Allianz
Can AI agents do business with Allianz?
3 tasks · 2 of 6 jobs measured · March to July 2026 · next wave Q3 2026
We gave AI agents three real jobs at Allianz: recommend an insurer, price a car insurance policy, and get ready to apply for a job. What follows is the plain record of what we asked for and what came back.
We ran each job with several kinds of agent, from the simplest ones that just read a page the way a script does, up to full browser agents that click through forms the way a person would. That range matters, because the simplest agents are also how AI systems read the web when they build their answers.
Named in half the answers. Recommended in none.
Whether an assistant brings Allianz up when someone asks where to buy car insurance, and whether it actually recommends it.
Someone asks their AI assistant where to get car insurance, and we wanted to see whether Allianz comes up at all.
We asked the three big AI assistants the questions a real buyer would ask, in three phrasings, twice each, which gave us eighteen answers to read through. The names that came back were often not insurers at all: in four of the eighteen the buyer was simply sent to a comparison portal (Check24, Verivox, Tarifcheck), and all four of those came from one provider's answers.
Answers naming each brand, of 18 audited answers · May 2026. Being named is not being recommended: no answer made Allianz the recommendation.
Across eighteen answers from three providers, Allianz was named in ten, twice inside a ranked list (positions 3 and 4), twice linked directly, and never as the recommendation. The visibility score of 19.79 out of 100 reflects that weighting: being named in passing counts far less than being the answer.
If you ask an assistant which car insurance to take, you get a confident, tidy answer, and Allianz may simply not be part of it. You never find out what was left out.
This is the expensive finding on the page, because it happens before anyone reaches allianz.de. The website mostly does its job, as the next two tasks show, but the shortlist is already formed by then.
Only a full browser agent ever reaches a price.
Whether an agent can get a real, personal premium out of the calculator without logging in.
Get a real annual premium for one specific driver: a VW Golf 2.0 TDI in Munich, claim-free class SF 15, 12,000 km a year, without logging in.
First a person filled in the Allianz calculator by hand for that driver, which produced 426.92 € a year and became our yardstick. Then we sent the agents at the same calculator and compared what they came back with.
±1.5 % All three runs landed within one and a half percent of the price a person got.
It clicked through all eight steps of the calculator three times and came back with 420.53 €, 432.49 € and 425.13 €, each within a percent and a half of what the person got. It never had to log in, and it could have gone on to apply.
It asked for the calculator page once and the security filter turned it away, even though the rejection page it got back still contained tariff text. The price is in the building, but the door is locked.
It reached the start of the calculator, but there is no shortcut to a personal price: you only get a quote by working through the whole eight-step form, so it stopped there.
If your agent can drive a browser, Allianz works today and the price you get is one you can trust. With anything simpler, your agent comes back empty-handed.
The calculator is genuinely usable by an agent, which most of the DAX cannot claim, but only by the most expensive kind. Every cheaper agent gets nothing, and the same filter that blocks them also blocks the crawlers that feed the assistants in the first task.
Every tested agent could read the job. The test stopped at sign-in.
How far an agent gets from finding a posting to actually beginning an application.
Find an open Senior Backend Engineer role on the Allianz career site, read it properly, and get as far as starting an application.
We picked a real posting in Munich and let the agents work through it. Reading it was no trouble, but the apply button leads to a sign-in screen that wants a candidate account before a single form field appears, so we stopped there and never logged in.
One plain request was enough to get the full posting, with nothing hidden behind scripts. The same kind of agent that was turned away at the insurance calculator walks straight in here.
The posting comes with clean, machine-readable job data, complete with title, location and dates, so it could be extracted and compared without guesswork.
For every agent we tested, the apply link leads to a SuccessFactors sign-in screen that wants a candidate account before a single form field appears. A deliberate choice rather than a fault, but it means no agent can start an application on a candidate's behalf.
Your agent can find, read and rank Allianz jobs as well as any recruiter, but the moment it should start the application for you, the login hands the work back.
The career pages are among the most readable we measured, and the login gate is a fair decision to make. It is worth making it knowingly, though, because agents that pre-fill applications are quickly becoming a normal way to apply.
The calculator blocks simple fetches. Whether that contributes to Allianz's weak recommendation presence remains unmeasured.
The same security filter that turns away the simplest agent at the car insurance calculator also turns away the crawlers that feed the AI assistants, and that is exactly where Allianz is missing from the answer. The two findings sit next to each other in our data. We have not measured a causal link between them, so treat this as a question the next wave should answer, not a proven result.
How well Allianz serves agents on the jobs we defined.
One number per job, with the pieces it is made of, and an honest blank where we have not measured yet.
| Job | Score |
|---|---|
| Buying Tasks 1 and 2 Allianz is named in roughly half the AI answers but never recommended, and only the most capable agents can reach a price. | 3.2 / 10 |
| Applying Task 3 The job pages are easy for agents to read, but every tested agent stopped at the sign-in. We have not measured how visible Allianz jobs are in AI answers yet. | no score yet |
| Service, procurement, investors, press Not measured yet. | · |
Brands are only ranked against each other on what every brand was measured on, which today is the buying journey.
Allianz works for capable agents, but barely appears where buyers decide.
On the two jobs we could run end to end, the website holds up: a browser agent gets a car insurance price a person would trust, and the career pages are among the easiest to read in the whole index. That is more than most of the DAX can say.
The problem sits earlier and lower. Anything less than a full browser agent is turned away at the calculator, and Allianz is named in half the answers without ever being the recommendation. That is what pulls the buying score down to 3.2 out of 10.
What would change the next result
Three things follow straight from the three jobs.
- Answer the buying question
- Publish content that answers the questions people actually ask about car insurance, so assistants have something of yours to quote before the comparison portals fill the gap.
- Let agents reach a price
- Offer a way to get a quote that does not require clicking through eight steps, so cheaper agents can get there too.
- Decide on the career login
- Keep the login or drop it, but make it a decision rather than an accident.
Your brand, the same test.
The free snapshot runs the same agents against your brand. One surface, one task, in 48 hours. No card, no call required.
Annex · sources and open questions
- Task 1 · visibility
- Q2 2026 answer audit: visibility score 19.79, 18 answers across OpenAI, Anthropic and Perplexity production models, three phrasings, two runs each. Answer traces held for citation. Per-brand mention counts recounted from the stored answer texts on 12 Aug 2026 (mentions-count-2026-08.yaml).
- Task 2 · pricing
- Human ground-truth run on 2026-03-25 (426.92 €, Komfort tariff) plus agent runs run-XI01-ALV-01/02/03. The simple reader received an HTTP 403 from the calculator page.
- Task 3 · applying
- Talent wave 2026-07-02, requisition 90293 in Munich, ground-truth copy of the posting saved before measurement. Valid JobPosting structured data present; apply flow ends at a SuccessFactors sign-in.
We test kinds of agent, not brand-name products, so anything below the assistant level is built to behave like its class rather than to be a specific tool. Task 1 is the exception: there we queried the production models directly.
The browser agent did not run in the July career wave, and we never logged in anywhere, so anything behind a sign-in is untested.
- 01 The usability half of the buying score still uses an older measurement run; a clean rerun is pending.
- 02 The link between the blocked calculator and the missing AI mentions is an inference, not a measured result.
- 03 The verdict word (low / mixed / high) follows a provisional rule until the scoring note is published.
Allianz can respond and we publish the reply in full next to the findings. Every brand gets 14 days before a major update goes live.
