Noarchive is a value placed in a web page's robots meta tag, or in the equivalent X-Robots-Tag Hypertext Transfer Protocol (HTTP) header, asking a search engine not to offer a publicly viewable stored copy of that page. The page can still be crawled, indexed and ranked. The directive exists because search engines once kept snapshots of what they crawled and served them through "Cached" links, which let readers see text a publisher had since edited, deleted or put behind a paywall. Google moved it to a list of unused rules on October 2, 2024. Microsoft has given it a second life: as of September 2026, Bing's documentation says a page marked noarchive is not linked in Copilot, formerly Bing Chat, and a 2023 policy excludes it from training Microsoft's generative artificial intelligence (AI) foundation models.

How the directive works

The instruction takes two forms. In HyperText Markup Language (HTML), a publisher adds a meta element to the page head with the name attribute set to robots and the content attribute set to noarchive. For files with no HTML, such as Portable Document Format (PDF) files, the web server sends X-Robots-Tag: noarchive with the response. Replacing the generic robots name with a crawler's own token, such as bingbot, confines the instruction to one engine, and values combine in a comma-separated list such as "noindex, noarchive".

Sequence matters. A crawler must fetch a page before reading it, so a web address (URL) disallowed in robots.txt never reveals its meta values. John Mueller confirmed in September 2024 that this blind spot lets blocked pages reach Google's index while their noindex goes unread. Robots.txt governs fetching; meta values govern what happens after the fetch.

What the value means depends on who reads it.

  • Yandex still offers a "saved copy" of results, and noarchive removes the link to it.
  • Google lists noarchive among "historical and other unused rules" that its search engine ignores.
  • Bing treats noarchive as three instructions: store no cached page, do not link the content in Chat or Copilot, and do not use it for model training. A second value, nocache, limits Copilot and training to the URL, title and snippet. A page carrying both is treated as nocache.

Publishers, through developers and search teams, usually set the value in a template, a content management system or a server rule. Search and AI platforms decide what it means, and can change that meaning without the publisher touching a line of code. Bing's April 2020 announcement of snippet controls combined max-snippet:-1, max-image-preview:large, max-video-preview:-1 and noarchive in one example: full previews, no cache link.

The robots meta tag predates the directive. At a World Wide Web Consortium (W3C) workshop in May 1996, representatives of Excite, Infoseek and Lycos agreed six terms for it: ALL, NONE, INDEX, NOINDEX, FOLLOW and NOFOLLOW, according to a session report by Michael Mauldin of Lycos. No archive term appeared.

Noarchive followed the cache. The Cached link was a staple of Google Search from its early years, and a results page captured on December 3, 2000 already shows it, according to Semrush. No public record from Google or the W3C dates the value's introduction. In later litigation, Google's expert witness John Levine called it a widely recognised industry standard that had existed for years.

That testimony shaped Field v. Google. Blake Field, representing himself, sued over access to 51 of his works through Google's cache. On January 19, 2006, the US District Court for the District of Nevada granted Google summary judgment on non-infringement, implied licence, estoppel and fair use, and found it qualified for the system-caching safe harbour in Section 512(b) of the Digital Millennium Copyright Act (DMCA). Field knew about noarchive and chose not to use it, and the court read that choice as permission.

Belgium reached the opposite view. On February 13, 2007, a Brussels court found that Google News and the cache infringed the rights of newspapers represented by Copiepresse. Google argued the publishers could have used robots.txt and noarchive; the court rejected implied consent on the grounds that copyright is a right to prior authorisation, not a right to opt out, according to law firm Simpsons. The Brussels Court of Appeal largely confirmed the ruling on May 5, 2011.

On July 27, 2007, Google product manager Dan Crow introduced the X-Robots-Tag header, extending values such as noarchive to PDFs, spreadsheets and video files.

The larger shift came on September 22, 2023, when Bing's Fabrice Canel redefined NOARCHIVE and NOCACHE for Bing Chat and for training Microsoft's foundation models, stating that tagged pages would stay in search results. In November 2023 Bing applied the same pair of tags as the opt-out for its generative AI captions.

The original purpose then disappeared. Google's Search Liaison, Danny Sullivan, confirmed the cache link's removal on February 1, 2024, arguing that pages now load reliably. Google announced Internet Archive links in its About this result panel on September 11, 2024, deleted the documentation for its cache: search operator on September 24 and moved noarchive into a historical section of its robots documentation on October 2. Its changelog said site owners need not remove the tag, since other services may use it. Microsoft removed cache links from Bing on December 11, 2024.

Why it matters for publishers and marketers

For publishers, noarchive was a revenue control. A cached copy of a subscriber-only article was a free copy. Bing acknowledged the pattern in 2023, noting that many paywall sites used only noarchive and recommending they add nocache so their articles could still be referenced in chat answers.

For search practitioners, the cache showed what an engine had stored. Sullivan pointed site owners to Search Console's URL Inspection instead, which works only for verified sites and shows nothing of anyone else's pages.

The stakes now sit in AI answers. At Bing, noarchive decides whether a page can appear in Copilot, and the traffic those answers send is small. Microsoft's own measurements, made public on September 17, 2026 when publishers refiled a brief in the New York Times copyright case, showed click-through in Bing Chat falling 87% to 93% against Bing web search for New York Times properties, and 51% to 94% for Ziff Davis domains. The tag therefore trades a presence in answers that send few clicks against keeping text out of them, and out of training. Bing frames the choice from the other side: nocache reduces citation depth and answer quality, according to its webmaster guidelines.

Limits and disputes

Compliance is voluntary; only cooperative robots follow robots meta values, as Mozilla's MDN Web Docs reference notes. No standards body defines noarchive. Request for Comments (RFC) 9309, published by the Internet Engineering Task Force in September 2022, codified robots.txt alone, and meta values sit outside it.

Precedence rules conflict: Google applies the more restrictive instruction, Yandex the more permissive, and Bing resolves noarchive plus nocache to the weaker nocache.

Repurposing draws the sharpest criticism. A survey by Michael Dinzinger, Florian Hess and Michael Granitzer of the University of Passau, published in April 2024, found noarchive or nocache on up to 1.27% of 60 million pages sampled from Common Crawl in November and December 2023, against a robots meta tag on 52.7% of pages. The authors argued that Microsoft benefits from low adoption because more content stays available to its AI products, and that reinterpreting old tags creates ambiguity. Their own paper illustrates the point: it attributes the training exclusion to nocache and the answer exclusion to noarchive, whereas Bing's documentation gives noarchive both.

Granularity is limited. Noarchive works page by page, and bingbot offers no separation between search indexing and AI grounding at crawler level, so blocking it removes a site from Bing, Yahoo and Copilot at once.

The legal question Field and Copiepresse split on, whether silence counts as consent, remains open for AI use.

Not the same as

Noindex removes a page from search results. Noarchive leaves the page indexed and ranked and restricts only reuse of the stored copy.

Nocache is Bing's weaker companion value, limiting Copilot and training to URL, title and snippet since September 2023. MDN still calls it a synonym of noarchive, which Bing's documentation no longer supports; Google does not use it.

Nosnippet controls the text excerpt. At Google it also keeps content from being used as direct input for AI Overviews and AI Mode, a scope extended to AI Mode in March 2025. Bing's data-nosnippet attribute, introduced on October 15, 2025, hides marked sections from snippets and Copilot while leaving the page indexed.

Web archives such as the Wayback Machine run their own exclusion processes. A 2006 Internet Archive forum reply said the Wayback Machine honoured noarchive; Search Engine Journal reported in October 2024 that the tag does not affect it.

Recent developments

Google's AI levers sit elsewhere: nosnippet, the Google-Extended robots.txt token and a Search Console setting, tested from June 3, 2026, that removes a whole domain from AI Overviews, AI Mode and AI features in Discover within one to two days, without stopping model training. The UK Competition and Markets Authority's (CMA) Publisher Conduct Requirement takes legal force on December 3, 2026, with page-level controls due by March 3, 2027.

Microsoft has moved the other way. As of September 2026, NOARCHIVE keeps content out of Copilot responses and grounding results, according to Bing's webmaster guidelines. Cloudflare, meanwhile, dropped a planned block on mixed-use crawlers such as Googlebot for sites refusing AI training; since September 15, 2026, new domains carrying advertising start with AI training disallowed and agent crawlers blocked on pages with ads.

Timeline

  • May 1996: A W3C workshop defines the robots meta tag with six terms, none relating to archiving
  • December 3, 2000: A Google results page from this date shows the Cached link
  • January 19, 2006: US District Court for the District of Nevada rules for Google in Field v. Google, citing Field's decision not to use noarchive
  • February 13, 2007: Brussels court rules against Google in the Copiepresse case, rejecting the implied consent argument
  • July 27, 2007: Google introduces the X-Robots-Tag HTTP header, extending noarchive to non-HTML files
  • May 5, 2011: Brussels Court of Appeal largely confirms the Copiepresse ruling
  • April 8, 2020: Bing announces snippet controls, with an example combining them with noarchive
  • September 22, 2023: Bing redefines NOARCHIVE and NOCACHE for Bing Chat answers and model training
  • November 2023: Bing makes NOCACHE and NOARCHIVE the opt-out for generative AI captions
  • February 1, 2024: Google confirms removal of the cache link from search results
  • April 2024: University of Passau survey reports noarchive or nocache on up to 1.27% of sampled pages
  • September 11, 2024: Google announces Internet Archive links in About this result
  • September 24, 2024: Google removes documentation for the cache: search operator
  • October 2, 2024: Google moves noarchive to a historical reference section
  • December 11, 2024: Microsoft removes cache links from Bing
  • March 5, 2025: Google extends nosnippet documentation to AI Mode
  • October 15, 2025: Bing introduces the data-nosnippet attribute
  • June 3, 2026: Google begins testing a Search Console control for exiting generative AI features
  • September 15, 2026: Cloudflare sets new advertising-carrying domains to disallow AI training and block agent crawlers by default
  • September 17, 2026: Microsoft's click-through measurements for Bing Chat become public in a refiled brief
  • December 3, 2026: CMA Publisher Conduct Requirement obligations scheduled to take legal force

Summary

Who: Publishers and site owners set noarchive through templates, content management systems or server headers. Search and AI platforms interpret it: Bing uses it for cache, Copilot and training decisions, Yandex for its saved copies, and Google no longer uses it.

What: A robots meta tag or X-Robots-Tag value asking an engine not to offer a stored copy of a page, while leaving the page eligible for indexing and ranking. At Bing it also excludes the page from Copilot answers and from training Microsoft's foundation models.

When: In use alongside cached links by the mid-2000s, tested in court in 2006 and 2007, extended to non-HTML files in 2007, repurposed by Bing for AI on September 22, 2023, and retired by Google on October 2, 2024.

Where: In the head of HTML pages or in HTTP response headers, read by crawlers after a page has been fetched. It has no standing in RFC 9309 or any other formal standard.

Why: Stored copies let readers bypass paywalls and see content publishers had changed. The cache has gone from Google and Bing, but in Bing the same value now determines whether content feeds AI answers and model training, which gives an old directive new commercial weight.