The tag that outlived its original job
Noarchive is a small instruction in a page’s robots meta tag or X-Robots-Tag header. It asks a search engine not to show a stored copy of a page while still allowing the page to be crawled, indexed and ranked. For years, that meant removing the “Cached” link that search engines once placed next to results.
The cached link is now gone. Google removed it on February 1, 2024, added Internet Archive links to “About this result” on September 11, and moved noarchive to a historical, unused rules list on October 2. Bing removed its own cache links on December 11, 2024.
Where noarchive still flexes
But the directive is not retired everywhere. Bing has repurposed it. As of September 2026, Bing’s documentation says a page carrying noarchive is not linked in Copilot, the assistant formerly called Bing Chat, and a 2023 policy excludes it from training Microsoft’s generative AI foundation models.
- Google ignores noarchive and points site owners to Search Console’s URL Inspection tool, which only works for verified properties.
- Bing treats noarchive as three instructions: no cached page, no link in Copilot, and no use for model training.
- Yandex still offers a saved copy, and noarchive removes the link to it.
A related value, nocache, is Bing’s weaker companion. It limits Copilot and training to the URL, title and snippet. If a page carries both noarchive and nocache, Bing treats it as nocache, according to PPC Land’s explainer.
Why marketers should care now
The commercial weight has shifted from cache links to AI answers. For publishers, a subscribed article previously leaked through a cached copy; noarchive was a revenue control. Now the same value decides whether a page can appear in Copilot and whether Microsoft can use it for training.
Microsoft’s own click-through data, made public on September 17, 2026 in the New York Times copyright case, showed Bing Chat clicks falling 87% to 93% against standard Bing search for New York Times properties, and 51% to 94% for Ziff Davis domains. The trade-off is therefore less about losing a traffic firehose and more about keeping text out of AI products.
A University of Passau survey from April 2024 found noarchive or nocache on only up to 1.27% of roughly 60 million pages sampled. Low adoption matters because it leaves most content available to AI systems that reinterpret old controls. In other words, if you are not actively managing these tags, your default may be participation.
What to do about it
Start with a template audit. Check whether your CMS, server headers or robots meta definitions still deploy noarchive, nocache, noindex or nosnippet. Then assign them to a clear use case rather than a reflex. If you want Bing Copilot to cite a paywalled article without exposing the full text, add nocache. If you want no presence in Copilot or training, keep noarchive.
For Google, the AI controls sit elsewhere: nosnippet for AI Overviews and AI Mode, the Google-Extended robots.txt token, and a Search Console toggle tested from June 3, 2026 to remove a domain from AI features without stopping model training. The UK CMA’s Publisher Conduct Requirement takes legal force on December 3, 2026, with page-level controls due by March 3, 2027, so governance is getting formal.
Also remember sequence: crawlers must fetch a page before reading meta values, so pages blocked in robots.txt never reveal their noarchive or noindex. Blocking bingbot entirely removes a site from Bing, Yahoo and Copilot at once because there is no separate AI opt-out crawler.
Source: PPC Land
