Responsible AI Policy

AI is not used everywhere in Oryon Data. Collecting data, computing metrics and producing charts are entirely deterministic: they run on code, not on a language model.

AI is involved where the work is writing or interpreting: the analyses and recommendations you request, daily summaries, the conversational assistant, written reports, analysis of your website's content, and checks on how visible your brand is.

Those features are named as AI features in the product — Oryon AI Insight, AI visibility, the Oryon assistant — and the list above is kept current in this policy.

The business data you connect to Oryon Data is sent to one artificial intelligence provider only: OpenAI. It produces the analyses, summaries, recommendations and reports you ask for.

One feature involves other providers: the checks on how visible your brand is inside AI assistants, which query OpenAI, Google Gemini and Perplexity to see whether your company comes up in their answers. Those checks send none of your connected data and no personal information: they send only a market question generated from your industry, your city and your areas of expertise — along the lines of "Which company in industry X in Y can help me with Z?" — and then check whether your name appears in the answer.

Put simply: your data goes to OpenAI; the visibility checks ask three assistants a public question.

Your data is not used to train, fine-tune or otherwise improve artificial intelligence models, whether generalized or personalized. Our use of our providers' APIs is governed by terms that exclude this.

One customer's data is never used to produce an analysis for another customer. Nothing is pooled across organizations.

Google user data and Meta Platform data are excluded from any such use, in accordance with the Google API Services User Data Policy and the Meta Platform Terms.

Oryon Data makes no decision about an individual based exclusively on automated processing. The platform analyses aggregated business data — traffic, campaigns, sales, visibility — and performs no profiling, assessment or scoring of individuals. The disclosure obligation in section 12.1 of Québec's Law 25 concerns automated decisions about a person; Oryon makes none.

Oryon's AI also takes no action inside your accounts. The platform's AI components are recorded in an internal registry, and not one of them has a write capability: they read data and draft recommendations you are free to apply or ignore. The only write the platform performs is exporting a report to Google Slides, which you request explicitly.

The recommendations the platform produces are proposals for your judgement. You decide what gets acted on, when and how.

We design AI features to explain their reasoning and point back to the data they rest on, so you can check them rather than take them on faith.

Language models can be wrong, can state something plausible but false, and can misread a context. We do not claim otherwise.

That is why the figures shown do not come from the model. When a recommendation cites a metric, it is anchored to that metric and the series of values is read directly from your data by the platform: the prose is written by the AI, the numbers are computed by code. A model therefore cannot invent a measurement.

The analyses provided are neither human professional advice nor a guarantee of performance or future results. Any decision taken on the basis of them remains your responsibility.

Language models reproduce biases present in the data they were trained on. In Oryon Data's context, that risk falls mainly on how recommendations are phrased, and on how some industries, markets and languages are served better than others.

We limit that exposure by restricting AI to interpreting your own data, by keeping it away from any assessment of individuals, and by treating French and English as two first-class languages of generation rather than translating from one.

If a recommendation strikes you as unfair, inaccurate or inappropriate, report it to hello@oryondata.com: those reports drive our fixes.

The analyses, summaries and reports produced for your organization belong to you. You can export, edit and distribute them as you see fit.

When you publish content Oryon Data produced outside your organization, we encourage you to read it first: it was drafted by a language model from your data, and by publishing it you become its author.

Data sent to an artificial intelligence provider to produce a result may be retained temporarily by that provider, for a period of up to thirty (30) days, for security and abuse-prevention purposes, and is then deleted in accordance with its applicable policies.

No zero-retention arrangement is in place with our providers today. We would rather write that down than imply otherwise. Organizations with specific requirements here can write to privacy@oryondata.com.

Some features analyse content from the web — pages on your site, reviews, search results. That content is written by third parties and can contain text designed to divert a language model from its task.

The platform therefore wraps all external content in inert data tags, together with an explicit instruction telling the model that the content is data and never an instruction. Sequences that would close those tags early are stripped repeatedly, until none remain, to defeat recombination attempts.

Every AI component in the platform is recorded in an internal registry stating what it does, which model it uses and which operations it performs. An AI feature missing from that registry fails our automated tests, so it cannot reach production undocumented.

Each run is logged with its steps, its consumption and its result, which lets us reconstruct what was produced, by which component and from which data whenever an answer is disputed.

Oryon Data's AI features evolve, and this policy evolves with them. Any updated version is published on this page and takes effect when it goes live; the update date appears at the top of the document.

For any question about our use of artificial intelligence, about exercising your rights, or about the processing of your personal information: Privacy Officer — Élody Sanchis, Co-founder, Oryon Data Intelligence Corp. — privacy@oryondata.com. General enquiries: hello@oryondata.com.