CITIES GLOBALLY UNITED / 22
How we use AI
We use AI openly, as a tool. It helps us read, check and draft; people decide, and every claim shows how it was verified.
What AI does here
- Creates the images of possible urban futures. They are artistic impressions, labelled “AI-generated”, not photographs and not forecasts.
- Helps monitor sources: once a month an AI agent re-reads the pages behind the promise registry and the signals and drafts suggested changes.
- Helps draft and translate texts into six languages. Translations still need review by native speakers.
- Helps prepare factual letters to city offices, which are sent from the project mailbox after address and duplicate checks.
What AI does not do
- It does not decide what is published: nothing from the monitoring goes live without approval by the project owner.
- It does not rank or score cities, and it does not turn plans into results.
- It does not invent figures: an entry carries a number only if it was read on the source page; otherwise it is marked as headline level.
- It does not publish what visitors send in without their agreement.
Human oversight
The project owner approves changes to the registry, signals and dossiers. Each change is recorded in the edition log with its date.
Is a generative model used?
Not on this website today. The assistant on the page “Ask the registry” is an automatic search over the texts of this edition; it does not generate sentences. If a generative model is added later, it will be used only to rephrase existing entries, its answers will carry a visible AI label and a link to the entry they came from, and a person will keep the final say on anything that is published.
Known limits
- AI can misread a page or a translation. If you spot an error, tell us and we will correct it and note it in the log.
- Some official sites cannot be read automatically. Those entries are marked “headline level: check the full text”.
- The images of possible futures are not evidence about any real place or plan.
Your messages
Messages sent through the feedback form are read by the project operator, who may use an AI assistant to help read and sort them. We do not train models on messages, and nothing you send is published without your agreement.
Found an error?
Write to us through the feedback form. A short note with the page and what is wrong is enough.
Open the feedback formUpdated: 2026-10-07. Sources retain their original publication dates.