LLMs.txt AI-generation

Fig. 1 — General view of the IP5 Agency — AI llms.txt Generator module in the CS-Cart admin panel.

IP5 Agency — AI llms.txt Generator: add-on for CS-Cart and MultiVendor

The add-on is available in 2 languages: Russian and English.

Customers increasingly ask AI assistants instead of traditional search engines. To appear in an AI-generated answer, an online store needs to be understandable to machines: what it sells, how its content is structured, and where important information can be found. The add-on generates three machine-readable formats for your store — llms.txt, llms-full.txt, and page-level Markdown — and helps keep them up to date. Unlike template-based generators, the main file is written by a real language model such as Claude or another provider compatible with the Anthropic Messages API, including OpenRouter. This allows the result to remain coherent and readable instead of becoming a mechanical export of database fields.

How the add-on works

The add-on is based on three main components:

• llms.txt — a concise AI-generated description of the store, including what it sells, key categories, featured products, contact information, and recommendations for language models on how to use the file.
• llms-full.txt — a complete structured export of the catalog, including categories, products, pages, and blog posts. It is generated directly from the database without calling an AI model, making the process fast and free.
• Page-level Markdown — any product, category, or CMS page can be accessed in clean Markdown format by adding .md to its regular URL, for example example.com/product-name.md. The content is generated on request, so nothing needs to be prepared in advance.

Each AI generation first creates a draft. The draft is not available through the public URL until an administrator reviews it in the admin panel and clicks Publish. This prevents an unsuccessful or low-quality AI response from immediately replacing the live file.

Compliance with the llms.txt standard

The llms.txt format the module generates isn't arbitrary — it follows the official specification from llmstxt.org line by line: a first-level heading, a short description as a blockquote, second-level topic sections with markdown hyperlink lists, and a dedicated Optional section for secondary material — exactly as described in the spec. Every link to a category, product, page, or blog post is a real markdown hyperlink [Name](url), not just plain text next to a URL — that's an explicit requirement of the standard, and we verified it against the reference example line by line.

The module also follows the guidance in CS-Cart's own documentation on Optimizing your site with LLM: the short llms.txt stays navigational (an optimal link count of 20–50, not the full catalog), while the entire product and category listing with descriptions moves into the separate llms-full.txt file — exactly the split the platform itself recommends.

1. Installation and initial setup

Go to Add-ons → Manage Add-ons → Install add-on, upload the add-on archive, and activate it. After installation, the add-on will be available under Marketing → IP5 Agency — AI llms.txt Generator.

Open the add-on's Settings tab and enter the API key and endpoint before starting AI generation. Without an API key, manual editing and the non-AI llms-full.txt file remain available.

Fig. 2 — Upper section of the add-on settings on the Settings tab.

Fig. 3 — Continuation of the add-on settings list.

Fig. 4 — Lower section of the add-on settings list.

2. General settings reference

The table below describes the settings available on the add-on's Settings tab.

3. Content generation and publication

Open the add-on management page under Marketing → IP5 Agency — AI llms.txt Generator. The page provides the main tools for generating, reviewing, and publishing the files.

Fig. 5 — Add-on management page with the main controls.

Fig. 6 — File status and generation information on the management page.

Click Estimate volume to see how many categories, products, pages, and blog posts will be included in the generated file, as well as the approximate number of input tokens before using the API budget.

Fig. 7 — Initial state of the management page before generating and publishing new content.

Click Generate llms.txt with AI. This creates a new draft while leaving the currently published file unchanged.

Fig. 8 — Management page while llms.txt is being generated.

After generation is complete, a draft containing the new text becomes available for review before publication.

Fig. 9 — Generated llms.txt draft before publication.

After reviewing the draft, it can be published. Publication updates the public llms.txt URL.

Fig. 10 — Management page after content generation and publication.

4. Manual content selection

Enable Manual selection in the settings to switch from quantity limits to selecting specific objects. On the management page, you can choose the categories, products, pages, and blog posts that should be included in the generated files.

Fig. 11 — Manual selection of categories, products, pages, and blog posts for inclusion in the generated file.

5. Manual editing

Both files can also be edited directly as regular text and saved manually. Changes made this way are published immediately without going through the draft stage.

Fig. 12 — Manual editing of llms.txt and llms-full.txt content.

6. Page-level Markdown

When the corresponding setting is enabled, any product, category, or CMS page is also available in clean Markdown format by adding .md to its regular URL — for example, example.com/product-name.md. The Markdown content is generated on request in real time without additional actions in the admin panel.

Fig. 13 — Example of a product page opened in Markdown format.

7. Scheduling

There are two ways to keep the generated files up to date without manual actions:

• Built-in soft scheduler — enable it in the settings and specify the required time in HH:MM format. The add-on regenerates and publishes the file when the next storefront request occurs after the specified time, no more than once per day. Server cron access is not required.
• External cron — if a CRON key is configured, an external request can be used to start generation according to a schedule through a system scheduler.

8. Generation log

Every generation attempt, whether successful or unsuccessful, is recorded with a timestamp, status, token usage, and an error message when applicable. The log can be used to monitor generation history and identify problems when communicating with the API.

Fig. 14 — Generation log with information about completed attempts, statuses, and token usage.