I tested new Claude "Skills" and I'm not impressed

I tested Claude's new Skills feature so you don't have to waste $17/month.

Watch my full breakdown of Claude Skills here:

A few days ago, Anthropic launched Skills - a feature that lets you build brand guidelines, frameworks, and custom instructions directly into Claude.

Sounds great.

Well, I spent hours testing it with real marketing use cases and here’s what actually happened.

First, you need the Pro plan to access Skills

First thing - this feature costs money. You’re looking at $17/month on the annual plan or $20/month if you go monthly.

If you’re already a Pro user, you can access Skills by going to Settings → Capabilities. You’ll need to enable “Code execution and file creation” first, which then unlocks the Skills section.

By default, everything’s disabled. Claude has some pre-made skills you can try, but I didn’t find them useful. The one you actually want is the “Skill Creator” skill (yes, very meta).

How I created my first skill

I wanted to test something practical - creating graphs using my brand colors and fonts. This is actually useful in my daily work as a marketer.

I told the Skill Creator I wanted to repurpose YouTube video scripts into graphs with my brand colors and fonts. I also wanted to pick between four different concepts from the transcript that could be visualized.

I said my brand color is pink (and uploaded a headshot with a pink background since I couldn’t find the exact hex code quickly). I also specified that my brand font is Poppins with all caps in headings.

The Skill Creator took about 10 minutes to process everything. It figured out the exact pink color, the font specs, and generated a zip file. It also created one example graph for me to check.

The output looked pretty good - it used my font with all caps and my pink color. So far, it seemed to be working.

Testing the skill with real content

After downloading the zip file, I uploaded it in Settings → Capabilities. Once enabled, I referenced the skill in a new chat and pasted a video transcript.

It generated four different graphs. They looked impressive at first glance.

But here’s the problem: The video transcript was about my “best marketing strategy in 2026” where I talked about building a founder brand through daily short-form videos. The video isn’t about numbers or data.

The graphs showed completely made-up information. One showed “expected business growth” with random percentages - something I never mentioned in the video. Another showed fabricated data points that didn’t exist in my content.

This is a big problem. Even though the graphs looked nice with my pink color, I couldn’t actually use any of them. The whole thing was pointless.

I tried to improve the skill

I didn’t give up. I went back to the original chat and added more specific instructions. I told it to use different graph types, and sometimes I just want concepts visualized (not data). I was super clear: Never make up facts that aren’t in the transcript.

Claude generated a new skill zip file with updated instructions. This time I got pie charts and diagrams instead of just bar charts.

I deleted the old skill, uploaded the new one, and tested it again with the same transcript.

But it asked me questions. “Do you want to analyze this transcript and create four graphics?” or “Are you just letting me know about the new skill for future use?”

Why is it second-guessing me? I expected it to already know what I wanted based on the skill instructions.

After saying yes twice, it finally generated four visualizations.

The results:

• First graph: A five-step process diagram. Looked okay, but text was cut off and the “repeat daily” arrow looked weird.

• Second graph: A 100-day commitment timeline. Actually pretty good and usable (though the font was tiny).

• Third graph: Another process flow. Made no sense - arrows going the wrong direction.

• Fourth graph: Risk vs reward comparison. Missing a header for “best case scenario.”

These small mistakes drive me crazy about AI outputs. I don’t want to edit if I’m using AI to save time.

Testing newsletter creation

Next, I tried creating newsletters from video transcripts. I already do this with my Make.com automation workflow (which uses Claude 4.5 via OpenRouter).

I created a new skill using the exact same prompts from my automation - my user prompt about what the newsletter should cover, plus my full system prompt for humanizing AI writing style.

Claude generated the skill. I uploaded it and tested it with the same video transcript.

The output was noticeably different from my Make.com automation, even though both use Claude 4.5 and the same prompts. I honestly preferred the output from my automation.

Then I asked Claude to also add graphs using my graph-creation skill. It found the skill, created four graphs, and updated the newsletter document to show “image here.”

But when I clicked on the images? Nothing. They were empty. The only way to use this was to copy the text, check if the images were worth it, download them separately, and upload them to another tool.

The graph quality:

• First one: Five-minute daily video process. Correct info, no obvious mistakes, but not engaging.

• Second one: 100-day habit formation timeline. Could work.

• Third one: Text too close to the graph. Wouldn’t use it.

• Fourth one: Missing header for one section. Wouldn’t use it.

My verdict: Not worth upgrading for

I upgraded Claude Pro again just to test this feature. But honestly, I’ve been using Genspark.AI for everyday AI questions and tasks.

I love Claude when I’m using it via API for my automations. But I won’t be using Skills.

If you’re already a Claude Pro power user with projects set up and you use it daily, you can probably do something valuable with this. But if you’re thinking about upgrading just for Skills? Not worth it.

I think it’s more valuable to learn to build automations. They don’t have to be complicated, but the outputs will be way more consistent and reliable. You’ll set up specific prompts and dynamic elements that keep changing as you add more information.

The AI world needs more reliability. Automations give you that reliability in outputs (not always, but more than this Skills feature).

This technology will keep improving. At some point, it might make sense to do everything in Claude (or whatever your favorite tool is). But for now, based on these outputs, it’s not worth it - at least not for marketing use cases where you can’t make mistakes and need to build trust with reliable content.

Keep Skills if you’re already a Claude power user and Pro plan owner. Otherwise? Not worth it.

---

Want my exact content automation blueprints?

📌 Join AI Marketing Masters to get the shortcuts to all the AI automations I’ve built: https://aimarketingmasters.com/