How this system works

A draft. Not a published communication, not a decision, and not advice. Someone in the organisation reads every draft, edits it, and decides whether it is used at all.

1. Someone picks a template, such as a press release or an FOI response, and describes what they need.
2. The system draws on that organisation’s own uploaded documents: the ones the person chose, read in full where they fit, or, if they chose none, the most relevant passages found by searching. House style, standing text and the template’s outline are always included.
3. That material, the request, and a quality checklist for the template are sent to an AI model, which writes the draft. So is any writing guidance the person asking has set for their own drafts, such as who they write for or words their department avoids. It shapes how the draft is written, never what it says is true, and the organisation’s house style comes first.
4. The draft comes back with numbered citations pointing at the documents it drew on. Before it is saved, the system checks that every citation points at a source that was actually retrieved.
5. A person reviews it.

The system searches only the organisation’s own uploaded documents. It does not search the open web, and it cannot reach any other organisation’s material.

• Provider: Mistral AI, a company based in France. It is the only AI provider this system can reach.
• Writing model: a Mistral model chosen by the system’s administrator. The model that wrote each draft is recorded with that draft.
• Search model: mistral-embed, used to find relevant passages in the organisation’s documents
• Database and application server: a single server in Roubaix, France

The application server and the database run on the same European infrastructure, in Roubaix, France, on an account held by the Digital Training Institute. This completed on 3 September 2026. Until then the application server was hosted outside the EU and the deployment was restricted to sample material rather than real departmental content; neither restriction applies now, and an earlier version of this page said they did.

Mistral receives text every time the system uses AI. For a draft, that is the request, the material from the documents chosen (in full where it fits), the template’s checklist and outline, and any writing guidance the person has set. When a document is uploaded, its text is sent so that it can be searched.

GovComms.ai does not train any AI model. Mistral’s terms allow it to use text it receives to improve its own models unless the account holder switches that off in Mistral’s settings. Its experimental Labs models can be trained on in every case, so this system does not offer or use them. We have not yet confirmed that the setting is switched off for this system’s account. This page will say so once we have.

Every generation is written to a log that no user can alter. Users can read their own entries and administrators can read their organisation’s, but the log can only be added to, never edited or deleted, and only the server can write to it.

For each draft generated, the log records:

• Who asked for it, and when
• The template and the title
• The AI model and provider that produced it
• How much text went in and came out, and how long it took
• A fingerprint of the exact instruction sent to the model, so it can be shown later that the instruction has not been changed. Where a person has since changed or cleared their own writing guidance, the instruction can no longer be rebuilt from it: that is the cost of their being able to erase it
• Whether the person’s own writing guidance was used, never the guidance itself, which only they can read
• Every document passage that was retrieved and offered to the model, identified individually
• Whether the response was cut short

Edits and approvals are recorded too, including a marker that a human made the change.

Every draft moves through three states: Draft, then In review, then Approved. A person moves it, and each move is recorded.

The system has no way to publish anything. There are no integrations with websites, social media accounts, mailing lists or press distribution services. A draft leaves this system only when a person copies it out. That is a deliberate design decision, not a feature that has not been built yet. The one other place text goes is Mistral, to write drafts and to search documents, as set out above.

• Publish, send, or post anything
• Make a decision about any person
• Read another organisation’s documents or drafts
• Search the open web
• Train an AI model on any organisation’s material. What Mistral may do with the text it receives is set out above

AI models can state things confidently that are not correct, and this one is no exception. In testing, one draft in eighteen cited a source that did not exist to support a figure it had invented. That is why citations are checked against the retrieved sources before a draft is saved, and it is the clearest illustration of why a person reviews every draft.

The checks catch citations pointing at sources that were never retrieved. They cannot tell whether a correctly cited source actually supports the sentence. Only a reader can.

If something this system helped produce has affected you, raise it with the public body that published it. They are accountable for what they publish and can escalate it to us. A direct complaints and appeal route is being set up, and this page will name it once it is open.