On Friday Google and OpenAI said they will honour one shared line in robots.txt that tells AI providers not to train on a site. For a brand, the useful question is not yes or no. It is which folder stays open.
One shared robots.txt signal for AI training
Google plans to support a shared AI training control in robots.txt, built on IETF work. A site that opts out with the shared token, without naming Google-Extended, will be treated as blocking Google-Extended; sites that allow Google-Extended stay eligible for clicks via grounding in Gemini Apps. Target: 2027 (Google Search Central, 30 Oct). OpenAI said the same day it will honour the signal and implement it “in the coming months” (OpenAI, 30 Oct). The draft behind it uses the token __usage_ai-training; its example blocks a whole site except /press-kit/. It has no formal IETF standing yet (IETF draft, 30 Oct).
My read: That press-kit example is the right shape for an FMCG site. Approved packshots and logos are what you want every model to get right. On the Festival Toffee 3D ad my job included the product sheet, the reference every shot was checked against. A model that has never seen your real pack will guess it. Campaign films, unreleased work and talent images go behind the line. A shared opt-out is a preference, not a lock. It works only with providers that honour it, it starts next year at the earliest for Google, and it does nothing about what was scraped before. And a blanket block can cost you Gemini citations.
Pay once, rework forever: AI and licensed art
Pitjantjatjara artist Sally Scales, who chairs the National Gallery of Australia’s First Nations advisory group, says AI is turning First Nations art into “Temu slop”; Temu, Shein and Amazon list many Aboriginal-style works, mostly unattributed. UNSW copyright scholar Kathy Bowrey describes organisations that pay once for artwork or images of Aboriginal people, then use AI to “update” them without permission (The Guardian, 1 Nov).
My read: Bowrey’s example is a brand problem. A team licenses an illustration or a face for one campaign; a year later someone runs it through an image tool for a “fresh version”. That is a new derivative work, and most old licences say nothing about it. I localise one hero film into several language versions; each one is a new file the contract must cover. My rule: if the source holds licensed art, a real person or a community’s pattern, the licence names AI derivatives, or the version does not get made.
iQIYI shoots the actor, builds the set afterwards
iQIYI opened what it calls China’s first AI Film Studio in Shanghai: a roughly 110 m² multi-camera stage whose footage syncs to NadouPro, its platform with nearly 30 AI models including Seedance 2.5 and Kling 4.0. Sets, lighting, costumes and props can be changed after filming; the actors’ performances stay. The first feature, Cop in Trouble, began filming on 28 October. No cost figures are given (iQIYI via PR Newswire, 1 Nov).
My read: Real performance, generated world. That is the split I would brief for any brand film with a presenter or a celebrity. It keeps consent simple too: the person signs for a performance they actually gave. My own films are fully AI-made. On the Mooo back-to-school film my role included the characters and the location sheets: a world only holds from shot to shot when someone draws it first. A studio does not remove that step. It moves it to after the shoot.
What I’d do today
Open your site’s robots.txt and list your folders in two columns: what models should learn (packs, logos, press kit) and what they should not (films, talent, unreleased work). The line takes a minute; the list is the decision. If you need films built on clean, cleared assets, I direct them shot by shot.
Sources
- Google Search Central — Supporting a shared AI training control for websites, 30 Oct 2026 — https://developers.google.com/search/blog/2026/10/ai-training-control
- OpenAI — Making it easier for websites to express AI training preferences, 30 Oct 2026 — https://openai.com/index/helping-websites-express-ai-training-preferences/
- IETF — Machine-Readable Training Preference Signal (draft-kelley-aipref-training-preference-signal-00), 30 Oct 2026 — https://datatracker.ietf.org/doc/draft-kelley-aipref-training-preference-signal/
- The Guardian — AI is stealing Indigenous Australian art and churning out ‘Temu slop’, artist says, 1 Nov 2026 — https://www.theguardian.com/artanddesign/2026/nov/02/ai-stealing-indigenous-australian-art-temu-slop-artist-says-ntwnfb
- iQIYI (PR Newswire, via Yahoo Finance) — iQIYI launches China’s first AI Film Studio, 1 Nov 2026 — https://finance.yahoo.com/technology/ai/articles/iqiyi-launches-chinas-first-ai-013900330.html