seo-text — an SEO skill for Claude

The whole cycle runs automatically in a few steps.

1. You set a Serpstat token, the main keyword, the language and the page type;
2. The skill pulls the results top, related queries, suggestions and questions from Serpstat;
3. It crawls competitor pages from the top and extracts texts and headings;
4. It runs the data through the pipeline: set-cover keys, LSI, collocations, structure, length;
5. It writes the final text by the prompt document and checks it against a checklist.

At the output you immediately see the keyword plan and the structure, and then the finished text — you can step in at any stage.

What the skill gathers from the results

Text quality is defined by data, so seo-text is not limited to one source. From Serpstat it takes frequency, related keys and competitor queries, from the crawl of the top the real texts and heading structure.

• The organic top and competitor pages for analysis;
• LSI and meaningful words from the leaders' texts;
• Collocations and the H1–H3 structure by the topics of the top;
• A target length by the median of the results;
• Question queries (PAA) for the FAQ block.

The same data underlies the copywriter brief and keyword research in the desktop app.

What you get

The result is not a wall of text but a ready-to-paste block. The skill returns a single HTML with the main text, a visible FAQ and valid JSON-LD FAQPage, plus the Title and Description meta tags by SEO norms.

Additionally a result card is assembled: in it you can conveniently view the text, copy the meta and the HTML itself. The tag structure is limited to a safe set (headings, paragraphs, lists, a table), so the material slots cleanly into any CMS.

Languages and page types

The skill writes in Ukrainian, Russian and English. English is full English-SEO with density in percent and readability rather than a machine translation from Russian. Each language has its own norms and cliché ban-lists.

The page type is chosen for the task: a store (categories, products), a blog (articles, guides) or a service (services, landings). For each type there is its own prompt document with rules of structure and length.

Built-in self-check

The main difference of seo-text is that it checks itself. After writing, the text is run through a checklist: keyword density, length, the cliché ban-list, the heading structure, the presence and correctness of the FAQ and meta tags. Violations are fixed before output.

So you get material that passes checkers without rework in circles rather than text you have to clean by hand afterward.

The skill works inside Claude Code as a skill: it is enough to install it and run it with a command. From the outside, only a Serpstat token is needed to gather data. There is no need to install a separate application — everything happens in your working environment.

If the desktop format is more convenient, the same engine is available as the SEO Bundle app on the home page. The logic and quality are the same, only the interface differs.