Copyright, other websites and AI training
This is an English translation for your convenience. The German version is the binding one. Read the German version
This page is for rights holders, publishers and anyone who wants to know how KIplatz handles other people's texts. It sets out what KIplatz reads, what it keeps and what it never keeps, what it learns from and how to ask for removal. We follow Austrian copyright law, in particular § 42h para. 6 UrhG (text and data mining) and § 76f UrhG (press publications), and the copyright chapter of the EU code of practice for AI models.
Have you found something at KIplatz that infringes copyright or other rights? Write to melden@kiplatz.at. We will remove it without delay.
How KIplatz handles other websites
- KIplatz does not search the internet in advance. There is no crawler that scours websites.
- When a question needs something current, KIplatz reads exactly the pages needed for that one question. If a page says “no to AI”, KIplatz leaves it out.
- What it reads stays only on the PC that asked and is deleted after a few hours, at the latest after 7 days. Other members get nothing of it.
- Stored for everyone are only Wikipedia (free licence), our own pages, pages with permission and KIplatz's answers in its own words, with a link to the source and the date. If an answer takes 10 words in a row from another website, it is not stored.
- Below every answer, “From where” shows where it comes from. If KIplatz answered without a source, for example with advice or a poem, it says so there too.
- If a rights holder contacts us, we block the content at once and reply within 48 hours. An email is enough; no formal warning letter is needed.
What KIplatz keeps for everyone
| What | For everyone? | Why it may be there |
|---|---|---|
| Wikipedia articles | yes | Licence CC BY-SA 4.0; the source is shown under every answer that comes from it. |
| A website the admin adds for everyone | yes, only with a reason | Own website, written permission of the operator, open licence or official work (§ 7 UrhG). The reason is stored with the address and date next to the page on the server. |
| Local knowledge and corrections by members | yes | Members' own texts, which they make available to us under the terms of use. |
| Answers by KIplatz | yes | In its own words, with title, link and date of the source. The quoted passage of another website is not stored with it. |
| Texts KIplatz reads when looking things up on the internet | no | They stay on the PC that asked, only for that person, and are deleted when they expire: weather after 6 hours, headlines after 3 hours, a search after 7 days. When a PC reads for another member, it keeps nothing. |
| Websites a member reads with “Website lesen” (read a website) | no | They stay private on that member's PC and only help with their own questions, much like “save page” in a browser. |
KIplatz is not a crawler that wanders through websites. It only ever reads a single page that was found or named for a question, never the pages behind it.
How KIplatz respects “no to AI”
Before KIplatz reads a page, it checks whether the operator has excluded programs using artificial intelligence. The page stays unread, and its snippet from the search is dropped as well, if any of the following applies:
- The robots.txt blocks the page for our reader “KIplatz” (until 7 October 2026 “CommunityAI-Wissen”; a block for that name still applies), for all programs (“*”) or even for just one of the known AI programs, such as GPTBot, ChatGPT-User, CCBot, Google-Extended, ClaudeBot, PerplexityBot, Applebot-Extended, Amazonbot or Bytespider.
- The robots.txt reports a server error or does not respond. The page then counts as blocked, as the robots.txt standard (RFC 9309) provides.
- The website carries a reservation under the W3C TDM Reservation Protocol, that is a file /.well-known/tdmrep.json, a header or a meta field “tdm-reservation”.
- The page carries “noai” or “noimageai” in the robots meta field or in the X-Robots-Tag header.
Our program only reads what is freely accessible without signing in and without paying. We do not get around paywalls, logins or copy protection. Our reader identifies itself as “KIplatz/1.0 (+https://kiplatz.at/urheberrecht/)”.
What KIplatz learns from
The KIplatz model is based on Ministral 3 by Mistral AI (Apache 2.0 licence). The community trains it further on its own computers. Training material consists of questions with answers about Wikipedia articles and sources with a clear licence, general knowledge questions from the community with their answers once several members have confirmed the answer, and checked corrections by members. Conversations and questions about a person are never training material, see How KIplatz learns.
For training, the rules above apply plus a stricter one: if a page contains any sentence about text and data mining or AI training, it does not go into the training list, even if the sentence allows it. Sites known for pirated works and sites with song lyrics never go in. Before every training run we check every reservation again. Write to melden@kiplatz.at and we will tell you whether anything of yours is in the training list, and take it out.
Nothing is taken over word for word
KIplatz writes its answers in its own words. If an answer takes over 10 or more words in a row exactly as on another website, KIplatz rewrites it before anyone sees it, or leaves the sentence out. Numbers, names and dates stay as they are in the source. No answer that contains ten or more words in a row exactly as in another source goes into the answer store or the training list. This way KIplatz keeps knowledge, not other people's texts. Corrections by members may be at most 500 characters long.
KIplatz may not be used to reproduce or distribute protected works, such as song lyrics, books or articles behind a paywall. This is also stated in the terms of use.
Notices from rights holders
If you find a work of yours in the shared knowledge, in a stored answer, in the training list or in an answer by KIplatz, an email to melden@kiplatz.at with the subject “Copyright” is enough. It helps to tell us:
- which work you mean and where it appears at KIplatz (address, question or line of the training list),
- where the original is published,
- who holds the rights and how we can reply.
If something is missing, we ask. A notice is not left aside just because it is incomplete. If you want KIplatz never to read your website, you can also set that yourself with a line in your robots.txt, for example “User-agent: KIplatz” and “Disallow: /”.
What we commit to after a notice
- At once: we block the reported content as a precaution in the shared knowledge, in the answer store and in the training list, even before we check it.
- Within 48 hours: the person who reported it gets a first reply from a human being.
- Within 7 days: we decide and write what we have done. If the notice is justified, the content is gone for good and does not come back, not even from the copies on members' PCs: they delete their copy at the next sync. On request we block the whole website for KIplatz.
- In the model: we train every new version of the model from the original, with the current training list. Removed content is therefore also gone from the model with the next version; we build new versions about once a month. In a serious case we switch all devices back at once to the previous version or to the unchanged Ministral 3.
We remove reported content quickly and without acknowledging any legal obligation. If you see your rights infringed, please contact us first: we handle a justified notice at no cost to you. A costly formal warning letter is not necessary.
Answers by KIplatz
Answers are written by an AI. They name their sources with a link so that you can read the original. Anyone who reuses an answer makes sure themselves not to infringe the rights of others, as with any other text.
Contact
For questions about copyright and training: melden@kiplatz.at. The provider is Philipp Rainer, Oberweis 16, 4664 Laakirchen, Austria (legal notice). We reply in German and English.
As of 8 October 2026.