- Patryk Marek
- News
- 1 likes
- 1604 views
- 1 comments
Markdown for Agents is a way of delivering page content to AI agents in a lightweight text format. Content Signals, in turn, make it possible to declare the rules for its use. In PrestaShop, it is worth combining these solutions with up-to-date product data and good SEO. However, each solves a different problem — neither on its own ensures that a store will be recommended by AI.
Markdown for Agents: less code, more information
An agent comparing products needs the name, parameters, price, and availability, not the menu, scripts, or repeated template elements. Models can process HTML, but cleaned Markdown can reduce the amount of unnecessary data. At the same time, it preserves headings, lists, tables, and links.
Cloudflare showed the scale of the difference on its own article: 16,180 tokens in HTML versus 3,150 in Markdown, which is about 80% less. This is the result for a specific page, not a guarantee for every store. The savings depend on the template, content, and the way tokens are counted. See the Cloudflare example.
How to make store content available to agents?
One mechanism is format negotiation: the client sends Accept: text/markdown, and the server handling it returns Content-Type: text/markdown. A regular user still receives HTML. The Vary: Accept header signals to caches that the response depends on the requested format. This is how the Markdown for Agents service in Cloudflare works, converting HTML after it is enabled.
You can also provide separate .md URLs and indicate them as alternative versions of the page. The llms.txt proposal complements this mechanism with a short guide containing a site description and links to detailed documents. It is guidance for the agents handling it, not a replacement for a sitemap or a mandatory standard for all search engines.
Content Signals: search, answers, or training?
Content Signals Policy separates three uses of content that are easy to mistakenly throw into one “AI” bucket:
search— building an index and presenting results with links and short snippets; it does not include summaries generated by AI.ai-input— passing content to a model as material for responses, for example in a RAG mechanism that retrieves information from sources while handling a query.ai-train— training or fine-tuning models on content.
An example declaration allowing search and use in responses, but excluding training, looks like this:
Content-Signal: search=yes, ai-input=yes, ai-train=no
In robots.txt, such an entry should be placed in the appropriate group together with a policy description, while preserving the store’s existing rules. Omitting the signal does not mean consent — it means no declaration in this mechanism. Content Signals are not a technical block on fetching and do not replace access control.
Note for Cloudflare: according to the service documentation, if the origin server does not set the Content-Signal header, Markdown conversion adds yes by default for all three uses. So check not only robots.txt, but also the final response headers.
What should be optimized in PrestaShop first?
The practical order of work should start with the quality of information, and only then move on to the format:
- Describe the product unambiguously. Provide the model, SKU or EAN, units, and variant features. Instead of “high performance,” write specifically what the product can do and what it fits.
- Keep it up to date. The price must include the currency and tax information, and availability must apply to the correct variant. Take cache and index refreshes into account after changes.
- Add shopping context. Publish specific FAQ answers, delivery rules, and return policies. Separate customer reviews from manufacturer parameters.
- Maintain consistency and privacy. The Markdown version should correspond to the public offer in a given language, indicate the original page, and not reveal account or cart data.
- Check the actual response. Compare HTML and Markdown after passing through the CDN. Verify prices, links, and whether an allowed agent receives a blocking page instead of the content.
How does PD Markdown4Agents Pro help?
Our PD Markdown4Agents Pro for PrestaShop prepares documents from store data, not from the entire rendered template. It supports products, categories, CMS pages, and manufacturers; for products, it includes, among other things, prices, availability, features, and combinations. It provides .md URLs and optional format negotiation.
It also generates llms.txt and llms-full.txt — in this module, the latter is an extended index with summaries and links, not a compilation of full descriptions. After enabling the appropriate integrations, it includes the ph_simpleblog blog, FAQ, and reviews from supported modules. It allows you to manage Content Signals in robots.txt. The statistics show content downloads, not the number of citations by AI.
Example: the same product page as HTML and Markdown
On 03.10.2026, we checked the public page of the PD Markdown4Agents Pro for PrestaShop module. A regular browser request returned the product HTML page. The same URL, called with the Accept: text/markdown header, returned a Markdown document with product data. The HTML page also indicated an alternative Markdown version in the Link header: /pl/markdown/product/344.md.
| Checked variant | Result | Practical meaning |
|---|---|---|
| HTML product page | HTTP 200, Content-Type: text/html, Link: rel="alternate" header to /pl/markdown/product/344.md. |
The browser gets a normal product page, and the agent is given the address of the text version. |
The same URL with Accept: text/markdown |
HTTP 200, Content-Type: text/markdown, Vary: Accept, X-Robots-Tag: noindex, nofollow, x-markdown-tokens: 7932. |
Format negotiation works without changing the canonical product page URL. |
/pl/markdown/product/344.md address |
HTTP 200 and the same Markdown document as in the variant with the Accept header. |
An agent that does not send the Accept header can fetch the indicated alternative version. |
Guessed address with the .md ending added to the product slug |
HTTP 404. | You should not assume a simple replacement of the URL ending. The correct address must be taken from the Link header or from the index file. |
The Markdown document header included, among other things, the product type, product identifier, canonical address, Markdown address, SKU, price, currency, availability, categories, and update date. These are the pieces of information an agent needs to read the offer without processing the page menu, footer, and scripts.
This test does not mean that Markdown alone guarantees rankings or AI citations. It only shows that the same public product can have an organized text version, consistent with the canonical HTML page and limited to data available to an unauthenticated visitor.
Does Markdown improve SEO and GEO?
Markdown is a technical convenience, not a promise of rankings. Google does not require special AI files or additional Schema.org markup for presence in AI Overviews and AI Mode. Availability, indexing, helpful content, and data consistency still matter.
In practice, GEO, meaning optimization for generative answers, is worth basing on facts that are easy to find and cite. Keep good HTML and Schema.org structured data in PrestaShop consistent with the offer, and treat Markdown as an additional way to deliver the same reliable information.
Testing HTML and Markdown consistency after an offer change
Choose one product and one service page. Save the HTML address and the corresponding Markdown address. After changing the public description, compare the name, the most important limitation, and the links. For the product, also compare the price, currency, and variant in the same context.
| What to record | Why it helps |
|---|---|
| Status and Content-Type | An error, a blocking screen, or HTML returned at a .md address is not a valid Markdown document. |
| Test date and path through the CDN | They make it possible to separate current content from an old cached copy. |
| Offer requirements and limitations | The text version should not omit conditions visible in HTML or add its own promises. |
Google explicitly indicates that llms.txt and special AI formats are not required for presence in its search. See the current Google guide on generative features. Maintain Markdown for audiences that use it, and evaluate results based on actual visits and citations.
Comments (1)