| 1 | Since early October, ChatGPT can build working tools inside the chat, such as a calculator or a split-the-bill tool | s03 | MacRumors, 7 Oct 2026 | ต้องลองบนเครื่อง Yes: the exact Thai prompt "ช่วยทำเครื่องมือหารค่าอาหาร สำหรับห้าคนหน่อย" must produce a widget on a Thai account |
| 2 | The feature is rolling out to free and paid plans | s03 | MacRumors (Plus, Pro, Business first; Free and Go a day later) | ต้องลองบนเครื่อง Yes: test on a Free account in Thailand |
| 3 | You type items and prices into the tool, then ask for changes in the chat | s03 | MacRumors, hack in research-news.md item 1 | ต้องลองบนเครื่อง Yes: test a Thai change request ("ค่าเบียร์ ไม่ต้องหาร...") |
| 4 | Give AI rough notes, the recipient and the tone, and it drafts a polite work message; one more line refines it | s04, s05, s10 | research-audience.md item 5 (draft, then refine) and item 4 (Thai reply from bullet points, [V] Nation, 15 Jul 2026) | ไม่ต้อง No (general chatbot use) |
| 5 | Works in almost any chatbot | s04 | brief.md revision 2, s04 row ("works in any chatbot"); hedged as "เกือบทุกตัว" | ไม่ต้อง No |
| 6 | Microsoft's 2026 worker survey: 89% of Thai workers who use AI start from an AI draft, then refine it | s04 | Microsoft Source Asia, 4 Aug 2026: "89% of Thai workers use AI-generated outputs as a starting point to refine" (survey of AI users) | ไม่ต้อง No. Verified by the Lead 2026-10-26: wording matches; population is Thai AI users, said aloud as "คนทำงานไทยที่ใช้ AI" |
| 7 | AI can list subscriptions from pasted bills, flag price rises, total per month and per year, and draft a polite cancel message | s06, s07 | research-audience.md hack 2 (bill and subscription audit); brief.md hack 3 | ลองด้วยข้อมูลสมมติ Dry run with dummy data recommended; script tells listeners to re-check the total with a calculator |
| 8 | Mind, age 26, and all amounts (500, 1,000+, 199, 149, 149, 497 per month, 5,964 per year, "almost 6,000") | s06, s07 | Illustrative example, said aloud as an example. Not a factual claim. | ไม่ต้อง No |
| 9 | There is a long weekend from 5 to 7 December | s07 | research-audience.md, Seasonal hooks: Sat 5 to Mon 7 Dec 2569, 7 Dec substitute day, [V] Bank of Thailand notice. No URL given. | ไม่ต้อง No |
| 10 | On 20 October, ChatGPT began rolling out Locked Chats | s08 | TechCrunch, 20 Oct 2026: "starting to roll out" | ต้องลองบนเครื่อง Yes: confirm it is live in Thailand |
| 11 | Locked Chats works on free and paid plans | s08 | TechCrunch | ต้องลองบนเครื่อง Yes |
| 12 | Lock with a 4-digit PIN, or face or fingerprint scan on supported devices | s08 | TechCrunch, 20 Oct 2026: Face ID or Touch ID on supported devices; Android not stated | ต้องลองบนเครื่อง Yes: check what Android offers |
| 13 | A locked chat does not show in history and cannot be found by search | s08 | TechCrunch | ต้องลองบนเครื่อง Yes |
| 14 | Resetting the PIN deletes locked chats permanently | s08 | TechCrunch | ลองด้วยข้อมูลสมมติ Test only with a dummy chat |
| 15 | The lock is not end-to-end encrypted, so never paste full account or card numbers | s08 | TechCrunch | ไม่ต้อง No |
| 16 | If the option is missing, it may not be enabled for your account yet; update the app | s08 | OpenAI help article [S] (gradual rollout per account) | ต้องลองบนเครื่อง Yes: find the real path before building motion graphics |
| 17 | AI answers sound fluent and convincing but can sometimes be wrong; the trap is trusting them without checking the source | s09 | research-audience.md item 10 ("Check the AI's sources before trusting it", [V] YouGov) and insight 3 ("show a verify step") | ไม่ต้อง No |
| 18 | Fix: ask "where does this information come from?", open and check that source yourself, you make the final call | s09 | brief.md revision 2, s09 row; research-audience.md item 10 | ไม่ต้อง No |
| 19 | Hide account and card numbers before pasting money details | s06, s07, s08, s10 | brief.md hack 3 and 4; supported by claim 15 | ไม่ต้อง No |
| 20 | Money may leak every month and time leaks every day into hard work messages | s01 | brief.md revision 2 hook framing; a relatable question ("เคยไหมครับ"), not a statistic | ไม่ต้อง No |