Scoring instructions for the AI model, translated from Polish (the model received them in Polish).
Batches of 25 families; input: the title and up to 700 characters of the abstract. In place of {areas} the model
received the list of topic areas below, and in place of {patents} a batch of patents. The "opis" field (a one-sentence
note on visibility) is not published: it is an untested hypothesis of the model.

You are scoring Google patents for SEO and for visibility in search and in AI answers.
For each patent, give:
- "rel": 3 if the patent is directly about Google Search or AI answers in search (ranking, indexing, crawling,
  quality assessment of pages and sources, selection of sources and citations, AI Overviews, AI Mode, snippets,
  local results and Maps, Shopping, Discover, News, search on YouTube);
  2 if it describes a mechanism that may affect the visibility of content or a brand (an assistant or a language
  model that uses search, grounding, RAG, query understanding, entities and the Knowledge Graph, fact-checking,
  content recommendations, ads in search results);
  1 if it is general machine learning or language processing with no connection to searching web content;
  0 if it is about images, hardware, speech, robotics, security or another area.
- "rodzina" (area): one of: {areas}.
- "opis" (note): only for rel 2 and 3: one sentence (up to 30 words) on what the mechanism does and what follows
  for the visibility of content. No em dashes. For rel 0 and 1, an empty string.
Answer with a JSON object {"grupy": [{"id", "rel", "rodzina", "opis"}, ...]}, with "id" as in the input.

{patents}

Topic areas (in the CSV under their English names): Information retrieval, intent and AI answers; Images, video and multimodal; Entities and semantics; Document structure; Crawling and indexing; Freshness and content changes; Links; Spam and abuse; E-commerce and products; Quality, facts and trust; Users, recommendations and personalization; Ads; Maps and local results; Outside search

Language model flag (a regular expression on the title and abstract, lowercase):
generative (model|ai|artificial|response|summar|answer)|large language model|language model|\bllm\b|chatbot|generative neural|foundation model|generative output|\bgemini\b|\bbard\b
