AnswerShare shows browsers and search-engine crawlers the human page. For some pages it shows AI crawlers, and other non-browser clients, a shorter machine version written from the same published facts. This page states how that routing works, what each kind of client received when we tested it on 2026-10-02, and what the published anti-cloaking definitions say. It describes the routing as the code does it today.
Search crawlers and people receive the same human page. The machine version is public, and you can request it yourself.
We Speak AI. And We Can Prove It.
On answershare.com, a page that has a machine version is routed from the request itself: the User-Agent and the request headers a browser sends. The visitor’s IP address and network do not decide what is served. Google’s spam policies govern participation in Google Search; they do not set what an independent AI system such as OpenAI, Anthropic or Perplexity is given.
curl.Source in this repository: functions/[[path]].js (onRequest: named AI-crawler routing, then the default rule that serves the machine version to any client that is neither a browser page load nor a search crawler), functions/_shared/request-classification.js (isSearchCrawler, isBrowserNavigation) and functions/_shared/bot-ip-gate.js (matchesL1Ua). The repository also holds an older IP and network check for named AI crawlers; its result is logged and does not withhold the machine version.
We requested the pages below from answershare.com through our evidence server, sending each User-Agent shown. Every request returned HTTP 200. Sizes are the HTML bytes returned.
| Client sent as | / (home) | /faq | /isthiscloaking/ |
|---|---|---|---|
| Browser (Chrome User-Agent, HTML Accept, Accept-Language) | Human 72,083 | Human 180,438 | Same page, 24,320 |
| Googlebot, Bingbot | Human 72,029 | Human 180,384 | Same page, 24,266 |
| Googlebot-Image, Google-InspectionTool, Applebot | Human 72,029 | Not tested | Same page, 24,266 |
| Google-Extended, GoogleOther, GoogleAgent-URLContext (Google’s AI fetchers; received before the router update, see the note in section 1) | Human 72,029 to 72,083 | Human 180,384 to 180,438 (Google-Extended, GoogleAgent-URLContext) | Same page |
| GPTBot, ChatGPT-User, ClaudeBot, Claude-User, PerplexityBot | Machine 22,399 | Machine 104,002 | Same page, 24,266 |
| OAI-SearchBot, Perplexity-User, Gemini-Deep-Research | Machine 22,399 | Not tested | Same page, 24,266 |
| curl and an unrecognised fetcher, sending Accept: */* | Machine 22,453 | Machine 104,056 | Same page, 24,320 |
On the home page, the only difference we found between the human page sent to a browser-style request and the one sent to search crawlers is a web-analytics script tag, which is not added for crawlers. We have not compared the machine version with the human page byte for byte; they are different documents by design. Our test client sends from one data-centre address and cannot show behaviour from other networks, so the statement that IP address and network do not decide what is served rests on reading the router code, not on tests from other networks. The test files are kept with their headers and SHA-256 hashes.
/, /faq and /for-ai. A browser opening those addresses receives the human page.curl -A GPTBot https://answershare.com/faq. The request needs no login or key.“Cloaking refers to the practice of presenting different content to users and search engines with the intent to manipulate search rankings and mislead users.”
Google’s spam policies, quoted verbatim, retrieved 2026-10-02. Google’s separate page on dynamic rendering says: “As long as your dynamic rendering produces similar content, Googlebot won’t view dynamic rendering as cloaking.” That page also says dynamic rendering “was a workaround and not a long-term solution” and recommends server-side rendering, static rendering or hydration. We cite it for its statement on similar content, not as an endorsement of our design.
developers.google.com/search/docs/essentials/spam-policies · developers.google.com/.../dynamic-rendering
“Showing us one piece of content and other visitors a different piece of content is misleading.”
Bing expressly permits bot-specific server rendering when the content stays the same: “acceptable and not considered cloaking.”
blogs.bing.com/.../Things-you-should-avoid · .../october-2018/...Cloaking-Oh-My
Yandex’s definition is structurally identical, with an explicit carve-out: “the layout of the page differs… while the content remains consistent” is not cloaking — the same adaptive-rendering exception Google and Bing recognize. Across every crawler policy reviewed, the recurring boundary is material content inconsistency used to manipulate indexing or ranking. None makes machine-oriented delivery, by itself, the violation.
Publishes crawler purpose, robots.txt controls, access guidance, and noindex behavior — and confirms each bot's setting is independent, so a site can allow OAI-SearchBot for search while disallowing GPTBot for training. No published definition says a machine representation is cloaking.
Publishes purpose, robots.txt behavior, and anti-circumvention commitments across three separately-controllable bots. No published machine-format cloaking definition.
Publishes crawler purpose, robots.txt controls, and published IP ranges. No published machine-format cloaking definition.
help.openai.com/en/articles/12627856 · support.anthropic.com/en/articles/8896518 · docs.perplexity.ai/docs/resources/perplexity-crawlers
Googlebot, Googlebot-Image, Google-InspectionTool, Bingbot and the other search crawlers we tested receive the human page.
The machine version is written from the same published facts and is shorter and more structured. If you find a fact that differs from the human site, tell us through the Contact page.
Machine-version pages carry source URLs in the text, so a reader does not need a second lookup to find where a statement comes from.
AnswerShare serves the human page to browsers and search crawlers, and a machine version to AI crawlers and other non-browser clients. Each search engine decides for itself how to apply its own policy, and we do not guarantee any ranking, citation or policy outcome. What each client receives is shown above so that you can check it.
We Speak AI. And We Can Prove It.
Sources 01–05 were retrieved on 2026-10-02 and the quotations above were found in them; sources 06–09 were last checked when this page was first assembled. Source 10 is internal (this repository). xAI/Grok: no official crawler documentation could be located at any candidate URL — deliberately excluded rather than cited unverified.
Related: AnswerShare home · Transparency · FAQ