Artificial intelligence

Last updated 15 August 2026

Our whole argument is that a diamond can be measured rather than described. That argument would be worthless if the numbers came out of a model that guesses. So the most useful thing this notice can tell you is where artificial intelligence is not involved, and then be straight about where it is.

Nothing we publish about a diamond is generated by AI

The simulator, the comparison tool and every figure in the Cut Registry are computed by a deterministic physical model. Light is traced through the facet geometry according to the optics of the material. There is no neural network, no training data, and nothing learned from other people's results.

The practical consequence is reproducibility. The same cut, at the same refractive index, under the same protocol, returns the same numbers today, next year and on your machine rather than ours. That is what makes the registry checkable, and being checkable is the entire point of publishing it.

If you ever see a diamond figure from us that you cannot reproduce from the published geometry and the stated method, treat it as an error and tell us.

Where we do use AI

We are a small company, and like most small companies we use general-purpose AI tools as assistants. Being specific about that is more useful than a vague reassurance.

Marketing imagerySome campaign and lifestyle visuals are AI-generated or AI-assisted. These are labelled. See the rule below on product photography.

Written copyDrafting and editing assistance for articles, captions and descriptions. A person decides what is published and is answerable for it.

Video and social contentEditing, framing and generation assistance for short-form content.

Internal workResearch, code assistance and document drafting, including parts of these notices.

What AI is never used for:

Producing, adjusting, smoothing or estimating any published measurement. Not the registry, not the simulator, not a grade, not a comparison.
Making any decision about a person: who we sell to, at what price, or on what terms.
Assessing a customer, a supplier or an applicant.
Answering you while pretending to be a human being.

How we label what AI makes

Since August 2026 the EU AI Act requires people to be told when content is artificially generated. We think the honest version of that rule goes a little further than the letter of it, so:

An AI-generated image is never presented as a photograph of the actual item you will receive. For a jeweller this is the rule that matters most. A generated visual can show a style, a mood or a setting; it cannot stand in for the stone in the box.
AI-generated or AI-assisted imagery is labelled as such where it could otherwise be mistaken for a photograph.
Renders from our own simulator are labelled as renders. They are computed images of a modelled stone, which is neither photography nor AI.
We publish no synthetic imagery or audio of a real person.
If we ever put a chatbot on the site, it will say it is a machine on first contact.

This standard applies to every listing we publish or update from 15 August 2026. Listings that predate it are being worked through as we reach them, and where an image misrepresents the item it is replaced rather than relabelled. We would rather tell you where the line falls than imply it had always been there.

Our position under the AI Act

Regulation (EU) 2024/1689 sorts AI systems by risk. Ours is the simple end of it.

Prohibited practicesWe operate none of them. No social scoring, no emotion recognition, no biometric categorisation, no untargeted scraping of faces, no manipulation of vulnerable people.

High-risk systemsNone. We do not use AI in employment, credit, education, essential services, biometrics or law enforcement.

Transparency obligationsThese apply to us as a user of generative tools, and are what the labelling rules above are for.

General-purpose AI modelsWe build none. We use third-party tools as a customer.

We are a deployer of AI, not a provider of it. We are not required to register anything, and we make sure that whoever uses these tools here understands what they can and cannot do, because a tool used without understanding is where the harm actually comes from.

A person is always accountable

Nothing reaches the public because software produced it. Every published image, sentence and figure passes a human who is answerable for it, and "the model wrote it" is not an excuse we will ever offer you. If something we published is wrong, it is wrong because we published it.

The screening panel on our testing page

Our lab versus natural page includes a panel showing how an automated screening instrument classifies a stone. Real instruments of that kind do use machine learning. Ours does not. The panel is a scripted illustration of the process, it is not reading a real stone, and it produces no origin determination. The page says so directly beneath the panel, and that caveat stays there.

Your data and AI

We do not put customer personal data into public AI tools, and we do not use your data to train models, ours or anyone else's. How we handle personal data generally is in the privacy policy.

Telling us we got it wrong

If you believe something we published was AI-generated and not labelled, or that a figure cannot be reproduced, write to [email protected]. We would rather correct it than defend it.

What is changing

The AI Act arrives in stages. The prohibitions on unacceptable practices have applied since February 2025, the rules for general-purpose AI models since August 2025, and the transparency rules and the enforcement regime from 2 August 2026. Obligations for high-risk systems follow on 2 December 2027. Nothing we currently operate falls into the high-risk tier, and we will update this notice if that changes.

This notice describes our own practice and is published for transparency. It is not legal advice. Related: Intellectual property covers what we allow AI assistants to read and cite from these pages, which we permit deliberately.