The problem
Every UK public body must answer Freedom of Information requests within 20 working days, whether under the Freedom of Information Act 2000 or, in Scotland, FOISA 2002. In our conversations with officers at Scottish local authorities, the same picture came up again and again. Volumes are rising, many requests ask questions that have already been answered and published, and every one still has to be logged, read, routed and tracked against the deadline. Councils, health boards, police, universities and government bodies all face the same pressure.
Why it matters now
- Councils across the UK were estimated to receive around 467,000 FOI requests a year in 2019, almost double the 2010 estimate and roughly ten times the volume sent to central government.
- UK government departments and agencies received 94,526 FOI requests in 2025, more than double the 2020 figure. About one in eight was answered late.
- Across all Scottish public bodies, where our conversations took place, requests reached 111,241 in 2025-26, up 13% in a year and the highest ever recorded.
- Appeals to the Scottish Information Commissioner rose 83% in the same year. Over one three-month period, 34% of those appeals were found to be machine-generated.
- In May 2026 the Information Commissioner's Office issued guidance after public authorities reported a rapid rise in the volume and complexity of AI-generated requests. A request written with AI is still a valid request.
Sources: mySociety, Freedom of Information in Local Government (2019 estimate); Cabinet Office, Freedom of Information statistics: annual 2025; Scottish Information Commissioner, Annual Report 2025-26; Information Commissioner's Office, May 2026.
What we heard
We started with a different idea. Our first assumption was that public bodies would want to detect requests written with AI. Officers put us right. Their focus was compliance: meeting the deadline, giving a consistent answer, and not doing the same work twice. Whether AI helped write a request does not change the duty to answer it.
So we changed the build. That is how Bield works: start with the problem worth solving, as the people closest to it describe it.
What we built
ClearFOI is a working prototype that triages incoming requests. It:
- classifies each request by subject, so it reaches the right team first time;
- checks it against requests already answered and published, so officers can see at once whether an answer already exists;
- gives the officer a structured starting point.
It is designed to work alongside the systems and data an organisation already has, not as another standalone system to manage.
Every decision stays with the public body. ClearFOI informs the officer; the officer decides. That is our human-in-the-loop approach.
The AI indicator
Each request also carries an indicator of how likely it is to have been written with AI. It is a likelihood, not a verdict, and it changes nothing about how the request is handled. What it gives the organisation is visibility. Public bodies are now starting to ask how much of their incoming correspondence is AI-written, and the same question is being raised about complaints. ClearFOI was ready to answer it because it was built around the stable problem, not the passing one.
Beyond where it started
ClearFOI began with councils, but nothing in it is specific to local government or to Scotland. Any public body that publishes a disclosure log can use the same approach: check new requests against what it has already answered, route them faster, and see how its correspondence is changing.
It can start with one organisation and grow from there. Because published answers are already public, the same approach can work across a whole sector, such as councils, health boards and NHS trusts, or universities, without any organisation sharing private records. Each body keeps its own requests and its own decisions; what they share is what they have already published.
Data first
We built and tested ClearFOI on published disclosure data, before any organisation's own records were involved.
What it shows
ClearFOI shows the Bield approach from start to finish: listen first, agree the problem worth solving, prove the build on safe test data, and keep people in the loop at both ends. AI informs decisions; people make them.