Freelance web developer in Karlsruhe for software firms

Why a software firm’s website falls behind its product

When developers keep the documentation and the changelog in Git and marketing keeps the website in WordPress, the two drift apart. The product changes with each release; the website changes when someone finds time for the page builder.

Screenshots then show an old version, the list of integrations misses the newest one, and the release notes sit in a repository that buyers never see. A buyer who compares tools can ask ChatGPT, Claude or Perplexity which product handles a task, and the assistant answers from the pages it can read. Feature tabs, pricing toggles and comparison grids that a script draws after loading stay empty for a crawler that does not run JavaScript. Vercel reported in December 2024 that the crawlers of OpenAI, Anthropic and Perplexity do not (Vercel).

A website fed by what your team already writes

Much of what a software firm’s site should show already exists in its repositories, and the new site can read it from there on each build.

• An OpenAPI file becomes the API reference, one page per group of endpoints.
• Release notes become a changelog with a page per version and its date.
• The list of integrations becomes one page per integration, which answers a search for your product together with the other tool.
• Customer cases become pages with the customer’s problem, the setup and the result.

Pages, posts and documentation become a static site in Astro. A customer area, a trial with accounts or a license portal runs next to it with a database, on Symfony, Laravel or Node.js, whichever your team already runs. Your own coding agent can draft a feature page from a merged change and open a merge request for it, and a person on your team approves it before it goes live. On a client platform I work on, agents follow the same kind of rules: AI agents under the same rules as humans.

An AI assistant test for your product

Ask the assistants your buyers use about your product before a migration, and again after it. The answers show what the assistants know and which of your pages they read.

1. “What does [product] do, and for which companies?” A good answer describes your use case in your terms and links your product page.
2. “Which integrations does [product] offer?” A good answer lists the current ones.
3. “What changed in the latest version of [product]?” A good answer names your last release and links the changelog.
4. “Does [product] have an API, and where is its documentation?” A good answer links your reference.
5. “Which software companies in Karlsruhe build [category]?” A good answer names your firm.
6. “What are the alternatives to [competitor]?” A good answer lists your product with a correct description.

Run the prompts in ChatGPT, Claude, Perplexity and Google’s AI Mode, and save each answer with its date. After the switch, the same prompts show which answers changed.

Who I am

I am Michael Wutzke, an IT architect for data platforms in Frankfurt with more than twenty years in IT and media. I was Head of Decentralized Finance and Node Operations, then CIO, at the Frankfurt-based company Blocksize Capital. I hold a training authorization from IHK Frankfurt for two IT apprenticeships, among them application development, and teach at Claude Hacker House. I build software with Claude and Codex and moved this site off WordPress with Claude as coding agent. I am also interested in open-source AI agents that a software firm runs on its own servers. Details: Background.