On 30 October 2026, OpenAI said it will honour a shared robots.txt signal that lets website operators state their preferences on AI training. The signal is designed to be read by participating providers, so one line in one file is meant to speak to several of them at once. I could not open the article itself, so I am working from the published summary and keeping the news part short.
What you will be told this means
The pitch will arrive within days. Someone will tell you that you can finally control what AI does with your website, and that you need to act now. A checklist will follow. Robots.txt rules for every AI crawler, an llms.txt file, a schema audit, and a package price to make you “AI-ready”.
So I ask the question I always ask. Is there a need for this? Yes, a small one. Operators have wanted a plain way to say yes or no to training, and a shared signal is better than a different rule for every crawler. That is a real improvement.
Then the second question. What actually changed? One more line in a file you already have, read by the providers who agreed to read it.
And what did not change? Whether an engine names your business when a buyer asks it who to use. Training preferences govern what models learn from. They say nothing about whether your clinic shows up when someone asks Google or ChatGPT which clinic in their area does same-day crowns. Those are two different questions, and the new signal only touches the first one.
Set it once and let the machine handle it
My position on llms.txt and the whole “AI-ready” list has not moved. Do it. Don’t worship it.
AI search engines are smarter than to put much weight on a simple text file. But if something costs nothing, harms nothing and can be automated, I let the machine handle it. The content engine behind this site writes the llms.txt entry for every post it publishes, and I never touch it. The training signal belongs in the same category. Decide which way you want it set, add the line, and move on.
What I will not do is treat that line as the strategy. A file that takes two minutes to set cannot be the thing that separates you from every other business that also took two minutes to set it.
The part that actually gets you named
Take a shop or a clinic that sells well on Google today. The owner sets the training preference, adds llms.txt, ticks every box on the checklist, and then asks ChatGPT which business to go to. Their name does not come up. Nothing on the checklist was ever going to change that.
The AI doesn’t pick the biggest brand. It picks the one it understands most clearly and can quote most precisely. That is something you can work on, and it lives in your pages, not in your robots.txt.
The pattern I keep seeing is that numbers survive and adjectives don’t. When I launched my own coffee store in Singapore in August 2026, Google’s AI Overview cited it within 48 hours, and everything the engines quoted from the new store was a specific fact. Prices, quantities, delivery details. Not one adjective survived. That is the work. Write down exactly what you are, who you serve and what makes you different, in terms a machine can quote back to a buyer.
Ranking and being named are also separate. If you rank well and still aren’t named, find out which one is broken before you fix anything. No checklist item will tell you that. Only checking the actual answers will.
What a business should do
- Decide your AI training preference, set it in robots.txt once, and add llms.txt if your platform can generate it. If an agency charges you more than an hour for this, you are paying for the wrong thing.
- Ask the engines your buyers use the five questions that matter to your business. Note whether you are named, and date the result, because answers change week to week.
- Put your time into pages that state specific, checkable facts about what you do. That is where the next month of effort belongs.
Source: OpenAI News: Making it easier for websites to express AI training preferences