Claude skills for marketing: what they are and how to use them

What are Claude skills?

A skill is a folder of instructions that Claude picks up when the task calls for it. Anthropic describes it as a file with expertise stored inside. In practice it's a text file, SKILL.md, that sets out how a particular task should be done at your company. You can add reference files to it, such as your brand guidelines, ten good example posts or the format of your quarterly report. Think of it as the onboarding document for a new colleague: you explain how things work once, and after that you point to it instead of repeating yourself every time.

When starting up, Claude reads the name and description of every skill. If you ask a question that matches one of those descriptions, Claude pulls the full skill file into its answer. Anthropic calls this "progressive disclosure", which is also what sets skills apart from saved prompts. Claude can consult the information itself, and, thanks to progressive disclosure, dozens of skills can be switched on without making a conversation slower or more confused.

Skills work in the Claude app, in Claude Code and through the API. You start them with a slash command, or Claude picks the right one autonomously based on what you ask.

Why Claude skills matter for companies and organisations

Beyond speed, skills bring extra benefits. Three things make the biggest difference, in our view:

• The knowledge sits in a file rather than in one colleague's head. The person who knows exactly how you structure a campaign letter writes out the steps once. From then on everyone can use it, even when that person is on holiday.
• Everyone gets the same result. The intern who starts today delivers work with the same structure and the same tone as your content manager.
• If you notice the output keeps making the same mistake, you adjust the file once and it's fixed for everyone. Discussions about tone or content also move from a separate text to one file you can maintain together. Without skills, every colleague improves their own prompt, in their own chat, and nobody else learns from it.

Are skills groundbreaking technology? Not at all. They're really just simple text files. However, they do bring a lot of value: they set down how you do things, which protects consistency, continuity and, to a degree, quality.

How do you use Claude skills?

Step 1: pick a task you repeat every week

Not your hardest task, but your most repeated one. A task that follows the same process every time and that you already know how to run, such as the newsletter, the monthly client report or product descriptions for the webshop.

Step 2: write down how you do it

Describe the steps in the order you take them. Describe what you look at, which decisions you make along the way and how you know it's finished. Add examples of work you were happy with. You can also have Claude write this out while you walk it through the task once.

Step 3: test with a real brief and improve the file

Give the skill a task you have completed yourself last week and compare Claude's output with your own work. Define improvements and write them into the file directly. You should expect a few feedback rounds before you obtain something you can trust.

Creating Claude skills takes three simple steps, and you don't need a developer for any of them. The only requirement is that you know exactly how the work should be done and what the output should look like.

Claude skills examples for marketing

Two examples that fit almost every marketing team:

Example 1: a skill for content in your tone of voice

Feed the skill file with your brand voice, the words you do and don't use, the structure of a LinkedIn post and two or three posts you were happy with. The more examples and material you can add, the better. Marketers who have been in the industry for a while feed their skill every relevant thing they've ever written for the brand, from old newsletters and best-performing social posts to landing pages from past campaigns.

Make a separate skill for each type of text. You write an email differently from a LinkedIn post, and a newsletter differs from a product page. One skill that has to do everything delivers mediocrity everywhere.

Then it's time to test the skill. Ask Claude to write a post about a new case. If your skill is well fed, you should get back a text that follows the right structure and no longer sounds like an average AI post. Fact-checking stays your own responsibility. Always do a final check and edit. The output is probably 80% right. The other 20% is up to you.

Example 2: a skill for competitor research

Include which competitors you follow, what data or KPI's to include and the format the report should come out in. Consider positioning, pricing, messaging and the channels they're on. The sources belong in the skill file too: sites, newsletters, social channels, as well as the timespan you want to compare.

If you have Claude run, for example, a quarterly analysis with this skill, you get an overview in the same format as last quarter. Because the structure is fixed, you see the differences straight away. You can also run the skill daily, so that every morning, you know what has changed in the last 48 hours.

A skill is good at collecting and structuring. The interpretation is up to you. Claude can indicate perfectly well when a competitor is adjusting its prices, but you have to work out what that means for you.

Where it could go wrong

Skills only work well when they're set up well. Here are some common mistakes with a strong effect on the quality of the output:

1. The skill is too broad. A skill that has to do everything does nothing well. Ten small skills that each do one thing up to standards beat one big skill that disappoints every time.
2. The skill contains no examples. Without example material you get average output. This is the most common mistake, and also the quickest one to fix.
3. The skill has steps but no limits. Anyone who reads Anthropic's own skill files will see that a large part consists of warnings about what not to do. Good skills also set down what can go wrong. If you keep getting the same miss back, that deserves a rule in the file.
4. The skill came straight off the internet. There are skills online which you can download as they are. A skill is an instruction to a system that has access to your files and your connections. Read what's in it before you switch it on, and watch out for fake versions of well-known skills.
5. The skill is out of date. A skill needs maintenance. Review the files regularly and check them for outdated information, such as slogans that have changed or nuances that deserve more attention.

How do you start working with skills in your marketing team?

Our tip: start with one task you repeat every week, write down how you do it, and improve the file after every use. That's enough to notice the difference between AI that improvises and AI that follows your instructions.

On 1 October we're running a webinar on Claude skills for marketing. What they are, how to build them yourself and which skills we see in use in practice. For marketers and communications managers; no technical background needed.

Learn more and sign up for the webinar on October 1st.

More relevant cases

• AI is not digital transformation
• Vibecoding for marketers

‍