How we use AI in content management at 1367 Studio

Everyone is asking the same question

Almost every project conversation this year has included some version of the same question: where does AI fit into our content?

It usually comes from a good place. Teams are publishing more, across more languages and more channels, with the same number of people. AI looks like the obvious answer.

The honest answer is that it helps, but not in the way most people expect when they first ask.

AI is only as good as the structure underneath it

The biggest difference between AI that helps and AI that creates work is the content model.

If your content lives in one large rich text field per page, an AI tool can rewrite it, but it cannot understand it. It does not know what a product name is, which paragraph is the offer, or which fields are used on three other pages.

If your content is structured, with named fields, defined types, and clear relationships, everything changes. You can ask for one field to be rewritten instead of a whole page. You can translate a section and leave the rest untouched. You can check what is missing across a hundred documents instead of opening them one by one.

This is why we spend time on schema design before anything else, and why Sanity is the platform we build content systems on. Structured content is not an AI feature. It is the thing that makes AI features usable.

Where we let AI help

We use AI for the work that is slow, repetitive, and easy to verify.

First drafts and variations. Getting from an empty field to something a person can react to is often the hardest part. A generated draft is a starting point, not the deliverable.

Migrations and cleanup. Moving content between systems, mapping old fields to new ones, and spotting gaps across large amounts of content is exactly the kind of work where automation pays off.

The repetitive layer around content. Metadata, summaries, and consistency checks across many documents take a long time by hand and are quick to review once drafted.

Agentic systems inside the workflow. Building agents that work with a content model, rather than bolting a chat box onto a website, is part of what we do for clients.

Where we do not

Nothing is published by AI. Every generated change lands as a draft and a person decides whether it goes live.

We do not let a model invent facts. Prices, results, certifications, partnerships, and quotes come from the client, not from a prediction. If information is missing, we flag it instead of filling the gap with something plausible.

We do not hand over brand voice. A model can match a tone it has seen, but deciding what a company sounds like is an editorial decision.

What this looks like in a project

In practice, the sequence is boring on purpose.

We model the content first, so each piece of information has a field and a name. We give the people who manage the content an editing experience they can actually use. Then we add AI where it removes a specific bottleneck the team has told us about.

That last part matters. The useful question is not whether a tool uses AI, but whether it makes things easier for the person managing the content. That is the question I brought into the workshops at Everything *[NYC] 2026, and it is the same one we ask before adding anything to a client project.

Where this is going

The tooling will keep moving. Content teams are already working with agents that can read a whole dataset, not just the document in front of them, and that shifts what a content operation can look like.

What has not changed is the foundation. Clear structure, clear ownership, and a person accountable for what gets published. Get those right and AI becomes genuinely useful. Skip them and you get faster ways to publish content nobody checked.

If you are thinking about how AI fits into your content operation, that is a conversation we are happy to have.

This text was written by SANITY AI BUT I’ve decided if it was good to go or not, AI can helps but the human should be on the top of that