Most companies end up with a subdomain because it was easier. The marketing team wanted a hosted CMS, the hosted CMS wanted a CNAME, and blog.example.com was the path of least resistance. Nobody made an architecture decision. An architecture decision happened.
The consequence is that your blog and your product site may be accumulating authority separately rather than together, and the way AI systems build a picture of a domain makes that split more costly than it used to be.
The Traditional Argument Was Never Fully Settled
Search practitioners have argued about this for two decades. Google's position has generally been that it can handle either, that subdomains are not inherently penalised, and that the choice should follow whatever makes operational sense.
Practitioner experience has generally been that consolidating onto subdirectories produces better outcomes, which is why so many companies have migrated blog.example.com to example.com/blog and reported gains. Both things can be true. A system that technically handles both can still produce different results depending on how signals accumulate.
I am not going to relitigate that argument here because AI retrieval adds a consideration that does not depend on resolving it.
Entity Consolidation Is the New Argument
The mechanism that matters is described in the entity based AEO guide. AI systems build an understanding of an organisation as an entity, then evaluate content in relation to that entity.
A subdomain is a technically separate host. Whether an engine treats blog.example.com as part of the same entity as example.com depends on how strongly the connection is signalled: shared Organization schema, consistent branding, cross linking, and matching sameAs references.
When those signals are strong, the split costs little. When they are weak, which is common because subdomains often run on separate platforms with separate templates and separate schema implementations, you have two partial entities where you wanted one.
Subdirectories sidestep the question entirely. Same host, unambiguously the same entity, no signalling required.
The Separate Platform Problem Is the Real Issue
Here is what I think actually drives the observed difference, more than any host level ranking logic.
Subdomains usually exist because the content runs on different infrastructure. Different infrastructure means the schema on your blog was configured by whoever set up the hosted CMS, and the schema on your product site was configured by your developers, and nobody checked whether the Organization objects match.
So you get two different Organization entities with slightly different names, different logo URLs, and different sameAs arrays. That is the disambiguation failure from the entity disambiguation guide, self inflicted through infrastructure.
The subdomain did not cause that. The separate platform did. But the two travel together often enough that the correlation is real.
When a Subdomain Is Genuinely the Right Call
Distinct audiences and distinct topical territory. A developer documentation site at docs.example.com serves a different reader with different queries than the marketing site, and there is a reasonable argument for keeping the topical signal clean rather than blending technical docs into a marketing domain.
Language and regional splits, where a subdomain per market is a legitimate structure covered in the multilingual guide.
Genuine platform constraints. If your ecommerce host cannot serve a subdirectory and rebuilding is not viable, a subdomain with strong entity signalling beats a rebuild you cannot afford.
What is not a good reason: the CMS was easier to set up that way. That is how most subdomains happen and it is not a decision, it is a default.
If You Are Staying on a Subdomain
Make the entity connection explicit rather than hoping it is inferred.
Publish identical Organization schema on both hosts, with the same @id, the same name string, the same logo URL, and the same sameAs array. Identical, not similar.
Cross link in both directions with real navigation links, not just a footer entry.
Keep branding consistent enough that a human landing on either would not question whether they are the same company, because the signals a machine reads are downstream of the signals a person reads.
List both in a single sitemap index, as covered in the sitemap guide, so the relationship is declared at the discovery layer too.
If You Are Migrating
This is a full migration with all the risk that carries, and the migration guide applies in full. Capture a citation baseline before you move, flatten redirects to one hop, and expect the staggered recovery pattern where live retrieval engines catch up quickly and training dependent ones take considerably longer.
My honest read: migrate if the subdomain exists by accident and there is no operational reason for it. Do not migrate a subdomain that exists for a genuine reason just to consolidate authority, because the migration cost is real and immediate while the consolidation benefit is gradual and partly recoverable through the signalling above.
Checking Which Situation You Are In
Ask an engine about your company and see whether it describes the whole business or only the part living on your main domain. If your blog covers topics the answer never mentions, the entity connection may not be landing.
The NotionCue Citation Tracker can run prompts covering both territories, which shows whether both hosts are contributing to the same picture or whether one is invisible. The AI Crawler Audit checks schema delivery on each host separately, which is where mismatched Organization objects surface.
Start your free NotionCue trial and audit both hosts as separate targets. Teams running split architectures frequently find the schema diverged years ago.
Free check that takes two minutes: view source on your main domain and your subdomain, find the Organization schema in each, and compare the name and logo fields character by character. If they differ at all, you have two entities.
Common Questions
Does Google treat subdomains as separate sites?
Google has said it handles both and treats them contextually rather than by a fixed rule. The practical concern in this post is less about Google's treatment and more about whether your own implementation makes the connection legible.
Is a subfolder on a CDN the same as a subdirectory?
If it serves from the same hostname, yes for these purposes. Reverse proxying a hosted CMS onto a subdirectory path is a common way to get subdirectory benefits without migrating platforms.
What about a completely separate domain for a content site?
That is a stronger version of the same split, with no shared host signal at all. It can work for genuinely independent properties and it is the hardest configuration to consolidate authority across.