Google Backs One Shared AI Training Control

Google announced on 30 October 2026 that it will support a shared AI training control in robots.txt, a single setting that lets a website express its AI training preferences across multiple AI providers at once. The plan is set out in Google's post on supporting a shared AI training control for websites, and it is based on work from the AIPREF working group at the Internet Engineering Task Force (IETF).

Here is what that means if you run growth at a SaaS company. You now have one decision to make with your legal and content owners, which is whether your site opts in or out of AI training, and one mistake to avoid, which is letting that decision bleed into the crawl rules that let answer engines fetch and cite your pages. Those are separate things. Conflate them and you can quietly remove yourself from the surfaces where your buyers are now forming a shortlist.

What Google actually said about the shared AI training control, and what it did not say

Google said it is announcing plans to support a shared AI training control in robots.txt. It said this will help websites express their AI training preferences across multiple AI providers in a single step. It said the work is based on the AIPREF working group at the IETF.

Google also framed it as building on a long history of giving website owners clear controls, including Google-Extended, the AI training control it has offered since 2023. That is the full substance of the announcement.

Google has not said when the control ships, what the exact syntax is, or which other providers will honour it. We would not write a rollout date into your roadmap on the basis of this post.

Why "training" and "answering" are not the same control

This is the part most teams get wrong, and it is worth being precise about. Training is a model learning from your content. Answering is a system fetching a page, often in real time, to produce a response and sometimes a citation.

A directive about AI training preferences is not, on its face, a directive about retrieval or citation. Google's post is about training preferences. It does not say that this control governs whether your pages appear in AI Overviews, Gemini responses or anywhere else.

So before anyone edits a file, map what each existing line in your robots.txt is doing. The table below is the distinction we walk clients through, and it is the one your legal team will not make for you.

How a SaaS team should separate the two decisions in robots.txt, as at Google's announcement on 30 October 2026

Decision	What it governs	Who owns it	Commercial risk if you get it wrong
AI training preference (the new shared AI training control)	Whether AI providers may use your content for model training, expressed once across multiple providers	Legal or data policy, with content and brand	You either give away IP you meant to protect, or you block something that was never affecting your visibility anyway
Crawl and fetch rules for answer engines	Whether a system can reach your pages to generate and cite an answer	Technical SEO, inside the growth team	You disappear from AI-assisted research without anyone noticing for a quarter
Classic Googlebot crawling	Standard indexing for organic search	Technical SEO	Organic traffic and pipeline fall off a cliff

Why the shared AI training control changes the shape of the conversation

Until now, expressing a preference meant handling providers one at a time, each with its own token. A shared AI training control compresses that into a single step, which is exactly why it will land on your desk rather than staying buried in a technical backlog. One line, one meeting, one position that applies broadly.

That convenience cuts both ways. A control that applies across multiple AI providers in one step is also a control that can be set carelessly in one step, by someone who was asked to "block the AI bots" and had no view on where demand now comes from.

For B2B SaaS, the discovery layer has already moved. Buyers are running category research through ChatGPT, Gemini, Perplexity and Copilot before they ever hit a comparison page. If you want the longer argument on that, our piece on how brand visibility in AI search builds in steps rather than flipping on covers how mentions accumulate.

What a SaaS growth team should do this quarter

Three things, in order. None of them require you to change a file this week.

• Agree a position, in writing. Get legal, content and technical SEO in one 45 minute meeting. The question is simple: do we permit AI training on our published content, yes or no, and why. Document the reasoning so you are not relitigating it in six months.
• Audit your current robots.txt line by line. Record what every user-agent directive does today, and tag each one as training, retrieval or indexing. Most teams we work with have not looked since 2023.
• Set a change-control rule. No one edits robots.txt without sign-off from the person who owns organic pipeline. That one rule prevents the expensive version of this mistake.

What we would not do yet: rewrite your directives in anticipation of syntax Google has not published, or treat an opt-out as a defensive marketing move. Protecting IP is a legitimate reason to opt out. Protecting traffic is not, because training preferences are not the lever that controls traffic.

What about measuring the impact if you do opt out?

Be honest with yourself: you probably cannot measure it cleanly. AI assistants pass little reliable referral data, which is why we argue that AI search ROI is effectively unprovable and should be funded anyway, using proxy metrics rather than last-click.

The practical move is to baseline before anything changes. Track branded search volume, direct traffic to your core commercial pages, and self-reported attribution on demo forms. If you want help structuring that, our measurement and attribution work is built for exactly this gap.

The shared AI training control is a governance decision, not an SEO task

The shared AI training control Google announced on 30 October 2026 is a policy choice about your intellectual property that happens to be implemented in a text file. Treat it that way. Decide your position with legal and content owners this quarter, keep it firmly separate from the crawl rules that feed answers and citations, and watch Google Search Central for the implementation detail before you touch a line.

Cover photo by Redmind Studio on Unsplash