I run a visa agency in Bali. The other half of my week goes into Apify Actors, and one of them scrapes a page type that only exists because a law says it must: the Impressum. Every business website in Germany, Austria and Switzerland has to publish one, and it has to carry things no marketing page ever would — the legal company name, the people who run it (Geschäftsführer, Inhaber, Vorstand), the register court and HRB / FN / UID number, the VAT ID, a real postal address and a real email.
For B2B lead work that is unusually good data. A Google Maps export gives you info@ and a phone number. An Impressum gives you the name of the person who signs things, plus a register number you can verify. One domain in, one row out.
That is the theory. Today I measured the practice, twice, and the gap between the two runs is the whole article.
Run one: the boring companies
The Actor's own prefill — five DACH companies of the size my Actor was built for. A muesli brand, a clothing manufacturer, an accounting SaaS, a hosting company, an Austrian chocolate maker:
{ "websites": ["www.mymuesli.com", "www.trigema.de", "www.sevdesk.de", "www.hetzner.com", "www.zotter.at"], "maxPagesPerSite": 4 } 5 of 5 returned a full company row in 4.1 seconds. No proxy, no browser, no retries. One of them:
{ "type": "company", "website": "https://www.hetzner.com", "impressumUrl": "https://www.hetzner.com/de/legal/legal-notice/", "companyName": "Hetzner Online GmbH", "legalForm": "GmbH", "managingDirectors": ["Martin Hetzner", "Stephan Konvickova", "Günther Müller"], "registerCourt": "Amtsgericht Ansbach", "registerType": "HRB", "registerNumber": "6089", "vatId": "DE812871812", "address": "Industriestr. 25, 91710 Gunzenhausen, Germany", "phones": ["+49 (0)9831 505-0"], "fax": "+49 (0)9831 505-3", "emails": ["info@hetzner.com"], "language": "de" } Note where that page lives: /de/legal/legal-notice/. Not /impressum. mymuesli's is at /policies/legal-notice (a Shopify path), Trigema's at /kundenservice/rechtliches/impressum/. Only two of the five sat on a path you could have guessed. The crawl is a cheap three-step — home page, then the best "Impressum / Imprint / Legal notice / Rechtliche Hinweise" link in the menu, then the usual paths as a fallback. The link in the footer is what actually does the work.
Run two: ten names you know
Same Actor, same settings, a list of brands instead: dm, About You, FlixBus, N26, Personio, Celonis, ÖBB, Red Bull, Migros, Swisscom.
Ten websites, 7.6 seconds, two company rows, one of them billable.
| website | result |
|---|---|
| oebb.at | full company row — FN 248742y, Handelsgericht Wien |
| n26.com | imprint found, no company name → free row with a note |
| dm.de | no Impressum found (4 pages checked) |
| flixbus.de | no Impressum found (4 pages) |
| celonis.com | no Impressum found (4 pages) |
| redbull.com | no Impressum found (4 pages) |
| migros.ch | no Impressum found (4 pages) |
| swisscom.ch | no Impressum found (4 pages) |
| about-you.com | could not be opened — connection closed, residential retry failed too |
| personio.de | could not be opened — HTTP 429 even from a residential address |
My first guess was crawl depth: four pages per site is not much for an enterprise site with a mega-menu. So I re-ran the six "no Impressum found" domains with maxPagesPerSite: 10.
46 pages fetched. Zero rows recovered. Every single one came back the same way, and four of the six stopped at seven or eight pages because the Actor ran out of plausible links before it ran out of budget. It is not a depth problem.
What is actually happening on those six sites
I went at them by hand afterwards, with a plain curl and a desktop user agent, to see what the fetcher had been looking at.
1. The page exists and contains nothing. dm.de/ and dm.de/impressum both return HTTP 200 and the same 11,392 bytes — an app shell. Neither body contains the string HRB, Amtsgericht, Handelsregister or Geschäftsführ. The legal text arrives later, from JavaScript. Same story in Switzerland: migros.ch/de/impressum.html is 32 KB of HTML with none of those words in it, swisscom.ch/de/about/impressum.html 89 KB with none either.
2. The geo redirect deletes the legal page. My input was flixbus.de. Look at what the row says it read: https://www.flixbus.com. From a European IP, flixbus.de/impressum is 89 KB of HTML with HRB, Amtsgericht and Handelsregister all present. From the datacenter the run used, .de redirected to the global .com site — and flixbus.com/impressum is a 404. The German legal page simply does not exist on the domain the scraper was moved to.
There is a hop out of this, and it also fails. flixbus.com publishes 49 hreflang alternates, and my Actor is supposed to follow the German one when it lands on an English page. Here is the German one:
<link rel="alternate" hreflang="de" href="https://www.flixbus.de" /> It points back at the host we were just redirected away from. The hop is a no-op, the URL is already in the seen-set, and the crawl ends.
3. The datacenter 403. redbull.com answered my hand request with 403 and 369 bytes. The run's fetcher got further — it reported pages checked, not a block — but it was reading whatever a bot is served, and that did not include an imprint.
4. The one that was honest about it. personio.de returned HTTP 429 to the direct request and to the residential retry. The Actor wrote the reason into the row and charged nothing for it, which is the only correct behaviour here: a site that rate-limits you has not sold you data.
Rows 3 and 9 in the table are worth separating out from the "no Impressum" ones for a practical reason. "Could not open" is retryable later from a different address. "No Impressum found on a 200-byte-perfect page" is not — that site needs a browser, and no number of retries will change it.
The two rows that came back, including the one I got wrong
ÖBB is a clean row, except for one field:
{ "companyName": "ÖBB-Personenverkehr Aktiengesellschaft", "legalForm": "AG", "managingDirectors": ["ÖBB-Personenverkehr Aktiengesellschaft"], "registerCourt": "Handelsgericht Wien", "registerType": "FN", "registerNumber": "248742y", "vatId": "ATU58044146", "address": "Am Hauptbahnhof 2, 1100 Wien, Austria", "emails": ["impressum@pv.oebb.at"], "supervisoryAuthority": "Oberste Eisenbahnbehörde (Abt. II/C) im Bundesministerium für Verkehr, Mobilität und Infrastruktur, 1030 Wien, Radetzkystraße 2." } managingDirectors is the company's own name. Austrian imprints often use "vertreten durch" to name a corporate representative rather than a person, and my name-shaped-string check happily accepted it. If you are building a mail merge on this field, test it against companyName and drop the matches — I am fixing it upstream, but your data may predate the fix.
N26 is the more interesting miss. The imprint was found at /en-eu/imprint, and the row carries HRB 271697 B, Amtsgericht Charlottenburg, DE305957096 and six emails — four of which are genuinely in the page, including the insurance ombudsman and the Berlin chamber of commerce. What it does not carry is companyName. The page says N26 Bank SE, but it says it inside an English sentence ("N26 Bank SE is registered in…"), not on a line of its own above an address, which is what the parser looks for.
The useful part is what happened next: the billing rule for this Actor is company name plus at least one of directors, register number, VAT ID or address, so the row shipped with a note and was not charged. The run summary says it out loud:
Run summary: 2 row(s) delivered, 8 free, 8 failed; charged company ×1 = $0.0035. One cent of three for ten of the best-known companies in the region. I would rather publish that number than a 90% success rate measured on a list I picked.
So who is this for
The honest answer from these two runs: the Mittelstand, not the brands. Shops, agencies, law firms, freelancers, local manufacturers, associations — companies whose website is a CMS with a footer link, where the Impressum is a static page because nobody ever had a reason to make it anything else. My earlier measurement on 130 real DACH websites of that kind read an Impressum on 110, and on those, company name landed on 99%, address 97%, legal form 92%, VAT ID 85%, register number 84%, directors 83%.
If your input list is a Google Maps export of Berlin agencies, this works. If it is the DAX 40, go to the Handelsregister instead.
Calling it from a script is one POST:
curl -X POST "https://api.apify.com/v2/acts/lergassy~impressum-scraper/run-sync-get-dataset-items?token=$APIFY_TOKEN" \ -H 'Content-Type: application/json' \ -d '{"websites":["muster-gmbh.de","www.zotter.at"],"maxPagesPerSite":4}' from apify_client import ApifyClient client = ApifyClient("YOUR_TOKEN") run = client.actor("lergassy/impressum-scraper").call(run_input={ "websites": ["muster-gmbh.de", "www.zotter.at"], "validateVatId": True, # paid add-on: EU VIES check }) rows = [r for r in client.dataset(run["defaultDatasetId"]).iterate_items() if r.get("type") == "company"] Two things in that snippet matter. Filter on type == "company" — the free error rows and the final run-summary notice are in the same dataset by design, because "we could not open this site" is information you want, not noise to hide. And validateVatId is a separate paid event that checks each VAT ID against the official EU VIES service; Germany does not return the registered name through VIES, so it confirms validity and not much else, and Swiss CHE- numbers are not in VIES at all.
Where it is the wrong tool: you need every company in a city (start with a directory or Maps scraper, then feed the domains here), you need financial statements (Bundesanzeiger), or your list is not DACH — there is no Impressum to read outside the German-speaking web, and this Actor will correctly tell you so, for free, ten times in a row.
The Actor: Impressum Scraper. The next thing I am fixing is the corporate managingDirectors string, and then a browser fallback for the app-shell sites — because dm.de serving the same 11,392 bytes for its home page and its legally mandated imprint is a problem no amount of crawling solves.
