What Agentic Web changes, and what it does not

Most product pages were built for a person with a mouse and a fast connection. An AI crawler gets the same page: a few thousand lines of markup, a cookie banner, tracking scripts, a carousel, and specifications that appear only when someone clicks an accordion. Some of those fetchers execute JavaScript, some do not, and none of them publish exactly what they render. Agentic Web exists to remove that uncertainty for the pages you care about. This note says precisely what it does and, at equal length, what it does not.

What it does: approved Markdown to verified crawlers

When a verified AI crawler requests a page you have enabled, your CDN returns a compact Markdown version of that page instead of the HTML. Everyone else, human visitors and unverified bots included, gets the normal page. The Markdown is generated from your own page, reviewed and approved by your team, and served with a verified date and the source URL on every copy.

Verification uses what each engine publishes about its crawlers: the documented user agent, plus the published IP ranges or reverse DNS where the provider offers them. A request that merely claims to be GPTBot is not verified and gets the normal page. The served content only ever reaches crawlers the engines themselves say are theirs.

# Gratwand Pro HS Hooded Jacket

Waterproof 3-layer hardshell for alpine climbing.

- Price: EUR 649 · In stock · Ships from CH, DE, US
- Weight: 470 g (size M) · Fabric: 3-layer, 28,000 mm hydrostatic head
- Fit: athletic · Sizes: XS to XXL
- Hood: helmet-compatible · Pockets: 2 chest, harness-compatible

## Best for
Technical alpine routes, sustained rain and wind, long days with a harness on.

## Care and repairs
Machine wash cold, no softener. Free repairs for 10 years through Nordkamm Repairs.

Last verified: 2026-08-04 · Source: https://www.nordkamm.example/gratwand-pro · Version: 12

Accuracy: only approved content

Nothing in the served Markdown is generated beyond what is on the source page. The extraction is a reduction, not an elaboration: the same product name, the same price, the same specifications, stripped of markup and navigation. A person approves each version before it is served, and the approval is recorded with who and when.

Every served page carries the date the content was last verified and the URL it was derived from, so the facts an engine stores stay dated and traceable to your own page.

Freshness and rollback

Nothing reaches an agent until a person has approved it, and a half-finished edit on your site never does. If a wrong price or a bad description was approved, an earlier approved version can be restored immediately, and every served page names the version it came from.

How it is measured: served, then cited

Two counts, kept apart. Served is a fact from your CDN logs: which verified crawler fetched which enabled page, and when. Cited is a fact from stored answers: whether that page’s URL was displayed as a source, on which engine, in which market, on which date. Bluemoon shows both side by side for each enabled page, with the counts.

What we do not do is turn those two counts into a causal claim. “Served 214 times, cited in 9 answers” is two observations. Whether the serving caused the citing is not observable from either, and we do not print a rate that implies it did. If you see a page move from never cited to cited after enabling it, that is worth noticing and worth testing against a page you did not enable. It is not proof.

What it does not do

• It does not execute JavaScript for you. The Markdown is derived from content reachable on the source page. If a specification exists only after a client-side call, it has to be made reachable before it can be approved and served.
• It does not change ranking. Not in search, not inside an engine’s retrieval. It changes what a verified crawler reads on the pages you enable, nothing else.
• It does not promise citations, ranking or conversions. We make no claim that delivery alone moves any of these. The measurement above exists so you can see what happened, not so we can assert it in advance.
• It does not override robots.txt. A crawler you block never asks, so it is never served. Readiness comes first; the free readiness check takes about 20 seconds.
• It does not touch human visitors. Unverified requests receive the normal page, always.
• It does not decide what an engine does with the content. Retention, weighting and display remain the engine’s, and we describe them only from what stored answers show.

Setting it up, and where to start

The integration is a small piece of configuration at your CDN that recognises verified crawlers and routes their requests for enabled pages to the served version. The setup guide linked from the Agentic Web page walks through it for the platforms we support, and your team keeps full control of which pages are enabled. Start with the pages that answer buying questions: product pages, comparison pages, pricing, availability. Enable a handful, approve the Markdown, watch served and cited for a collection window, and only then widen.

If you take one line from this note: Agentic Web changes what agents can read, not how they rank you. Everything else is measurement.