Anthropic did block the pages. The share-link path was disallowed in robots.txt, and Google indexed the URLs anyway.
Over the weekend of July 25–26, 2026, hundreds of conversations shared through Anthropic's public "share" links were sitting in Google and Bing results, findable with a `site:claude.ai/share` query, among them chats about health, relationships and other private subjects. Wired reported it first. The pages themselves never carried a noindex tag, and per Google's own crawling documentation that is the difference that decides the outcome: a robots.txt disallow stops the crawler from fetching a page, not the search engine from listing it. Google can index a URL it learns about from an external link without ever reading what is on it.
Google spokesperson Ned Adriance told Search Engine Land on July 28, 2026 that Google "always respect[s]" site owners' indexing directives. The gap traced to Anthropic's configuration, not to a Google override. Anthropic pulled the exposed conversations out of the search indexes within days of the Wired report. Bing had indexed the same URLs. Microsoft declined to comment as of July 28, 2026.
Credibility: MEDIUM. Directly verified against one outlet (Search Engine Land, July 28, 2026), which itself cites Wired's original report and carries an on-record Google statement. That is two-source corroboration by citation, not by independent fetch: Wired's original piece was not opened in this run. No known bias, and Google's technical explanation is checkable against its own public crawling and indexing documentation.
Fact bullets:
- Hundreds of shared Claude chat URLs, findable via `site:claude.ai/share`, were indexed in Google and Bing as of the weekend of July 25–26, 2026. Wired reported it first.
- Anthropic blocked the share-link path via robots.txt but did not add noindex meta tags. Per Google's documentation, a robots.txt block alone does not prevent indexing.
- Google spokesperson Ned Adriance told Search Engine Land (July 28, 2026) that Google "always respect[s]" site owners' indexing directives. The gap traced to Anthropic's configuration, not a Google override.
- Bing also indexed the same URLs. Microsoft declined to comment as of July 28, 2026.
Sources:
Search Engine Land: "Google indexed Claude Chats because Anthropic didn't block your private chats from search engines," July 28, 2026, https://searchengineland.com/google-indexed-claude-chats-because-anthropic-didnt-block-your-private-chats-from-search-engines-483748
Stay ahead of AI search
The changes that move your rankings and AI citations — and the exact move to make — before they cost you. Free, three times a week.
For any business using or building AI tools that hand out public share links (a chat export, a dashboard snapshot), the operative fact is that robots.txt is a crawl control, not a privacy control.
Google's statement is worth reading closely, because it is accurate. It respected the directive it was given. That directive said do not fetch this page; nothing in it said do not list this page. Those are different jobs: robots.txt manages what a crawler spends its time on, and a noindex tag decides whether a page can appear in search at all.
There is a wrinkle worth knowing before someone on your team fixes this the obvious way. Adding a noindex tag to a page that is already disallowed in robots.txt does nothing, because the crawler is never allowed to fetch the page and read the tag. For noindex to be seen, the page has to stay crawlable. For anything that must never surface at all, the setting that actually delivers that is authentication, not a directive in a text file.
None of this is exotic. It is the oldest file on the web behaving exactly as documented, and it still caught the company that thinks harder than most about how machines read text. The block was real the whole time, which is precisely why nobody went back to look at it.
Before the next vendor call, the thing to get in writing is narrow: is a link this tool generates indexable by default, and who at the vendor will confirm that.
Recommended actions:
- Ask your web team to list every URL pattern the site blocks in robots.txt, plus any AI tool your team uses that mints public share links, and confirm whether those pages can still be listed without being crawled. Where privacy matters, the fix is a noindex tag or an X-Robots-Tag header, not another disallow line.
- Ask the vendor in writing, before adopting any AI tool that creates shareable links, whether shared URLs are indexable by default and whether noindex is applied. A product label that says private is not a statement about search engines.
- Record the decision rationale whenever a page must stay crawlable for AI-citation purposes but excluded from classic search results. Noindex and AI-crawler access are separate settings your web team controls independently, and conflating them risks either accidental exposure or accidental loss of AI-citation eligibility.
Search Engine Land: "Google indexed Claude Chats because Anthropic didn't block your private chats from search engines," July 28, 2026, https://searchengineland.com/google-indexed-claude-chats-because-anthropic-didnt-block-your-private-chats-from-search-engines-483748