Is your information ready for an AI knowledge service?

An AI knowledge service starts with a useful promise: help people find and understand the information they need to get something done. Preparing that information gives your team a practical place to begin, before choosing a tool or connecting a shared drive.

You do not need to reorganise every document in the organisation first. Start with one audience, one recurring question and the sources that should support a dependable answer. That creates a manageable review with a clear purpose.

Choose the question before the collection

Ask a colleague to show you a question they answer regularly. What do they consult? What do they know that is missing from the document? Where do they check an exception?

For an illustrative charity example, volunteers may need to explain how someone can access a local support service. The answer might depend on the referral route, opening hours and eligibility guidance. Start with those sources, rather than uploading the whole volunteer handbook, every meeting note and last year's service directory.

Write down who will ask the question, what a helpful answer should include and when it should direct someone to a person. A focused question makes it easier to decide which material belongs in the first trial.

Establish which version people should trust

For each source, identify its owner, intended audience and review date. Check whether the guidance is still in force and whether another document describes a different process. Keep a record of the decisions you make.

Two files with similar names can contain genuinely different instructions. Ask the responsible team to resolve the difference before treating either as authoritative. Preserve older versions through your normal records process, while making clear which version the knowledge service should use.

Recent Government Communications guidance recommends reviewing outdated and duplicate public content and monitoring how AI search presents it. Its focus is government communications and public search. The useful parallel for an internal knowledge service is to maintain a clear source of current information; it is not evidence that tidying documents guarantees accurate answers.

Keep the context with the answer

A sentence can be correct but unhelpful when separated from its conditions. Keep the service area, exceptions, effective dates and relevant definitions close to the guidance they qualify. Use meaningful headings and links that colleagues can follow.

Check whether tables, scanned pages and attachments remain understandable when processed by the proposed tool. Test an actual extract, including its headings and footnotes, rather than assuming a readable PDF will be interpreted correctly. Record where manual preparation is needed.

The government's AI-ready datasets guidance connects suitability with context, quality and ongoing stewardship. It concerns government datasets, so it provides a useful reference rather than a certification standard for a business or charity's document library.

Make access part of readiness

Separate general guidance from information about particular people or confidential operations. Record who should be able to access each source and ask how the proposed service will enforce those boundaries.

A colleague's permission to read a document does not settle whether it should be copied to another system. Review the proposed information flow, supplier arrangements and organisational requirements with the appropriate owner before connecting material.

Use synthetic examples for early trials where possible. Include a test in which a user asks for information outside their permissions. If the access arrangements are unclear, narrow the trial to material approved for that audience while resolving the wider questions.

Test answers against the source

Build a small set of representative questions. Include a straightforward request, an exception, an ambiguous question and one the approved material cannot answer. Agree what the service should do in each case.

Ask reviewers to compare answers with the sources, check that links open for the intended user and identify missing qualifications. Where the information is insufficient, a useful response may ask a question or direct the user to the responsible team.

Try updating a source and checking the answer again. This helps reveal whether the update process works and whether the old version remains available in a way that could cause confusion. Keep the findings visible so a promising demonstration becomes a clearer decision about what to improve next.

Give the information a life after launch

Name who will maintain each source and who will investigate a reported problem. Agree how changes reach the knowledge service, how quickly they need checking and how an unreliable source can be withdrawn from use.

Make those responsibilities part of the service team's ordinary work. A useful starting output is a short source register, an agreed set of questions and a list of issues with owners. That gives you something concrete to review before expanding the trial.

Bring one recurring question and the documents your team currently uses to answer it. Our data and systems readiness work can help turn those into a practical starting point for AI Readiness, with a focused route towards a chat and knowledge service.