The test.
Wave 8 put three DAX aerospace and defence primes under one question each: when a buyer's agent researches the locked variant, does it reach the brand, the right product, and a quote path it could hand to a human? One frozen task per brand, public, locked before measurement. 18 datapoints per brand, across OpenAI, Perplexity, and Anthropic, in German [S2]. The task slate is set by the public Task Selection rule [S4].
- Airbus — the H145: twin-engine medium helicopter, eight passenger seats in the standard layout, MTOW 3.8 tonnes, final assembly in Donauwörth. Close: an RFQ via Airbus Helicopters regional sales.
- MTU Aero Engines — PW1100G-JM engine MRO for the A320neo at MTU Maintenance Hannover. Close: an RFQ via a B2B sales contact.
- Rheinmetall — the Lynx KF41: tracked infantry fighting vehicle, around 44 tonnes combat weight, Lance 2.0 turret. Close: an RFI via B2G defence sales.
No brand in this wave runs a checkout. There is no cart for a helicopter, no order button for an engine overhaul, no basket for a fighting vehicle. Every close path is a direct-sales RFQ [S1]. That makes Wave 8 a different kind of test: not "can the agent buy", but "does the agent arrive at the right variant with a callable next step".
One caveat, stated once for all three. This release publishes AI Visibility, the measured discovery score. The fleet pass per agent class, AI Usability, and the composite land in Wave 8b. Until then, every score line below carries that scope. Confidence C [S2].
In aerospace there is nothing to buy. There is only the path to the quote.
Airbus: one letter in the variant decides the whole answer.
The H145 is a twin-engine medium helicopter for EMS, police, and para-public operators. The single-engine H125 sits near it in the Airbus light range, a different product family with a similar name. For an emergency-medical fleet, that difference is the whole answer: EMS missions are flown twin-engine, because flying on after an engine failure is a regulatory requirement for the mission profile, not a preference [S1].
So the Wave 8 task locks the variant: twin-engine, eight passenger seats, MTOW 3.8 tonnes, final assembly in Donauwörth. An agent that recommends an H125 to a fleet replacing ageing twin-engine machines has not made a near miss. It has answered a different question, confidently.
Discovery passes. Airbus surfaces in the AI shortlist on the unbranded twin-engine-medium probe, AI Visibility 43.9 [S2]. The open question is the path after the shortlist: airbus.com routes RFQs through a regional-sales selector that asks the visitor to identify a country before a contact appears. Whether an agent traverses that selector or strands on a generic global contact form is exactly what the Wave 8b fleet pass tests [S1].
Recommending the single-engine sibling to an EMS fleet is not a near miss. It is a confident wrong answer.
Score: AI Visibility 43.9 · Usability pending (Wave 8b fleet pass) [S2]. See the Airbus BrandScore.
MTU: the most visible prime in the wave sells a service, not a thing.
MTU Aero Engines posts the highest AI Visibility of the three, 52.5 [S2]. It is also the one brand in the wave whose product is not an object at all. The task is an MRO contract: maintenance, repair, and overhaul for the PW1100G-JM geared turbofan that powers the A320neo, run at MTU Maintenance Hannover [S1].
The disambiguation burden is the story. MTU holds stakes in several commercial engine programs at once; the PW1100G-JM stake is 18 percent [S1]. A fleet manager asking about A320neo MRO needs PW1100G-JM data and a Maintenance Hannover contact. V2500 content answers the previous-generation A320ceo. Widebody-program content answers a different airframe entirely. Same brand, same site, wrong engine.
Whether an agent lands on the program-specific surface or strands on a generic maintenance contact form is the Wave 8b question. Discovery, the part Wave 8 measured, is the part MTU has already won [S2].
The brand is the easiest of the three to find. The right engine page is the test.
Score: AI Visibility 52.5 · Usability pending (Wave 8b fleet pass) [S2]. See the MTU Aero Engines BrandScore.
Rheinmetall: one site, two audiences, and the agent has to pick right.
rheinmetall.com serves two readers at once: the investor reading corporate pages, and the defence-procurement officer reading platform pages. An audience selector routes between them. A NATO procurement officer researching a tracked Marder-replacement needs the Defence side, and on it the right platform [S1].
The variant test is sharp here too. The Lynx KF41 is the production vehicle: around 44 tonnes combat weight, the Lance 2.0 turret, a 30 or 35 millimetre gun depending on configuration. The KF31 is the lighter demonstrator generation. The Boxer is a wheeled vehicle on a separate procurement track. An agent that returns Boxer where Lynx KF41 was asked has changed the procurement category, not missed a detail [S1].
Discovery holds. Rheinmetall surfaces at AI Visibility 49.4 on the unbranded tracked-IFV probe [S2]. The Wave 8b questions are the structural ones: does an agent traverse the audience selector, reach the KF41 platform page, and surface a defence-sales contact it could hand to a human, rather than a region-by-region matrix built for trade-fair visitors?
The agent's first decision on rheinmetall.com is not which vehicle. It is which audience.
Score: AI Visibility 49.4 · Usability pending (Wave 8b fleet pass) [S2]. See the Rheinmetall BrandScore.
Siemens: the neighbour with a checkout, and a login in front of it.
The aerospace primes have no checkout to test. Siemens, the Wave 2 industrials anchor, has one, which is what makes it the comparison case [S3].
The SIMATIC S7-1500 controller surface let a browser-class agent run the selection tool end to end and click through to order in all three runs, with full spec fidelity, before hitting the Siemens-ID login wall. AI Visibility 44.0, Composite 5.6, the strongest industrial result in the index so far [S3].
That marks the fourth surface mode in the heavy-industrial picture: not "no checkout at all", which is aerospace and defence by construction. Not "the close captured by retail", which is the chemicals story. Not "the spec behind a quote form", which is the distributor story. But "the close exists, and a login closes it."
Anchor: AI Visibility 44.0 · Composite 5.6 / 10 (Wave 2, scored) [S3].
Found, not yet quotable: what the three tell us.
Three primes, three locked variants, one shared shape. Every brand surfaces. No brand sells. The close lives in a human sales channel, and the agent's job is to arrive there with the right variant in hand.
Airbus. The variant decides the mission: twin or single is the whole answer, and the RFQ path runs through a country selector built for humans.
MTU. The most findable brand in the wave, with the heaviest disambiguation burden: the right engine program, on a site that serves many.
Rheinmetall. Two audiences on one domain, three platforms that a procurement officer must not confuse, and a B2G contact path at the end.
What the sector shares with the rest of the index is the direction of the risk. The agent will not fail loudly here. It will shortlist confidently, from whatever the surface hands back. A wrong variant, a generic contact form, an investor page where a platform page was needed: each produces an answer that looks complete and is not [S1].
A quote-ready sector fails quietly: the shortlist looks complete, and the path to the quote is not on it.
So the work is on the surface, and the stakes are variant-deep. A locked variant an agent can verify and an RFQ contact it can hand to a human, on the prime's own page, separate the brand that enters the panel from the one that stays a name on a shortlist.
