Bing will let sites block AI training and still show up in AI answers. Google's version works differently

Published

A lot of agencies slapped "block all AI bots" on client sites last year. Fair enough. But AI tools use your content in two very different ways, and on Oct 30 Microsoft and Google both made moves that make that split matter. If you block the wrong one, your client can quietly drop out of AI answers.

Training vs grounding, in plain words

Training is when an AI company uses your pages to build its next model. You don't get a link or a visit out of it.

Grounding is when an AI tool pulls your page in real time to answer someone's question, often with a citation and a link back. That's the part that gets a local business named in an AI answer.

Most people want to say no to the first and yes to the second. Until now, the controls mostly lumped them together.

What Bing announced

On Oct 30, 2026, Krishna Madhavan, Principal Product Manager at Microsoft Bing, announced two separate preferences you can set independently:

Both default to Allow if you set nothing. So you can block training, leave grounding alone, and stay eligible to be cited in Copilot.

Microsoft says none of this changes eligibility for regular Bing search results. It also warns that choosing Disallow or DisallowExtended for grounding will reduce your visibility in AI experiences. And allowing either one doesn't guarantee inclusion, traffic or payment.

The Bing tags

Training for a whole site, in robots.txt:

User-Agent: __usage_ai-training
Disallow: /

Training or grounding for one page, as a meta tag:

<meta name="Bingbot" content="ai-training:disallow">
<meta name="Bingbot" content="ai-grounding:disallow">

Or as a header: X-Robots-Tag: Bingbot: ai-grounding:disallow. Use ai-grounding:disallowextended for the Microsoft only option.

One gotcha from Microsoft: these settings are about how content is used, not whether Bing can crawl it. An Allow here won't undo a robots.txt rule that blocks Bingbot.

It's not live yet. Here are the dates

Bing only sees a new tag when it recrawls the page. IndexNow can nudge that, but Microsoft says it doesn't guarantee a recrawl date.

What Google announced, and why it's different

Same day, Google's John Mueller posted that Google plans to support a shared AI training control in robots.txt, based on work from the AIPREF working group at the IETF. That's the same __usage_ai-training token Bing uses. Google's crawler docs say support is coming in "early 2027." The blog post just says 2027.

Here's the part people will miss. Google is only adopting the training signal. It is not adding a separate grounding switch like Bing did.

If a site disallows __usage_ai-training and doesn't mention Google-Extended, Google plans to treat it as if the site disallowed Google-Extended. And per Google's own docs, Google-Extended covers training Gemini models and grounding in Gemini Apps and Grounding with Google Search on Vertex AI. Google's post says sites that allow Google-Extended "are eligible to appear in responses and receive clicks via grounding in Gemini Apps."

So on Google, blocking training with the shared token can also cost a client citations in the Gemini app. If you want "no training, but yes Gemini citations," that clean split isn't on the menu.

What it doesn't touch: Google says Google-Extended does not affect a site's inclusion in Google Search and isn't a ranking signal there.

An explicit Google-Extended rule still wins over the shared token, and Google says its existing company specific controls keep working.

OpenAI and Anthropic

Search Engine Roundtable reports OpenAI and Anthropic are on board with the shared training signal too. The draft itself, "Machine-Readable Training Preference Signal," was filed Oct 30 by Kevin Kelley of Anthropic.

One caveat: that draft is an individual Internet-Draft. The IETF Datatracker says it "is not endorsed by the IETF and has no formal standing in the IETF standards process." Nate Hake, who has been active in the working group, also pointed out on X that it's not actually done at the IETF level and some big disputes remain. Treat it as where things are heading, not a finished standard.

Why blanket AI bot blocks can backfire

The common move was a robots.txt that blocks every AI crawler it can name. That kills training, sure. It can also kill the crawls that feed live AI answers, which is the whole reason a plumber or dentist shows up when someone asks ChatGPT, Copilot or Gemini for a recommendation.

For most local businesses, being cited is the goal. Training is the trade nobody's paying for. The new controls let you be choosy, at least on Bing.

What to check on client sites this week

Plain takeaways

Sources

Know when prices change

One short email a week. What changed in local SEO tools, and what it costs you. No pitch.