keyword, the angle, any source material).
The research step: how the AI finds and verifies sources, with a hard rule that every statistic gets primary-source-checked.
The drafting steps: four passes that each produce a separate file. A direct draft to nail the substance. A prose draft to make it flow. A voice draft that pulls in our voice reference document. A humanizing pass that strips out AI-sounding patterns.
The reference material: style guides for our voice, examples of writing we like, the rules we want enforced.
The output format: title, meta description, exact file location, what to do when it is done.

Total length: about 250 lines. Total time to write the first version: 20 minutes. Time saved every week since: hours.

You do not need Claude Code to do this. The same idea works as a Custom GPT in ChatGPT, a Project in Claude, or just a saved prompt template you keep at the top of every chat. The format does not matter. The discipline of writing it down does.

The marketing skills we actually run at Markham Square

Here is the stack of skills currently running our marketing, what each one does, and where to go deeper on it.

The content brief skill

The intake skill. You give it a topic and the target keyword. It outputs a complete brief: audience, search intent, angle, recommended outline with section word counts, must-includes, must-avoids, internal links to existing content, and a competitor benchmark table. About forty-five minutes of work compressed to five.

The long-form writing skill

The producing skill. Takes a brief and runs the full multi-pass workflow described above. Outputs a final, edited, voice-matched draft. The post you’re reading right now came out of this skill.

The repurposing skill

The amplifier. Takes a finished blog post and outputs the LinkedIn post, the Twitter thread, the Instagram caption, the newsletter excerpt, and the email teaser, each one tuned for the platform’s tone and length. One source piece becomes a week of distribution in about ten minutes.

The competitor scan skill

The intel gatherer. Pasted competitor URLs in, get back a structured analysis of positioning, pricing signals, messaging themes, and content gaps. We run it monthly to keep our positioning honest.

The publishing skill

The capstone. Takes a finished post out of the writing skill, transforms it into the format the site expects, drops it in the right directory, and moves the draft to the published folder. This is the skill that closes the loop. Without it, you’d still be doing the last-mile copy-paste-and-format dance every time you ship.

Why a human stays in the loop

None of these skills are autonomous. Every one of them stops at a review point and waits for a human (usually me) to read the output, edit it, and approve before it moves on.

This is not a workaround until the AI gets better. It is the design.

The skill takes a job from “I have to do every step” to “I have to review every output.” The cognitive load drops by an order of magnitude, but the judgment stays human. Voice questions, factual claims, brand calls, “is this actually good?” — those are still your job. The skill just hands you an editable draft instead of a blank page.

When we cut content production from 20 hours to 1 or 2, the 1 to 2 hours that remain are almost entirely human review and editing. That is the work we want to keep doing. The 18 hours we got back is the part nobody wanted to do anyway.

If a skill ever produces output you can ship without reading, the skill is too narrow or your standards are too low. Fix one of those before you trust it more.

Build the publishing skill last (this is the actual unlock)

Here is the move most people miss. Once you have a skill that produces good output, you are still doing the manual work of shipping that output: pasting it into your CMS, formatting the markdown, adding the meta description, uploading the image, scheduling the social posts, hitting publish. That last mile can be 30 to 45 minutes per piece. Without a skill for it, your producing-skill’s time savings get eaten by the publishing tax.

Building the publishing skill is what closes the loop. For us, it takes a finished post in our content drafts directory and:

Reads the title, meta description, and category from the file.
Reformats it for the site’s expected structure.
Writes it to the right folder in our codebase.
Moves the draft folder to the published archive.
Reports back what shipped.

It is unglamorous. It is also the difference between “AI saves us a few hours” and “AI runs our content marketing.” Build the producing skill first, get it working, get the human-review pattern down. Then build the companion publishing skill. Now you have a workflow that actually goes from idea to live in under two hours total.

How to apply this to your marketing this week

Here is the framework, end to end:

Pick the workflow. Look at your last four weeks. Find the marketing task you did the most times that took the most cumulative hours. That’s the one.
Document it manually. Write down every step you take. Every input you collect. Every decision you make. Every reference document you check. Don’t optimize yet, just record reality.
Codify it as a skill. Turn that document into instructions for the AI. Spell out the inputs to ask for, the steps to run, the references to use, the format to return.
Run it. Review. Iterate. Use the skill on a real piece of work. Note where it fell short. Update the skill. Run it again. Three or four cycles in, it will start producing output you trust.
Build the publishing companion. Once the producing skill is solid, build the skill that handles the last mile.
Move to the next workflow. Pick the next-highest-scoring task. Repeat.

That is the playbook. There is no step where you sit down and write a clever one-off prompt.

The tools that make this work

You do not need a 40-tool stack to build skills. Here is what we actually use.

ChatGPT or Claude: the assistant that runs your skills. Either is fine. ChatGPT has Custom GPTs and Projects; Claude has Projects and (if you’re using Claude Code) skill files. Pick one and live in it.
Wispr Flow: voice input. Wispr Flow lets you dictate into any app on Mac, Windows, iOS, or Android. When you’re writing a skill, talking your instructions in is two to three times faster than typing them. This one came in clutch for us.
Your existing marketing tools: email platform, CMS, social scheduler, analytics. Skills sit on top of these, not in place of them.

When you’re ready to chain skills together end-to-end (the producing skill feeds the publishing skill automatically, no human-prompted handoff), look at OpenAI’s AgentKit and Agent Builder, launched at DevDay 2025. It gives you a visual canvas for stitching skills into a single agent workflow. You don’t need it on day one, but it’s the natural move from assistant to true teammate once your skills are humming.

Common mistakes when you’re building skills

Trying to codify a workflow you’ve never done by hand. Skills are bottled experience. If you don’t have the experience, there’s nothing to bottle. Do the task manually three or four times before you try to write a skill for it.

Building one giant skill instead of small composable ones. A skill that “writes the post AND repurposes it AND publishes it” is harder to fix when one step breaks. Three small skills you can call in sequence are easier to maintain.

Skipping the human-in-the-loop step. The skill is supposed to hand you an editable draft, not a fait accompli. If your skill ships without you reading the output, you’ve built an unguided missile.

No measurement. Pick one metric per workflow (post-publishes-per-week, click-through rate, reply rate, time-to-ship) and track it before and after the skill exists. If the number doesn’t move, the skill isn’t paying for itself.

Treating the skill as finished. Skills decay. Voice shifts, what works on social shifts, what your audience cares about shifts. Plan to revisit each skill every quarter. Update the references. Tighten the steps. Cut what no longer fits.

FAQ

Do I need Claude Code or some technical setup to build skills?

Nope! You can build a skill just by talking to Claude or ChatGPT.

How long does it take to build a useful skill?

The first version of a focused skill takes a long afternoon. Getting it to the quality bar where you trust it takes another three or four real-use iterations spread over a couple of weeks. Total investment: maybe ten hours. Time saved: hours every week, every week, indefinitely.

What if my workflow changes?

Update the skill. Skills are living documents, not static automations. Plan a quarterly review of every skill you have running.

How much time can I realistically save?

It depends on the workflow and the volume. For our content production at Markham Square, we went from about 20 hours a week to 1 or 2. Your numbers will depend on how much marketing you produce and how disciplined you get about codifying the parts that repeat.

What about brand voice? Won’t AI flatten it?

Only if you don’t feed it your voice. Every one of our skills includes a voice reference document with examples of writing we love, sentence patterns we use, and patterns we avoid. The skill is told to match this on every pass. After three or four iterations, the output reads like us.

Where should I start?

Pick the one marketing workflow that eats the most of your week, document it by hand, then codify it. Don’t try to build five skills at once. Get one to the point where you trust it, then build the publishing companion, then move on.

The Wrap-Up

Most “AI marketing” advice will leave you typing prompts into a chat box for the rest of your life. That is the thought partner level, and it tops out fast. The actual unlock is the assistant level: write your repeatable marketing workflows down as skills, run them with a human in the loop, and build the publishing skill that closes the loop. From there, chaining those skills together gets you to true teammate, and that’s where AI starts running real parts of your marketing instead of just helping with them. That is what took our content marketing from 20 hours a week to 1 or 2, and it’s how every small business I know that’s getting real leverage out of AI is doing it.

Pick the workflow that hurts the most. Codify it. Run it. Review it. Iterate. Build the companion publisher. Then do it again with the next one. That’s the whole playbook.

As always, feel free to ask me any questions on LinkedIn. And until next time, happy marketing!

How to Use AI for Small Business Marketing