Connecting Twitter / X to AI: following along without anything going out unseen
Twitter, now X, can be connected to AI. The assistant searches recent posts, reads the timeline of an account and looks at public profiles. It can also publish a post, but that is a public statement on behalf of the organisation: it happens only on explicit request and only after the full text has been shown.
What the X connector can do
Search recent posts with intent
Search on a name, a hashtag or an account across what was posted recently, within a period you specify. Reposts can be left out, so you keep the real reactions instead of the same message a hundred times.
The posts of one account
The most recent posts of a specific account can be retrieved, without the replies and reposts if you prefer. Useful for seeing what a competitor or a partner has put out lately.
Look at a public profile
Given a username the assistant looks at the public profile. That is often the first step before requesting the posts of an account, and it confirms straight away that you have the right one.
Publish a post
The assistant can publish a post, optionally as a reply to an existing one. This is the only thing that genuinely goes out, so it happens only on explicit request and only after the text has been put in front of a person.
A question, not a dragnet
The assistant asks a sharp question instead of exploring broadly. That produces a list someone can still read, and it keeps hundreds of irrelevant posts from passing through the model.
What teams actually do with it
Communications
- Situation
- During an outage the press officer wants to know whether people are already posting about it and in what tone.
- What the agent does
- The assistant searches the brand name over the past few hours, leaves reposts out and groups the posts by topic and tone.
- Result
- A factual picture of what is happening before a statement is drafted, instead of an impression from two random posts.
Customer service
- Situation
- Customers ask questions by tagging the company account, but nobody watches that consistently.
- What the agent does
- The assistant periodically retrieves the posts mentioning the account and picks out the ones that need an answer.
- Result
- Open questions reach the work queue instead of sitting there for days.
Education and research
- Situation
- A communications officer has to report monthly on what the institution has published.
- What the agent does
- The assistant retrieves the posts of the own account for the period and lists them by theme with the core of each one.
- Result
- A monthly overview ready in minutes that someone only has to check.
Events
- Situation
- A hashtag is used around a conference and the organisers want to know what is being said.
- What the agent does
- The assistant searches the hashtag within the days of the event, excludes reposts and summarises the recurring remarks.
- Result
- Usable feedback for the next edition, gathered without anyone reading along all day.
Marketing
- Situation
- An announced message is ready and has to go out at the right moment.
- What the agent does
- The assistant drafts the text from the knowledge base, shows it in full and publishes only after a person has approved it.
- Result
- A post that matches the house tone and a clear record of who approved it.
Connecting the X account
-
1
Connecting through the platform
An administrator connects the organisation's X account in IntraGPT. Access can be withdrawn at any moment, after which the connector can do nothing at all.
-
2
Publishing as a separate switch
Reading along and publishing are two different decisions. Most organisations start with search and read only, and enable publishing once it is agreed who may ask for it.
-
3
Using the assistant
After that someone asks in the chat what is being said about the organisation, or puts a draft forward. A proposed post is always shown in full first.
-
4
Every post traceable
What was requested and what was published lands in the log. For a published post that establishes when it went out and in which conversation it was requested.
Publishing is the risky side, not the reading
A search on X only reads what is already public. Publishing a post is something else entirely: a statement on behalf of the organisation that is visible at once and cannot be quietly undone.
Where the data lives: Nederland of de EU
Show first, publish after
The assistant presents the full text before anything is published, including the post it would be replying to. Without an explicit approval nothing goes out.
Publishing can stay off
If you only want the connector for reading along, you leave publishing disabled. The assistant can then search, read timelines and look at profiles, and nothing more.
Not everyone gets to ask
You decide which roles may reach the connector. In practice only the communications team gets the ability to have something published, the rest read along.
Processing on our own server
The retrieved posts are processed on our own server in the Netherlands with open-weight models. The drafts the assistant writes do not pass an external model provider and nothing is trained on them.
Questions about the X connector
Can I connect Twitter or X to ChatGPT?
Not in a way an organisation should want. You would put access that can publish on behalf of the company account into a public service, without roles and without a log. In IntraGPT, X sits inside the platform as a connector, you decide per role who may reach it, and every request is logged.
Does IntraGPT work with X?
Yes. X is in the connector catalogue. The assistant can search recent posts, read the posts of an account, look at a public profile and, if you want, publish a post. That last one can stay switched off.
Can the AI tweet by itself?
Only if you enable it and only on explicit request. The assistant shows the full text first, and the post is published only after approval. A post on X is public instantly and gets copied before you can delete it, which is why the heaviest safeguards sit here.
Why can the assistant not find an older post?
Because the search on X looks at recent posts rather than the full archive. For the question of what is being said about the organisation right now that is no obstacle. For a look back across a whole year, use a source where those posts are kept, such as your own reporting.
Can the assistant schedule posts for later?
No. Publishing happens at the moment a person approves it, or it does not happen. Anyone who wants to publish at a fixed time keeps using their own scheduling tool. The upside is that nothing ever goes out at a moment when nobody is watching.
Can the assistant read or answer direct messages?
No. This connector covers public posts, timelines and profiles. Direct messages are not part of it. If you want a private channel with customers handled, a messaging channel such as WhatsApp or a shared mailbox is the logical place.
What does an X connector cost?
Connecting is part of the project we run together. The work depends on whether it is read only or publishing too, how many accounts and standing searches are involved and which roles get access. X also charges for business access to the platform itself, which you arrange directly with X.
How long does setup take?
Connecting itself is a short job and searching can start the same day. The time goes into establishing which searches genuinely matter and into the agreement about who may have something published.