What it is
Split-horizon DNS means your DNS servers give different answers for the same hostname depending on where the query comes from. It is also called split-view DNS or split-brain DNS. A laptop on the office network asks for app.example.com and gets 10.20.5.15. A phone on a mobile network asks for the same name and gets 203.0.113.40. One name, two address books.
Normal public DNS has one authoritative answer for everyone, subject to caching. Split horizon changes that by keeping separate zone files or separate private zones with the same domain name. The internal copy can contain private addresses and internal-only names. The public copy contains only addresses that work from the internet. Both copies answer to the same domain name, but they are managed as different sets of records.
This is not a load balancer and it does not check who the user is. It keys off network location: source IP address, VPN pool, cloud VPC, or the resolver that asks the question. If a client can reach the internal resolver, it gets the internal answer. If it uses a public resolver, it gets the public answer. That distinction matters when you design remote access and cloud networks.
How it works
You keep at least two zone copies. The public zone sits with a public DNS provider and is reachable from the internet. The internal zone sits on domain controllers, an internal DNS service, or a private hosted zone in a cloud provider. A query from the office or VPN hits an internal resolver. The resolver matches the client to a view or private zone and returns the internal record. A query from the open internet hits the public zone and returns the public record.
The feature names differ by platform. BIND uses views, matched by client address ranges. AWS Route 53 uses public hosted zones and private hosted zones associated with VPCs. Azure has public DNS zones and Private DNS zones linked to virtual networks. Google Cloud has public zones and private zones scoped to a VPC network. Windows DNS can run separate zones and conditional forwarders, and it can apply DNS policies. The idea is the same: the resolver's location decides which copy it reads.
Client DNS settings decide which answer a person gets. If a VPN client is configured to use the internal resolver and the internal search suffix, it sees internal names. If the VPN sends no DNS settings, or the laptop keeps a public resolver, the user gets public answers and internal names may fail. TTL values also matter. A low TTL helps during a cutover, but cached answers can still linger. Negative caching can make a missing name look broken for longer than you expect.
Why it matters for a business
Users get one simple name for an application. Staff type app.example.com at the office, on VPN, or at home. Inside the network, the name points to a private address and traffic stays on the internal path. Outside, it points to a public load balancer or reverse proxy. You do not have to teach people two names or publish private addresses in public DNS.
Split horizon also helps during a move. You can point the internal view at a new private address while the public view still points at the old endpoint, then switch the public record when the new path is ready. It keeps admin tools and staging systems out of public DNS while still giving them readable names. It can also reduce public exposure of internal network layout, because private addresses are not advertised to the internet.
The trade-off is operational. You now have two sets of records with the same domain name. Every change needs a decision about which view or zone should receive it. Split horizon is not an access control. A DNS answer tells a client where to connect, but it does not authorize the connection. Pair it with firewalls, identity checks, VPN policy, and network segmentation. Treat it as a naming and routing tool, not as a security boundary.
Common mistakes
The first mistake is treating split horizon as a security control. Hiding an internal address from public DNS does not stop someone from reaching the service if the service is exposed. It does not authenticate users. Put admin panels behind a VPN or an identity-aware proxy, and keep the network rules tight. DNS is a signpost, not a lock.
The second mistake is letting the zones drift. A public zone and an internal zone with the same name need a clear owner and a change checklist. Teams often update the A record for a web app and forget the MX, TXT, or verification records that also live under the domain. If you manage DNS as code, keep both zones in the same repository and review changes together. Compare shared records on a schedule.
Remote access is another common failure point. A VPN profile that does not push the internal DNS server, or does not push the right search domain, leaves users on public answers. Test from the office LAN, a VPN session, a home network, a mobile hotspot, and a cloud VPC. Check IPv4 and IPv6. A missing AAAA record or an IPv6 route that bypasses the VPN can produce a different answer than you planned.
Watch caching and DNSSEC too. After a cutover, old answers can stay in resolver caches until the TTL expires. Lower the TTL before the change and raise it after testing. If the public zone is signed with DNSSEC, the internal zone needs its own plan. A validator that expects a signed public answer may reject an internal answer that is not signed. Test with the same resolvers your users actually use.
How to get started
Start with an inventory. List the names that need different answers inside and outside: intranet sites, VPN gateways, admin tools, internal APIs, and staging systems. For each name, write down the public value and the internal value. Keep website, public API, and mail records in the public zone unless there is a clear reason to split them. Fewer split names mean fewer records to maintain.
Choose where each zone lives. Self-hosted BIND or Windows DNS can work if you already run internal resolvers. Cloud private DNS is a natural fit when your workloads sit in a VPC or virtual network. The public zone can stay with your domain registrar or a managed DNS provider. Make sure the private zone is attached to every network that needs it, including VPN and shared services networks.
Build a test matrix before you change production. Query the name from each network and each resolver with dig or nslookup. Confirm the answer, the TTL, and the address family. Check that clients receive the right DNS server and search suffix. Document the expected answer for each location so support staff can spot a bad result quickly. Monitor query logs for public answers to names that should be internal.
If you are unsure about the zone layout, get a second opinion before you build it. You can run the work in-house, ask your DNS provider's support team, or book a review. Cloudverg's Expert architecture session is $450/month and includes a 90 minute call with a senior architect plus a written summary. That is one option among others, and it fits teams that need a decision checked before a migration.
Frequently asked questions
Is split-horizon DNS the same as split-brain DNS?
Yes. The terms are often used interchangeably for a setup where one domain name returns different records based on where the query comes from. Some teams say split-brain when the two zones drift by accident and split-horizon when the difference is planned. The technical setup is the same.
Does split-horizon DNS work with a VPN?
It works when VPN clients use the internal DNS resolver and receive the right search domain. Push those settings in the VPN profile and test from a real remote connection. If the client keeps a public resolver, it will get public answers and internal names may fail.
Can split-horizon DNS break email?
It can if MX, autodiscover, SPF, DKIM, or DMARC records exist in only one zone or point to different names than expected. Keep public mail records in the public zone and test mail client discovery from outside. Use public certificate names for public services so TLS validation still works.