Why do websites show different IP addresses?
Different IP addresses across websites can result from IPv4 and IPv6 connections, destination-based proxy rules, VPN split tunneling, or a network change between checks. Compare the address family and observation time before treating a difference as a routing problem. A different city for the same address is a separate geolocation question.
To collect evidence, open outbound IP detection and select Start detection. It asks named HTTP destinations which public address they observe from your browser. The table shows addresses, statuses and HTTP times; Copy report includes observation timestamps. This guide does not start external probes.
Read more
Exit probes originate in your browser, rather than our server, so they describe your browser’s path. The homepage shows its own observed address and a smaller, separate comparison. Browser extensions, proxy rules and VPN routes can affect which address a destination sees.
Stop detection preserves completed observations and stops unfinished requests. Run again retries all destinations; Retry missing retries failed or stopped ones. Subset retries keep other observations, so inspect their timestamps when comparing a mixed-time report. A row can retain a labeled Previous result after a retry fails, but that old address is not counted as a fresh observation. Category and search filters change the table view, not the destinations requested. Hide IPs conceals address text and redacts addresses in the copied report without changing your connection. Otherwise reports contain public addresses, so review them before sharing.
Compare different IP results by address family
Compare IPv4 with IPv4 and IPv6 with IPv6. Different families are neither agreement nor a routing mismatch. If a run returns only one address of each family, it cannot establish same-family consistency. A dual-stack connection can legitimately use both.
Read more
When two destinations report different addresses in the same version, the tool reports different observed exits. Possible explanations include destination-based proxy rules, a corporate gateway, several upstream connections, or an address change during the run. The result establishes the observed difference; it does not establish its cause. Check your intended routing rules before interpreting the difference as a fault. If an organization intentionally separates work and public browsing, different exits can be expected behavior.
Matching results have a narrower meaning than complete privacy. They show that the available same-version observations agree during this run. They do not prove that another application, another hostname, or another protocol follows the same route. A run with two matching responses and one unavailable response remains incomplete for that missing destination. Read the observed count together with the summary, rather than treating the summary as a certificate that every request reached the same exit.
The copied report includes UTC observation times and HTTP completion durations. That timing includes browser, transport and service work, with possible connection reuse. It is not ICMP ping or a reliable provider ranking from one sample.
Which destinations explain different IP results?
The outbound tool has ten default targets: NetEase, ByteDance, Cloudflare China, ChatGPT, Claude, Cloudflare, GitLab, Perplexity, and separate ipify IPv4 and IPv6 endpoints. Each table row links to its requested URL. A listed target is an attempted measurement; its presence does not mean that it will return a readable address.
Read more
The homepage comparison instead starts with this site’s address endpoint, ipify’s combined endpoint and Cloudflare’s trace endpoint. These three defaults are a different set from the outbound tool’s ten. Operator-configured echo nodes can extend either set.
The ipify documentation distinguishes its address-family endpoints: outbound uses api.ipify.org for IPv4 and api6.ipify.org for IPv6; the homepage comparison uses api64.ipify.org, which can return either family. Other outbound targets may also return either family. Switch the outbound IP version view to inspect observations separately, or use the IPv6 test for focused connectivity checks.
The trace-format targets expose address information through Cloudflare trace. NetEase and ByteDance instead depend on readable response headers. CORS restrictions, endpoint changes or network failures may prevent any of these observations. A trace response from ChatGPT or Claude describes that specific endpoint connection; it does not test account access, model availability or every request made by the service.
The Domestic category describes a group of target services, not a verified direct route or a fixed physical server location. Domain names and ownership do not establish where a request was served. Configured nodes carry operator-declared locations, which are separate from visitor geolocation. The default outbound set does not contact Google, and no finite target set represents every website in a region.
Investigate proxy and VPN routing
Investigate different exit addresses with outbound IP detection, without changing network settings during the short measurement window. If you then change a proxy rule, rerun and compare the two reports while keeping other conditions as similar as possible. The tool does not operate your VPN or change system routes. On a managed device, use the settings and troubleshooting process authorized by your organization. An unexplained difference is a reason to inspect configuration, not a reason to disable protections indiscriminately.
Unavailable responses and other limitations
Unreadable responses can arise from DNS, TLS, extensions, routing, outages or cross-origin restrictions. CORS affects whether a page can read a response. Failure alone does not establish that a destination is blocked.
Read more
Each exit probe has an eight-second deadline. Timeouts, rate limits and unexpected bodies remain separate. HTML error pages, private addresses and conflicting family data are not accepted as successful observations. Wait before retrying a rate-limited source.
The outbound tool does not perform DNS leak detection, WebRTC candidate gathering, packet capture, or a traceroute. Those mechanisms complement HTTP observations by observing different parts of a connection. The WebRTC leak test gathers browser candidates with its own disclosure and limitations. The DNS lookup queries a named resolver and is not a measurement of which recursive resolver your browser used. Keeping these distinctions visible makes the results easier to interpret alongside other tools.
Named services report these observations; HTTPS does not independently audit their infrastructure. Preserve destination and time when documenting a problem. This tool cannot infer every application’s routing policy from the listed browser connections.