Connect Contentful to AI: know what is in the space, and in which language
Contentful can be connected to AI. The assistant looks up entries inside the space you point it at, per content type, and can prepare a new entry with the same fields as the entries already there. Such a new entry is a draft. Publishing is a separate act in Contentful and stays that way here: nothing goes live unless somebody asks for it.
What the assistant can do in the space
Look up entries per content type
The assistant searches inside one kind of item, for example article, product or location. That way no random pile of content comes back, only the type of entry the question is about.
Find the right entry in a full space
Even with thousands of entries in there, the assistant gets to the handful that matter. It states whether an entry is published or still waiting as a draft.
Prepare an entry as a draft
A new entry gets the same fields filled in as the entries of that type that already exist. It stays a draft: visible to the editors, not yet picked up by the website or the app.
Publishing stays a separate step
Going live is an act in itself in Contentful, and the assistant performs it only on an explicit request. A half-finished text therefore never slips into the shop window unnoticed.
Writing in the right language
The fields of an entry sit separately per language. The assistant works in the language the question is about and checks when that is unclear, so a Dutch text does not land in the English field.
One space per connector
The connector belongs to the space you point it at. If you keep separate spaces for brands or markets, each space gets its own connector and content cannot get mixed up.
Where this saves work in practice
Content team
- Situation
- Before a new campaign it has to be clear which articles on a theme already exist and which of those are still drafts.
- What the agent does
- The assistant looks up the articles on that theme and lists them with their date and their status.
- Result
- The team knows what can be reused and what genuinely still has to be written.
E-commerce
- Situation
- A new product line means dozens of product texts, in two languages and all following a fixed pattern.
- What the agent does
- The assistant prepares an entry per product with the same fields as the existing products, in the agreed languages.
- Result
- The editors start at proofreading instead of at data entry, and publish themselves once the series is complete.
Customer service
- Situation
- An agent wants to know what the site says about a warranty period, but the site search does not know the internal wording.
- What the agent does
- The assistant looks up the relevant entry and returns the text exactly as the customer receives it.
- Result
- The same answer the customer reads, without the employee having to open the CMS.
Real estate
- Situation
- A new branch has to appear on the site with opening hours and contact details, in the same shape as the existing branches.
- What the agent does
- The assistant looks at how an existing branch is built up and prepares a new entry on that model.
- Result
- The location is ready with the same fields filled in, waiting to be checked and published.
How the connection to your space is made
-
1
Linking the space
In IntraGPT you switch the connector on and point out the space it works with, using an access key you issue in Contentful yourself and can withdraw yourself.
-
2
Choosing content types and roles
You decide which content types take part and which roles in IntraGPT may work with the space. A support assistant can then only read, while the editorial team may also prepare entries.
-
3
What the assistant may and may not do
Looking things up is on by default. Preparing an entry and publishing it are separate choices, so you decide whether and when the assistant may put anything into the space.
-
4
What lands in the log
Every action is recorded with who asked for it and what the outcome was. For a prepared or published entry you can therefore see when it happened and which entry was involved.
Content that may not go out yet
A space usually holds more than what is visible on the site: prices that do not apply yet, texts for a launch and material under embargo. None of that should go live by accident or surface in the wrong answer.
Where the data lives: Nederland of de EU
Unpublished stays unpublished
Whatever the assistant prepares is and remains a draft. Going live happens only when a person asks for it, so a launch text does not go out a week early.
Only the content types you point out
The connector works within the content types that were agreed. A type with internal notes or with data that must not go out is simply left off it and stays invisible.
Traceable in the log
Every action lands in the log with its outcome. That makes it possible to check afterwards which entry was created or published by whom, so nobody has to rely on memory.
On our own server
The model runs on our own server in the Netherlands. Unpublished entries stay inside that environment and are not sent to a model provider or used for training.
Frequently asked questions about Contentful and AI
Can I connect Contentful to ChatGPT?
ChatGPT has no connection to your space. IntraGPT offers the same way of working, but with a connector tied to one space that shows only the agreed content types and records every action. Publishing stays something a person asks for.
Does IntraGPT work with Contentful?
Yes. Contentful is one of the connectors IntraGPT supports as standard: looking up entries, preparing an entry and publishing an entry. The content simply stays in Contentful; no copy of your space is created.
Will the assistant put content live on our website by itself?
No. A new entry is created as a draft and stays one until somebody asks for publication. That is a separate act in Contentful, and the assistant performs it only on request. To rule it out, leave publishing out of the connector: the assistant can then prepare entries but take nothing live.
How do we prevent things being written in the wrong language?
By making the language explicit. In Contentful the fields of an entry sit separately per language, so a title exists side by side in Dutch and in English. The assistant always works within one language and checks which one it should be when the question does not say. An English field then stays empty instead of filling up with Dutch text.
Does it work if we have several spaces?
Yes, with a connector per space. Many organisations deliberately keep brands, markets or a test environment apart, and that separation stays intact. Each space gets its own key and its own connector, so an entry cannot land in the wrong space and you can decide per space who works with it.
What if our content types have very particular fields?
That is the rule rather than the exception and it is no problem. The assistant looks at how existing entries of that type are built up and follows the same pattern. Only for a content type that is still virtually empty do we agree during setup which fields are required, so that nothing has to be guessed.
What does a Contentful connector cost?
The connector is part of the project we run together. The work is driven by the number of spaces, how many content types take part, the languages you use and whether the assistant may prepare and publish as well.
How long does setup take?
The connection is made quickly and can be tested the same day. Most of the time goes into walking through the content types and the languages: which take part, which fields are required and who may have something prepared. Start with looking things up; in practice that already answers a lot of questions.