We generate article engagement recommendations. TASKS 1....

Prompt

We generate article engagement recommendations. TASKS 1. Comment CTAs — follow these rules exactly (do not shorten or replace them): Generate exactly 3 audience engagement CTAs based on the news article. Return them in the cta array of the JSON object. Write every CTA entirely in the requested language; do not mix in foreign-language sentences or phrases. You may retain article-specific proper nouns, official names, brand names, standard abbreviations, acronyms, numbers, and necessary technical terms when appropriate. Do not copy foreign-language prompt text into the CTAs; translate the meaning into the requested language. Use natural, grammatically correct wording as a native speaker would write it; do not translate English sentence structures literally. Each CTA must be a complete, open-ended question. Each CTA must mention that readers can share their opinion in the comment box. Each CTA must refer naturally to the article topic, people, event, or situation. Do NOT write generic CTAs like "तुमचे काय मत आहे?" or "इस खबर पर आपकी क्या राय है?" unless connected to article-specific context. Do NOT repeat the same idea in different wording. Keep the tone thoughtful, respectful, and suitable for news readers. For crime, death, accident, legal, political, disaster, or sensitive news, avoid emojis and avoid sensational language. Do NOT make accusations beyond the article content. Use careful wording such as आरोप, प्रकरण, तपास, or समोर आलेली माहिती where appropriate. STRICT LANGUAGE RULES - All generated text must be written only in the language specified by article.languageName and article.languageCode. - Use the native writing script of that language. - Never use Chinese, Japanese, Korean, or an unrelated writing script. - Do not translate service IDs or topic IDs. - IDs must remain exactly as provided. - English brand names may remain in English only when they appear in the supplied article or candidate data. - Before returning the response, verify that every generated CTA and service title uses the requested language and writing script. SELECTION RULES - Comment CTAs always return exactly 3 strings in cta. - Do not return serviceId, atf, midArticle, or faq. - Do not return topicId, confidence, or heading_cta1. - Do not select a closest fallback. No match → omit that id. - Tasks are independent. A topic match must not change CTAs or astro; an astro Match must not change topicId. CONTENT RULES - Comment CTAs follow the comment CTA rules in task 1 only. Do not override them. - Service titles (atf, midArticle, faq) should not use clickbait or misleading claims. - Use respectful wording for crime, death, violence, accidents, disasters, or other sensitive articles. - Never include urls. - No markdown. One JSON object only. Return {"cta":["...","...","..."]}. Shape only — replace placeholders with real catalog ids. Do not invent ids. User message article.languageName: {{LANGUAGE_NAME}} article.languageCode: {{LANGUAGE_CODE}} article.category: {{CATEGORY}} article.title: {{TITLE}} article.body: {{BODY}} Apply the system tasks to this article. Task 1 comment CTAs always run. Do not recommend astrology. Do not match discussion topics. Do not default to the first list row for astro or topic. For Marathi, Kannada, Tamil, Malayalam and Telugu, the CTA rules are the same. Those languages add their own script rules, a "LANGUAGE STYLE" block and a "FINAL VALIDATION" checklist.