Bing Just Split "Train on Your Pages" From "Cite Your Pages." Your Old Tags Have Until March 1.

On October 30, Microsoft announced three new preferences for how Bing's AI systems may use a website's content: AI Training, AI Grounding, and a middle setting Microsoft calls Limited Grounding. For the first time, a site can tell Microsoft "don't train models on my pages" and "do cite my pages in Copilot answers" as two separate decisions. The same day, Google said it will adopt a shared robots.txt training signal in 2027 and treat it as a Google-Extended opt-out.

For most immigration practices, the headline is good news. The fine print is where firms get caught. If your site carries an old NOARCHIVE or NOCACHE tag — and many firm websites do, without anyone at the firm knowing — those tags are quietly deciding today whether Copilot can use your pages. Microsoft has set a calendar for replacing them, and the first date is December 15.

Two Questions, Not One

Until now, Bing bundled two very different uses of your content. One is training: your pages become part of the data a model learns from. The other is grounding: when someone asks Copilot a question, Bing pulls your page in real time and uses it to build the answer, often with a citation back to you.

For a small firm, those are not the same thing. Plenty of attorneys are uneasy about their practice-area pages, FAQ answers, and blog posts being absorbed into model training. Very few want to disappear from the answer when a family asks Copilot "what happens after an I-130 is approved" or "do I need a lawyer for an RFE on my adjustment of status." Grounding is how your firm gets named in that answer.

The new preferences separate the two:

• AI Training — allow or prevent your content from being used to train AI models. This can be set site-wide in robots.txt using a new `__usage_ai-training` group, or page by page.
• AI Grounding — allow or prevent your pages from being used to build AI answers. This is page-level only, set with a `Bingbot` robots meta tag or an X-Robots-Tag header. Allow is the default.
• Limited Grounding — the middle option. Your pages can still inform Microsoft's own products, including Copilot and Bing's AI experiences, but are kept out of the grounding services Microsoft sells to other companies. (In the tag itself, the value is `disallowextended`.)

Microsoft is explicit about the trade-off: blocking grounding can reduce your visibility in AI answers. Neither grounding setting changes how your pages appear in classic Bing search results.

The Tags You Didn't Know You Had

NOARCHIVE and NOCACHE are old directives. They originally told search engines not to show a cached copy of a page. In 2023, Microsoft repurposed them as its first AI controls. Since then, Bing has treated NOCACHE as permission to use only a page's title, URL, and snippet in its chat answers, and NOARCHIVE as a request to keep the page out of those answers and out of training altogether.

That is the problem. Most firms never chose those tags. They arrive with a WordPress privacy plugin, a "protect my content" setting a vendor switched on, a hosting template, or a site migration years ago. A firm that wanted to keep its content out of training may have added NOARCHIVE and, without realizing it, also kept its asylum, DACA, and naturalization pages out of Copilot answers.

This is a different question from the robots.txt AI bot list many firms have reviewed this fall. Robots.txt governs which crawlers may fetch a page. NOARCHIVE and NOCACHE sit on the page itself, in the HTML head or in a server header, and nobody sees them unless they look at the source code.

The Calendar Microsoft Set

Microsoft laid out a transition with four dates:

• December 15, 2026 — sites that currently use NOARCHIVE or NOCACHE can record temporary grounding preferences in Bing Webmaster Tools.
• February 1, 2027 — Bing starts honoring the new AI Training and AI Grounding preferences as it recrawls pages.
• March 1, 2027 — NOARCHIVE and NOCACHE stop being read as AI training or grounding preferences. Sites that want to keep a restriction must have set the new preferences by this date.
• September 30, 2027 — temporary grounding preferences recorded in Webmaster Tools expire, or sooner for a page once Bing recrawls it and finds a new on-page preference.

Two details matter. First, after March 1, a page that still carries only the old tags, with no temporary preference recorded, defaults to Allow for grounding until Bing finds a new preference. Second, once Bing detects any new AI Training or AI Grounding preference anywhere on your site, it ignores any NOARCHIVE or NOCACHE tags still on that site. Half-finished updates can change behavior on pages you didn't touch.

Google's Parallel Move

On the same day, Google's John Mueller said Google intends to support a shared AI training signal developed at the IETF's AIPREF working group, written in robots.txt as `User-Agent: __usage_ai-training`, and aims to support it in 2027. If a site opts out of training with that token and doesn't specifically address Google-Extended, Google plans to treat it as a Google-Extended disallow. Existing company-specific controls keep working alongside it.

Treat that as a direction, not a finished standard. An IETF participant quoted in the coverage noted that significant areas of dispute remain. But the shape is clear: training and answer visibility are becoming two separate switches across the major AI providers. One caution if you add the token: keep it in its own robots.txt group. Putting it in the same group as `User-agent: *` applies the `Disallow` to every crawler, including Googlebot and Bingbot.

What to Do Before December 15

This takes about an hour, most of it with your web vendor.

1. Check for the old tags. Open a few key pages — home, a practice-area page, an attorney bio, a recent blog post — and view the page source. Search for `noarchive` and `nocache`. Ask your vendor to check the server's X-Robots-Tag headers too, since tags set there never appear in the HTML.
2. Find where they came from. Note whether the tag comes from a plugin, a theme setting, or the server configuration. That determines who changes it.
3. Decide training and grounding separately. Write down the firm's position on each. A common choice for a practice that depends on being found: no training, grounding allowed.
4. Set the new preferences on purpose. If you want to block training, add the `__usage_ai-training` group to robots.txt in its own block. Leave grounding at the default Allow unless you have a specific reason to restrict it. Then remove the legacy tags so nothing contradicts your choice.
5. Verify your site in Bing Webmaster Tools. If you are keeping a restriction and can't update pages quickly, you'll need access there to record a temporary preference after December 15.
6. Watch the AI Performance report. Bing Webmaster Tools already shows citations, cited URLs, and the grounding queries that surfaced your pages. Note a baseline now and recheck after February 1.

The Bigger Picture

For two decades, a firm's visibility choices were mostly about search results. AI answers have quietly added a second layer, and until now the controls for it were borrowed tags built for something else. Microsoft and Google are now pulling those decisions apart: whether your work trains a model, and whether it gets cited when someone needs help.

Those are choices a managing partner should make deliberately, not inherit from a plugin. The firms that settle them before March 1 will know exactly why they do or don't show up when a family asks Copilot who can help with their case.

Clientory helps immigration law firms become visible to the families and individuals who need them most — turning AI search into consultations, and consultations into clients.

See Your AI Visibility Score → CLIENTORY.ORG