Should your blog be on a subdomain or a subfolder for AI search?

Put a new blog in a subfolder. The reasons are operational and classic-search ones, not measured AI citation effects, and that distinction is the whole point of this page: a subfolder is easier to run, keeps one analytics property and one set of redirects, and avoids an argument every time somebody asks whether the hostname is costing you anything.

If your blog already lives on a subdomain, leave it there unless something else justifies the move. A migration consumes engineering time, puts every URL through a redirect chain, and buys you a benefit nobody has measured for AI answers. The same quarter spent rewriting the twenty pages that answer your buying questions has published evidence behind it.

The claim you will meet while researching this is that answer engines evaluate trust at the domain level, so a subdomain inherits less of it. The idea is plausible and it is also unsupported. Searching the live results for this question in October 2026 returns SEO vendor blogs asserting the effect, with no study comparing citation rates for the same content served from a subdomain and from a subdirectory. Asserted and measured are different things, and on this site we try not to blur them.

So the decision is a practical one with a default. New content goes in a subfolder because it is simpler. Existing content stays where it is because moving it is expensive and the gain is speculative. What follows explains what the evidence does support, so you can make the case either way in front of a sceptical engineering lead.

Does a subdomain lose AI visibility compared with a subdirectory?

No published measurement shows a subdomain losing AI citations against a subdirectory. The honest state of the evidence is that nobody has run the experiment, which would require serving comparable content from both and comparing citation rates over months.

What makes the claim sound settled is that it borrows an older argument. The subdomain-versus-subdirectory debate in classic SEO is decades old, and it was always about how link signals consolidate. Generative engines retrieve and quote pages rather than ranking them by link graph alone, so the older argument does not carry over automatically, and repeating it with "AI" attached is not evidence.

There is one mechanism that gives the concern a reasonable basis, and it is worth stating precisely instead of overstating it. Chen and Koudas, University of Toronto, October 2026, found that the engines partly decide what to search for from what they already know, with websites in a model's prior knowledge for a question having 1.60 times the odds of being cited at equal retrieval rank. If an engine treats a new hostname as an unfamiliar website, a brand-new subdomain starts without whatever familiarity the main domain has accumulated. Whether engines do treat hostnames that way is exactly the thing nobody has measured.

Our position, then, is narrow. Prefer a subfolder when you have the choice, say the reason is operational simplicity and classic search, and do not tell a client that a subdomain is costing them citations, because you cannot show it. A supplier who quotes a migration on that basis is selling a project, not a finding.

What does the measured evidence say about site-level and page-level signals?

The measured evidence says most of what separates cited pages from ignored ones sits inside a website rather than between websites. Chen and Koudas, University of Toronto, October 2026, using a within-website design across two 90-day collections totalling 158,099 answers and 468,037 citations, found 13% of the variation in citation counts between websites and 87% between pages of the same website.

Inside a site, two features survived all five of their robustness tests: how well a page's title matched the question asked, worth +14.1% citations per standard deviation (95% confidence interval 10.7 to 17.6), and how well its best paragraph matched, worth +15.5% (11.1 to 20.1). Neither has anything to do with the hostname. A page on a subdomain written in the words buyers use beats a page in a subfolder written in internal language.

The same paper found the structural features people usually argue about behaving differently across sites and within them. FAQ blocks, summary boxes, author boxes, schema.org markup and page speed all looked positive between websites and none was significant within one, which the authors attribute to Simpson's paradox. They present their findings as associations, not causal effects, and one preprint is not a literature.

Where your own domain does matter unambiguously is in being the thing cited at all. Our study of how four AI engines answer pharmaceutical process skid sourcing questions found between 73% and 89% of citations pointing at suppliers' own capability pages instead of directories or marketplaces, consistently across six categories. The engines were reading suppliers' own sites. Which hostname those pages sat on was not what decided whether they were quoted.

What does Google document about subdomains and subdirectories?

Google documents a preference in one specific case and a cost in another. For geotargeting, Google's URL structure documentation, last updated 21 October 2026, recommends a country-specific subdirectory on a generic top-level domain as one of the ways to make a multi-regional site easy to geotarget, listing it as the recommended option alongside a country-code domain.

The cost shows up in the migration guidance. Per Google's documentation on site moves with URL changes, last updated 20 August 2026, the Change of Address tool applies when "moving from one domain or subdomain to another", and is not needed for "moving paths within the same domain". That distinction is the practical difference between the two structures: a subfolder reorganisation is a path change, while moving a blog off a subdomain is a hostname move with the heavier process attached.

The same page tells you what that process costs. Google advises using server-side permanent redirects where technically possible and keeping them in place for at least a year, which means the migration is not finished when the content moves. It is finished a year later, assuming nothing in the chain breaks in between.

What Google does not publish is a statement that subdomains rank or get cited worse than subdirectories. Neither does it publish anything about hostname structure and AI Overviews or AI Mode specifically. So the documentation supports a mild preference for subdirectories in one scenario and a clear picture of what a hostname move costs, which together argue for choosing a subfolder up front and leaving an existing subdomain alone. Our page on whether a redesign or migration hurts AI visibility covers what to watch if you do move.

Subfolder, subdomain or separate domain: which should you pick?

Pick the subfolder unless your platform forces a subdomain, and keep a separate domain for a genuinely separate brand. Those three structures cover almost every case, and the choice is usually decided by who runs the platform rather than by search. The table sets out what each one is good for and what it costs you.

Structure	Best for	What it costs	What the evidence says for AI search
Subfolder, example.com/blog	Almost every B2B blog and resource library	Your CMS has to live under the main site, which is sometimes an engineering constraint	No measured citation advantage, and the simplest structure to run and to measure
Subdomain, blog.example.com	A platform you cannot host in a path, or a separate team with its own stack	Separate analytics and redirect handling, and a hostname move later if you change your mind	No measured citation penalty. The concern rests on an unmeasured assumption about how engines treat hostnames
Separate domain	A distinct brand, a joint venture, or content that must not be associated with the main site	Everything starts from nothing, including whatever familiarity the engines have with your main domain	The one structure with a clear downside, because brand and site familiarity measurably affects what an engine searches for

One case genuinely argues for a subdomain and it has nothing to do with search: a documentation or community platform you cannot reverse-proxy into a path without fighting your stack every release. Choosing it knowingly is fine. What is not fine is choosing a separate domain for a content library because the brand team prefers the name, since that is the option where the evidence does point at a real cost.

What should you do if your blog is already on a subdomain?

Leave an existing subdomain in place and spend the migration budget on the pages themselves. The measured page-level levers, title and paragraph wording matched to the questions buyers ask, are available on any hostname, and they are the ones with a confidence interval attached instead of an assumption.

Three checks are worth running on a subdomain blog, because a split hostname does hide a few real problems. Confirm the subdomain is not excluded by its own robots.txt, since hostnames carry separate files and an inherited staging rule is a common find. Confirm the pages render without JavaScript, because a separately built blog platform is where that breaks. And confirm your main site links to the content, because an orphaned hostname that nothing points to is harder for anything to discover.

Then look for the duplication that split structures produce. Teams with a blog on one hostname and a resources section on another often answer the same buyer question twice, which splits your own retrieval chances and leaves neither page the obvious one to cite. Consolidating those pairs is usually worth more than relocating either hostname, and two pages competing for the same AI answer works through how to pick the survivor.

If a migration is already justified for other reasons, a replatform or a consolidation of three CMSs, then fold the structure change into it and keep the redirects for at least a year as Google advises. What does not justify it on its own is AI search. If you want the question settled with evidence from your own pages before the argument reaches a planning meeting, our free visibility assessment scores them and reports what is actually costing you citations.

What else do people ask about subdomains and AI search?

The five questions below about subdomains come up when a marketing team inherits a blog on a different hostname and has to decide whether moving it is worth a quarter of engineering time.