How we
use AI

We design AI-enabled products and systems for our clients. We also use AI in our own work. These technologies can speed up research, reveal patterns across large amounts of information and make it easier to explore several directions. They do not replace context, judgement or personal accountability.

This page explains how we approach that work.

What AI helps us do

Depending on the project and the tools approved for it, we may use AI to organise research signals, summarise documents, analyse data or code, explore structures and angles, assist with testing, or develop internal tools.

Output from an AI tool is not treated as ready to publish or deliver. The level of human review depends on the risk, context and intended use.

What people remain responsible for

A person frames the question, selects the method and relevant sources, interprets the results and decides what is worth keeping. They check and correct the work before it is published or delivered, with expertise appropriate to the subject.

That responsibility extends to the final form of the work: its point of view, recommendations, tone, design decisions and fit with the client's context. A plausible answer is not necessarily an accurate one.

Sources, accuracy and corrections

AI can invent a reference, flatten an important nuance or present old information as current. When a material factual claim supports public content, we aim to connect it to an accessible source, check its scope and date, and make relevant limitations clear.

Human review reduces the risk of error but cannot remove it. If we identify a material error in published content, we review it and correct it where appropriate.

Client and personal data

A convenient tool is not automatically a suitable environment for confidential information. We do not put confidential client data into a public consumer AI service without appropriate authorisation and safeguards. Any use depends on factors such as the agreed purpose, data minimisation, the provider's terms, access controls and the relevant contractual framework.

For details of how we handle personal data, read our privacy policy.

Rights, originality and confidentiality

AI assistance does not remove our responsibility. We must still respect copyright, licences, permissions for brands and supplied material, and the confidentiality commitments we make to clients.

We assess material suggested by AI tools before reusing it. Where its origin, usage rights or degree of originality is uncertain, the right decision may be to find another source, recreate the element or leave it out.

When we disclose the use of AI

We make it clear when someone is interacting with an AI system where applicable rules require this. We also disclose generated or manipulated material when withholding that information could reasonably mislead the audience, or when a specific transparency obligation applies.

This does not mean that every text prepared with AI assistance carries an individual label. The context, the nature of the material, meaningful human editorial control and the applicable rules all matter.

Limits are part of the method

Tools, models and their outputs change. They can reproduce bias, miss essential context or fail in unpredictable ways. We therefore adjust our checks to the potential consequences of an error and bring in additional expertise when the subject calls for it.

Questions or corrections

You can contact izzy to ask about this approach or flag something you think we should review.

This page describes our working practices. It is not legal advice or a compliance certification, and it does not replace our privacy policy or the terms agreed with a client.