Thailand’s food-advertising rules, in force since 28 July 2026, define “advertising” as any means that lets the public see or hear a claim — not just TV or print. That reaches an AI Agent’s chat reply and a staff member’s DM just as easily as a commercial, and the Food Act’s two advertising offences apply either way, whether the claim is a lie or just unapproved.
What changed on 28 July 2026
The Thai FDA’s Notification on Food Advertising Criteria B.E. 2569 was published in the Royal Gazette (Volume 143, Special Issue 182 Ngor) on 27 July 2026 and took effect the next day, replacing the 2021 version (Royal Gazette notice; reported in The Nation). It bans advertising for infant formula and follow-up formula outright, and sets specific restrictions for kratom- and cannabis-containing food ads — proof that the rule reaches far past TV spots into whatever channel a seller actually uses to describe a product.
Why a chat reply carries the same two risks as a TV ad
Two separate offences sit under the Food Act B.E. 2522, and a DM trips either one. Section 40 bans advertising a food’s benefits, quality, or efficacy if the claim is false or deceptive — penalty up to 3 years in prison or a ฿30,000 fine under Section 70. Section 41 goes further: advertising a food’s benefits, quality, or efficacy for commercial purposes “through broadcast, print, or any other means” requires submitting that claim to the FDA for approval before it runs, true or not — penalty up to a ฿5,000 fine under Section 71 if you skip that step (legardy.com: Thai food and drug law). “Any other means” is exactly where a chat reply, a DM, or an AI Agent’s canned answer about what a supplement does sit — the same spot the 2026 notification just made explicit.
Set up in 1CS
The fix isn’t rewriting every reply by hand — it’s a rule the AI Agent can’t skip. Open AI > AI Agent (AI เจ้าหน้าที่) and click Edit Brain (แก้ไขสมอง), then open the Persona (บุคลิก) settings. Alongside the example 1CS already gives for a different industry — “ไม่ใช่หมอ ห้ามวินิจฉัย” (not a doctor, must not diagnose) — add your own line in the same Additional rules (กฎเพิ่มเติม) field: something like “ห้ามอ้างสรรพคุณการรักษาหรือลดน้ำหนัก เว้นแต่ข้อความได้รับอนุญาตโฆษณาจาก อย. แล้ว” (no healing or weight-loss claims unless the FDA has already approved that exact wording). That rule applies across every persona, not just one. Run a few real questions through Chat Preview (ตัวอย่างการสนทนา) before the AI Agent goes live, the same way you’d check any other change to its Brain; the Thai help center walks the same screen step by step at Edit Brain.
A safe answer and a checklist to copy
Give the team one answer to paste when a customer asks what a product does, rather than improvising:
ขอบคุณที่สอบถามนะคะ สินค้านี้ขึ้นทะเบียนกับ อย. แล้ว แต่ทางร้านยังไม่ได้รับอนุญาตให้โฆษณาผลลัพธ์แบบที่คุณถามถึง จึงขอแนะนำให้ใช้ตามวิธีบนฉลากสินค้า และสอบถามแพทย์หรือเภสัชกรเพิ่มเติมได้เลยค่ะ
Words that research on Thai enforcement flags as exaggerated — and therefore deceptive under Section 40 — belong on a list both your team and your AI Agent’s brain should avoid: “หายขาด,” “รักษา,” “การันตีผล,” “มีปาฏิหาริย์,” “ดีที่สุด,” “ลดได้จริงใน 7 วัน,” “ขาวใน 3 วัน” (lexbangkok.com: Thailand supplement advertising rules). None of this blocks selling the product — it blocks the one sentence that turns a sale into a Food Act case. Sellers already keeping prices visible in chat to dodge a different fine, or filing for a permit before a live-sale giveaway, are used to the same pattern: the paperwork sits before the message goes out, not after a customer forwards it to the FDA.
