Can AI agents buy from German pharma and healthcare brands?

Five agent classes, one job per brand, one wave.

The five classes span the full spectrum a researcher's or patient's AI agent might run on: a plain reader doing an HTTP fetch with no JavaScript, a search assistant routing the question through an AI search engine, a coding agent driving Playwright over the Chrome DevTools Protocol, a full browser used by computer-use agents, and an autonomous operator that completes the order or booking on its own. [S1] The search-assistant row is scored separately as AI Visibility and is queued for the next enrichment pass.

Each brand was tested at the close-state its surface actually supports. The B2B brands close on a cart. The care networks close on a booking or contact form. Bayer closes on a pharmacy handoff. The tasks differ; the method of measuring whether five agent classes reach the close is identical.

One task per brand, three AI providers. Qiagen carries Confidence B; the other four carry Confidence C, each grounded on a single task at its own close-state. Enough to ground each surface on a clear job, not to judge a brand's whole catalogue. Every score below carries that one scope. [S5]

The agent runs are point-in-time. WAF and bot rules fire dynamically, so a surface measured in April 2026 may read differently on a re-run. Treat the scores as directional, not precise rankings.

Before the five brand stories, the whole wave on one grid. Read each column top to bottom: only one brand answers every agent class.

Only Qiagen is filled top to bottom. The split is about what an agent runs on. All four other brands close for a browser; below the browser, a wall appears. Sigma-Aldrich and aspirin.de block the plain reader and the coding agent outright. Helios and NephroCare let them see the place but not the match. And Bayer is the only brand a browser still cannot buy from: the right product sits in a legal footer with no button beneath it.

Qiagen is the cleanest agent surface in this wave. A plain reader fetches qiagen.com and finds the catalog number 874611, the price €5,480, the 42 EGFR mutations the kit detects, and the CE-IVD classification, all in the initial HTML, no JavaScript required. A coding agent gets the same content via curl. Three browser runs each completed an add-to-cart on the therascreen EGFR Plus RGQ Kit with no login. [S4]

The most specialised diagnostic in the wave is the one an HTTP fetch can buy.

The architecture choice, Angular Universal server-side rendering with Schema.org Product JSON-LD embedded, is exactly what a machine reader wants. No special agent-readable layer was added. The catalogue itself is the agent surface, because the catalogue itself is server-rendered.

The catch sits one step upstream of the surface: broad category questions almost never surface Qiagen at all. Expert intent finds the brand; mass-market intent doesn't. Every agent that reached the page completed the cart; the only thing between Qiagen and a clean score is the discovery layer above it.

Score: AI Visibility 11.51 · AI Usability ~80 · Composite 6.8 / 10. [S4]

Sigma-Aldrich, the Merck Life Science surface, runs one of the cleanest B2B product catalogues in the DAX 40, and the simplest agent classes never see it. A plain reader's connection aborts before any response lands. A coding agent's curl returns HTTP 000, the code for a connection that never opened. The browser passes three of three runs at score 96.3, identifies TRI Reagent BD T3809 for whole-blood RNA at 200µl, and adds three pack sizes to cart without login. [S3]

A €154 reagent you can order with a browser, invisible to anything simpler.

The catalogue is built. The protection layer above it deletes half the agent classes before they see anything. A specifier who lets ChatGPT do the lookup against a text-only fetcher gets routed to Qiagen, Thermo Fisher, or Bio-Rad by default, because Sigma's content is not reachable on the HTTP path.

Score: AI Visibility 38.33 · AI Usability ~54 · Composite 5.2 / 10.

Helios runs Germany's largest private hospital network, and the booking portal actually works. A browser navigates from a Berlin orthopaedics query to the Helios Berlin-Buch clinic page, clicks Termin buchen, and reaches the Helios appointment portal (patienten.helios-gesundheit.de) two steps into a six-step booking, with the orthopaedics department already selectable. [S3]

A plain reader fetches helios-gesundheit.de and finds clinic names but no department detail; the specialty filter renders only after JavaScript. A coding agent reaches more structure but still misses the orthopaedic match. The data is in the page; the JavaScript render hides it from text-only fetchers.

The booking step a browser reaches and a plain reader can't.

Score: AI Visibility 40.56 · AI Usability ~44 · Composite 4.6 / 10.

NephroCare runs the world's largest dialysis network, and the patient handoff is a generic contact form. A browser navigates from a Munich hemodialysis query to NephroCare München-Ost Neuperlach, confirms HD and Online-HDF as offered treatment types, finds the holiday-dialysis path, and lands on a contact form. [S3]

The discovery layer works for browsers; the close throws the context away. No patient registration, no appointment booking, no treatment-specific intake. The center finder reaches the right facility with the right therapy type, then drops the agent onto a free-text field. An agent that establishes the facility, the therapy, and the patient over four steps, then loses all three at a contact form, has accomplished nothing the patient can use.

Four steps to the right center and the right therapy, then a generic contact form.

Score: AI Visibility 45.0 · AI Usability ~44 · Composite 4.7 / 10.

Bayer built aspirin.de as a content showcase, then never put a buy button on the product an AI agent would actually recommend. A plain reader gets an Akamai 403 with a bot-registration page. A coding agent gets the same 403. The browser reaches aspirin.de, navigates to the Spannungskopfschmerzen (tension headache) content, and confirms Aspirin Forte 1000mg as the right product for tension headache with neck strain, in the legal Pflichttext at the footer only. [S3]

Forte has no product page on aspirin.de. No slider card. No /aspirin-forte URL. No Jetzt-kaufen button. The pharmacy handoff via px-reach.de exists, but only for the Aspirin Tablette 500mg variant on its own card. An agent that correctly identifies Forte from the content cannot purchase it, because the brand's commerce layer doesn't surface it.

Everyone in Germany knows Aspirin. No AI agent can buy Aspirin Forte 1000mg from Bayer.

This is an architecture gap, not a content gap. The content team got the indication right. The commerce team built the px-reach.de handoff. The bridge between them was never built for the variant the indication picks.

Score: AI Visibility 22.08 · AI Usability ~30 · Composite 3.1 / 10.

Place the five on the two axes that build the score, how well an agent finds the brand against how far it gets once there, and the inversion is hard to miss. The brand that built the most usable surface sits at the far left, the hardest to find.

The B2B brands beat the consumer brands by a wide margin. Qiagen and Sigma-Aldrich, the two pure B2B catalogues, post the two highest agent success scores in the wave. Helios and NephroCare, the patient-facing care networks, land in the middle: discovery works, the handoff is the weak part. Bayer, the most recognised consumer brand in the wave, trails it.

The split tracks how each surface was built, not how much was spent on it. The two B2B catalogues expose structured product data, catalog numbers, specifications, pricing, and let an agent add to cart without a login, because their users are professionals running procurement. AI agents run the same task pattern, so they sail through. The consumer and patient surfaces prioritise discovery, compliance, and brand presentation over agent-completable transactions, the exact opposite of what an agent matching an indication to a purchasable product needs.

One finding holds across the wave. Akamai and Cloudflare protection policies, often set by IT security teams with no view of agent visibility, are the single largest binary determinant of whether an agent reaches a brand at all. Sigma-Aldrich and aspirin.de return nothing on the HTTP path because a security team made that call; qiagen.com serves its full catalogue because its team didn't. The WAF decision is the new GEO decision, made by people who don't know they are making it.

Healthcare in Germany is intermediated by law. Prescription drugs need a prescription, OTC drugs go through pharmacies, hospital care needs referrals, diagnostic kits go to certified labs. So the scoring question is never did the agent buy the product. It is whether the agent reached the right handoff with the right product or service identified and accurate information surfaced. Four of the five brands in this wave do. The one that doesn't is the one with the strongest brand.

The most famous brand in the wave is the one an agent leaves empty-handed.

A surface in server-rendered HTML and a next step that survives the handoff are what separate an agent that reaches your product from one that defaults to a competitor, or hands your patient nothing. Qiagen put its catalogue in server-rendered HTML, and every agent that ran read it, no special agent layer added. Bayer put the right answer in a legal footer with no way to act on it.