6 the continuation of the evaluation of the content that will be created in Claude

Your existing content can be gold if you know how to improve it. Finding the time is a challenge in itself. I’ve found that using Claude makes what can be a daunting task more manageable.

If it sounds difficult, you don’t have to build a large content testing workflow from scratch. With Claude, you can start by researching one article, iterate on it, and create skills that you can reuse and refine over time. I went from one study to a full library of skills that I refine every time I use them.

Claude can help you uncover trending posts, flag outdated information, check brand voice, and more. You just need to invest the time in repetition so that the number of your Claude skills is compounded.

Here are six types of content tests you can use with Claude. The first four work at the article level, so you can start with one clip today.

Page rank testing

If you’re not comfortable creating skills or workflows, page-level research is a good starting point. These four tests run on a single article, with no table of contents, no data export, and minimal setup required. At the end of each session, ask Claude to develop a skill that you can use again in future page searches.

1. Brand voice consistency

Content libraries move over time due to changes in brand voice, staff, products, and services. A brand voice consistency check can help you identify what needs to be revised so that the piece better aligns with your brand’s guidelines.

Unless you have detailed product guides with lots of examples, let Claude deliver your voice guide to high-quality content. This helps remove the disclaimer language common in many product guides, such as “conversational but authoritative” or “informative, not too structured.”

Choose three to five articles as your common carriers. Download them as download files if you can, then send them to Claude immediately and ask him to explain:

• How do articles usually open (for example, do they open with a direct claim, a counter-statement, or a concrete situation?)
• Sentence structure and paragraphs: average length, width, when sentences get longer or shorter, how paragraphs tend to close, and so on.
• Personality dimension: three to five “We say X, but not Y” pairs, each with an active and passive example.
• Vocabulary: words and phrases to use, and words and phrases to avoid.
• This has never been done by this genre: certain constructions, phrases, and principles are not present in all pieces.

Instead of “conversational but authoritative,” you should get:

• Marked as “Your essays open with a direct claim instead of a descriptive paragraph, sentences are between 15 and 20 words and rarely exceed 30, the transition is active (ie, ‘here’s why it’s important’) instead of formulaic (‘more than that’).”
• Examples of pairs like “We can say ‘the data show three things,’ not ‘there are many factors to consider.'”

The goal is not to produce a voice guide that your writers can use. To build one that an LLM can understand and use.

Once you’re happy with your output, have Claude save it as a skill and check out the article about it. If he calls things you don’t agree with, have Claude review the skill until you’re satisfied with the output.

You can use it to find brand conflicts in old content, test new content for alignment, and even start producing branded content, though you’ll still want to edit it.

Dig deeper: How to train Claude to sound like your brand

2. Comparative comparison

If you need to improve content performance, cover comparison can help you spot gaps in the headlines in your articles.

Use the Claude extension in Chrome to have Claude scrape content from the top three to five pages that rank for your target keyword. Ask Claude to analyze your content against scrap content and highlight any gaps. You want to know:

• What your competitors are doing well.
• You’re good.
• Where you can improve your piece.

If you select this information from the table, tell Claude to extract it that way. Or, if you’d like to have a downloadable DOCX, request that.

If Claude recommends things that you would never add to your content, be aware of them when you include the information in the skill.

3. New research

Old content adds up, and it’s hard to focus on updating it when you’re also creating new content. Creating a skill that identifies what needs to be updated so you don’t have to read every old article can save you a lot of time.

Feed Claude an old article and ask him to flag everything sensitive at the time: statistics with years attached, named instruments or platforms, “current” or “latest” trend indicators, and any claims that depend on a specific market or regulatory context that may have changed. You are not asking it to rewrite anything. You are asking for a list of issues so you can review them.

If you have new products that may not have been covered in older articles, discontinued products, or similar changes, include that context in your input so Claude can flag content for removal or addition.

Dive deep: How to turn Claude Code into your SEO command center

4. AEO and AI recovery

Answer engine optimization (AEO) focuses on making content visible in AI-generated responses. AI tools, including ChatGPT, Perplexity, and Google AI Overviews, often present content that directly answers questions. If your article buries the answer under three paragraphs of introduction, or builds key information in a way that is difficult to extract, it is less likely to appear in those answers.

Give Claude an article that matches the target question and ask him to check it:

• That the piece answers the question directly and prematurely. LLMs scan from top to bottom and measure the maximum opening.
• Whether the key statements are clear enough that you can quote them directly. LLM is less likely to issue or cite ambiguous language because it is difficult to verify.
• That’s where an FAQ-style section can help.
• Whether the page has clear authority signals, such as outbound links to primary research, first-person information, or specific examples.

Save this as a skill, and it becomes an additional editor to help improve AI visibility.

Library quality assessment

This test requires access to your performance data or content inventory, either through a connector or by manual transmission.

5. Performance evaluation

When people think of content testing, they often think of analyzing a library of content to uncover performance issues. Performance analysis is exactly that type of research.

Before you start, make sure you have access to your data through a connector like BigQuery or the Semrush API. Otherwise, export the data you normally use to evaluate high-value content (traffic, clicks, engagement metrics, conversions, rankings, etc.) as input.

Ask Claude to prioritize: pages that have experienced significant performance declines in the past six to 12 months, pages that have high visibility but consistently low click-through rates, and pages that have been alive long enough to rank but never have.

Be specific about what constitutes a slowdown in your website’s performance, as traffic varies across sites and industries. Ask Claude to come up with a priority list of what to investigate and why. From there, the page-level analysis above provides a diagnostic framework.

If you’ve done this analysis before, give Claude the previous output so he can better understand what you’re looking for.

Dig deep: How to build a powerful second brain Claude Code for agency work

6. Topic gap analysis

Entities are a big part of AEO and semantic search. A topic gap analysis can help reveal whether you have enough content to build authority for companies related to your product.

The basic question is: Which library of your content to cover should be?

To get started, you’ll want a list of those organizations as part of your installation. For example, at my agency, we want to be known for SEO and AEO. If you have a clear list of services or products, you can use that instead of a business list.

Using Cowork or Code, have Claude analyze your sitemap and compare it to your target organizations. Alternatively, if you have a Screaming Frog export with URLs, page titles, and meta descriptions, use that as input for accurate analysis.

Ask Claude to identify underrepresented or missing subject areas based on the businesses or products you target. If you want a ranked list, you can use Semrush MCP for Claude to check the search volume for potential keywords.

Not every gap is worth filling. Filter the results to suit your audience’s needs, and re-emphasize any changes you make to Claude to create the ability to match the output you want. Your playlist can play directly into your content creation workflow or be passed on to your content team.

Don’t try to check everything at once

Content testing has stalled not because teams don’t have data, but because the scope feels too big to start. Choose one test and one topic. Run it, save the skill, and use it again in the next episode.

For me, repetition is part of the fun. I enjoy taking a skill and developing it, then combining it with other skills to uncover other content opportunities. I hope you can build one of these this week.