AI use

I use AI tools. I do not delegate judgement, authorship, or responsibility to them.

AI can be useful for exploring a question, testing an argument, organising material, and improving a draft. It can also introduce confident errors, flatten important distinctions, obscure the origin of a claim, and expose information that should not have been shared.

This page explains how I use these tools through Proper Tools and the constraints that apply.

A tool may assist the work. It does not become responsible for the work.

What AI may be used for

Depending on the work and the information involved, I may use AI tools to:

• organise notes, source material, or competing accounts;
• produce an initial structure or working draft;
• explore alternative interpretations and framings;
• identify possible objections, omissions, or inconsistencies;
• assist with editing, summarisation, translation, or coding;
• generate questions for further research or discussion; and
• test whether an explanation is understandable to its intended audience.

AI output is treated as working material. It is reviewed, revised, and tested against the relevant evidence before it becomes part of advice, research, or published work.

A model response is not treated as a source merely because it is fluent, detailed, or correct-sounding.

What remains human

I remain responsible for the claims, omissions, recommendations, and consequences of work delivered through Proper Tools.

• I decide which evidence is relevant and how much weight it can carry.
• I distinguish sourced fact from inference, judgement, and speculation.
• I verify consequential claims against appropriate sources.
• I decide what advice to give and state the assumptions on which it depends.
• I review the final work in context rather than approving isolated passages.

AI systems do not decide what a client should do. They do not approve conclusions, accept risk, or speak on behalf of Proper Tools.

Known limitations

AI systems can produce invented facts and citations, omit relevant context, reproduce errors from their inputs, and sound more certain than the evidence permits.

They may also:

• misread ambiguous instructions or documents;
• lose important information as context grows;
• produce plausible but unsupported explanations;
• reflect biases present in training data or supplied material;
• converge too readily on the framing implicit in a prompt; and
• give inconsistent answers to substantially similar questions.

These are not edge cases to be solved by adding a warning label after the fact. They shape which tasks I use AI for, what information I provide, and how the resulting work is checked.

When AI is not used

AI use is not automatic. Some work is better done without it.

I will not use an external AI service for client material where:

• the engagement prohibits it;
• the client has asked that it not be used;
• the information is too sensitive for the available controls;
• applicable law, regulation, policy, or contract does not permit it;
• the task requires an evidentiary record the tool cannot support; or
• using the tool would add risk without adding meaningful value.

Traditional research, analysis, writing, and facilitation remain ordinary parts of the practice. AI is one available tool, not the default explanation for how work gets done.

Client information

Client information is handled according to the engagement, the sensitivity of the material, and the controls available in the tool being considered.

The governing principle is data minimisation: do not provide an analytical tool with information it does not need.

• Identifying details are omitted, generalised, or replaced where the task does not require them.
• Material from different clients is kept separate.
• Personal or casual-use accounts are not used for confidential client work.
• Sensitive material is not submitted to a third-party AI service merely because doing so would be convenient.
• Access, history, retention, and training settings are reviewed before a service is used for client work.

Where stronger isolation is required, the work may be performed without AI, with a locally controlled system, or within another environment agreed with the client.

No technical arrangement makes disclosure risk disappear. The first control is deciding not to disclose information the task does not require.

External AI services

When Proper Tools uses an external AI service for client work, the service and account configuration are assessed against the needs of that work.

Relevant considerations include:

• whether submitted material is used to train or improve models;
• what conversation history and operational metadata are retained;
• which people or administrators may access the workspace;
• where data is processed and stored;
• what contractual and technical controls are available; and
• how data can be exported or deleted when the work ends.

I do not knowingly opt client material into model training or product-development programmes.

Vendor systems and terms change. Controls are therefore reviewed as part of the work rather than assumed to remain true indefinitely.

Records and transparency

AI involvement should be documented in proportion to its importance to the work.

Where useful or agreed, Proper Tools can provide a curated record of matters such as:

• which tools were used and for what purpose;
• representative prompts or instructions;
• sources supplied to or retrieved through the tool;
• material outputs that materially affected the work;
• checks, corrections, and human decisions applied afterward; and
• known limitations relevant to the result.

This is an account of the workflow and evidence, not a claim to reproduce a model’s private or hidden reasoning.

Full raw interaction logs are not retained or supplied by default. Where a particular record is required for audit, assurance, or regulatory purposes, its scope and retention should be agreed before the work begins.

Retention

AI working material is kept only while it remains needed for the work, its review, or an agreed record of how a conclusion was reached.

Retention may differ where the engagement, applicable law, an audit requirement, or a documented client need requires a longer record.

External services may retain conversation content, metadata, safety records, or backups after material has been deleted from the visible account. Their documented retention practices form part of the assessment described above.

Further information about personal data and security is available in the privacy notice and on the security page.

Client choice

Clients may ask how AI is expected to be used in their work, request particular constraints, or ask that specified material not be processed through AI tools.

Where a requested restriction materially changes the scope, timing, method, or cost of the work, I will explain that before proceeding.

Questions about a current or proposed engagement can be raised through the ordinary contact route.

Changes to this page

Material changes to this page will be recorded here.

• 4 August 2026 — Substantially revised to clarify the purposes for which AI may be used, the work that remains human, client choice, data handling, vendor assessment, documentation, and retention.

Last updated: 4 August 2026