Pollreal
पोल बनाएँरैंकिंगबोर्डसाइन इन
हिन्दी
  • English
  • 한국어
  • Español
  • Français
  • हिन्दी
  • Deutsch
  • 日本語
  • 中文
  • Русский

AI Notice

यह दस्तावेज़ केवल अंग्रेज़ी और कोरियाई में प्रकाशित किया जाता है। नीचे अंग्रेज़ी संस्करण दिया गया है; अंग्रेज़ी पाठ ही आधिकारिक है।

Version 1.0 — effective 4 November 2026

Pollreal uses generative artificial intelligence for one feature: the AI comparison card shown on each poll candidate. This notice is given in advance under Article 31 of the Korean Framework Act on Artificial Intelligence and explains what the AI does, what it does not do, and how to report a wrong card.

This document is published in English and Korean. The English text is the authoritative version; the Korean text is a translation provided for convenience. If they differ, the English text prevails.

1. What the AI comparison card is

When a poll is created, each candidate — a maker name and a product or service name — gets a card that summarises the maker’s official information: a product title, 2 to 10 short factual specifications, up to 3 notes, links to the pages used, and the official page when one was found. Each card is labelled “AI comparison card” and “AI summary — check the source”, shows the date it was checked, and says “Official source not confirmed” when no page on the maker’s own domain could be used.

2. How the card is made

  • Model: Claude (claude-haiku-5-5) by Anthropic, used through its API.
  • Inputs: only the maker name, the product name and the poll category. No personal data and nothing about who created the poll or who votes is sent.
  • Method: the model searches the web (at most 3 searches, with marketplaces, price-comparison sites, app stores, blogs, communities and video sites excluded) for the maker’s own page, reads it (at most 3 pages) and submits the card through a fixed form.
  • Checks before storage: our server verifies that every link in the card points to a page the model actually opened or saw in search results, that a source is marked official only when it is on the maker’s own domain, that marketplaces are never counted as official, and that lengths and counts stay within limits. A card that fails these checks is not stored.
  • Attempts and sharing: at most 2 attempts per product; at most 200 cards a day site-wide. One card is made per product and shared by every poll that includes that product. The card is written in English with a Korean translation made at the same time; pages in other languages show the English card.

3. Limits — check the source

  • The card may be incomplete, outdated or wrong, may miss the exact edition or plan, or may describe a different product with a similar name. Prices, plans and specifications change after the card is made.
  • The card is not a recommendation, not an endorsement by the Operator or the maker, and not part of the vote. Members vote for whatever reasons they have; the card only helps them look things up.
  • The maker of the AI model that writes the cards — currently Anthropic, the maker of Claude — and that maker’s competitors may appear as candidates in polls. Cards are factual summaries of official pages and never influence how votes are counted, ranked or displayed; a card about Claude is made and checked under exactly the same rules as any other card.
  • Always open the linked official page before relying on a figure.

4. What AI is not used for

  • Counting votes, computing shares or ranking candidates — these are plain arithmetic on the vote records.
  • Deciding whether a vote is accepted — those rules are fixed thresholds (one per member, one per device, network limits, blocked network lists).
  • Reviewing anomalies — the “Under review” notice is raised by a fixed rule (a one-sided burst of votes from brand-new members); a person decides what follows.
  • Moderating content or deciding about accounts — reports and sanctions are handled by a person.
  • Writing or translating members’ content — polls, posts and comments appear exactly as written.

Interface texts in languages other than English and Korean were translated with the help of AI and are awaiting review by native speakers. Legal documents are published in English and Korean only.

5. Data sent to the AI provider

The maker name, product name and category are sent to Anthropic’s API, which processes them under Anthropic’s API terms. No personal data of any member leaves the Service for this purpose. The resulting card, its sources, the model name and the cost of the call are stored by the Service.

6. Reporting a wrong card

Use the Report button on the poll (reason “Other”) and describe what is wrong — or write to hello@pollreal.com with the link. The Operator re-checks the card against the official page and regenerates, corrects or removes it. Makers can additionally claim their product (see the Community Rules) and tell us directly.

7. Contact

Dr.Seoul · hello@pollreal.com

सेवा की शर्तेंगोपनीयता नीतिसमुदाय के नियम

हर वोट के लिए: एक सत्यापित फ़ोन नंबर, वोट देते समय वोट करने वाले के अपने डिवाइस पर फ़िंगरप्रिंट या चेहरे की जाँच, कोई VPN या डेटा सेंटर नहीं, और एक डिवाइस पर एक वोट। AI तुलना कार्ड आधिकारिक पेजों का सारांश देते हैं — स्रोत ज़रूर देखें।

शर्तेंगोपनीयतासमुदाय के नियमAI सूचना