AI transparency: where we use AI
We work with AI every day. This page says which systems those are, for what work we use them, what happens to your input, how you can tell AI was involved and where a person sits in between.
Which rules apply where, before anything else
So here is the honest division. What we do is the same everywhere: a person reads what goes out, we do not train general models on client content, and we say where AI is involved. What differs is which law you can hold us to.
| Where you are | What frames AI there | What that means for this page |
|---|---|---|
| The Netherlands, Ireland and the rest of the European Union | The EU AI Act, alongside the GDPR | The commitments below sit inside that frame. Which duties apply to which application we assess case by case, because that differs per application |
| United Kingdom | No single AI act. AI is addressed through the regulators that already cover the domain in question, and through the UK GDPR for anything involving personal data | The commitments below still apply, but as our own commitments and not as compliance with a British AI statute. We make no claim of the second kind |
We say that plainly because the alternative is worse. A supplier who tells a British client that they are AI Act compliant is either confused about which country the client is in, or hoping the client will not check. Where our own software stands under that regulation, and why the role is settled per use case, is at the AI Act position for Jarvis.
Two AI pages, two directions
This page is about our own use of AI. Our AI Policy is about the opposite direction: what AI systems from outside may do with the content of this site.
# The two dots run against each other because the two directions really do. Neither one stops, which is why the figure does not either.
Beyond that, this page covers AI in our services, in our own software and in the free tools on this site. What happens to your personal data and which parties receive it is in the privacy statement. What we agree contractually is in the terms and conditions.
Which AI we use and what for
We use AI for research, summarising, draft text, classifying data, analysis of search and advertising data, the measurements for the AI Index, the free generators on this site, the agent functions in Jarvis and internal support. Per product, the interface or the documentation states what AI does there exactly, where that matters to you.
Internal support includes our own lead and client administration: that is maintained by AI sessions rather than by staff. Fixed rules go with it about who may read and change which data, and a check afterwards that measures whether those rules were followed. Which data ends up there and how long we keep it is in the privacy statement.
| System | What we use it for | What goes in | Who looks at it |
|---|---|---|---|
| ChatGPT (OpenAI) | Research, summarising, draft text, classifying data and working through search and advertising data | What we put in ourselves for the work in front of us | A person reads it over and changes it before anything goes out |
| Claude (Anthropic) | The same work, plus drafting and maintaining copy and technical documentation | What we put in ourselves for the work in front of us | A person reads it over and changes it before anything goes out |
| Gemini (Google) | The same work, mostly for questions that sit close to search behaviour | What we put in ourselves for the work in front of us | A person reads it over and changes it before anything goes out |
| The language model behind the free generators, called through Make | Writing ad copy, review replies and the brand check on this site | What you enter in that form, and your email address if you give one | Nobody in advance. What comes out goes straight to you, you are the editor |
| The agent functions in Jarvis | Working with the memory, the tasks and the connections of your organisation | What your organisation puts in itself, within the configuration approved for you | Your organisation decides which steps a person has to sign off |
| gpt-5-mini (OpenAI) inside the AI Index | Measuring which brands this model names. Here the AI is the object being measured, not the tool | The questions from the measurement design, no client data | The method and its limitations are published with the dataset |
Which model runs behind a free generator differs per tool and is recorded per tool. At this moment: the AI brand checker uses GPT-4o, the ad generator, the review reply generator and the business profile poster use GPT-4o mini, and the separate model comparison on the brand checker page uses GPT-4o mini, Gemini 2.0 Flash and Perplexity Sonar. If one of those changes, this line changes with it.
Not everything that looks like AI is a language model. The local visibility check, for instance, fetches data from a maps and search service and does its own arithmetic on it. Which parties receive data about you through the use of this site, and where that data sits, is in the privacy statement.
How you can tell AI was involved
Where the law or the situation asks for it, we make clear that you are communicating with an AI or that something was synthetically created or edited. This is how it stands per kind of output.
The question underneath all of it is the same one for you every time: do I still have to check this myself before I use it? The answer differs per kind of output, and that difference is exactly what you lose when you read six separate headings. So it is set out below first as a strip, with the place where a person sits in between marked per kind. The detail per kind follows underneath.
# Three states, and there are no more than three: a person looks beforehand, you are the first reader, or the question does not arise here.
# This strip summarises what the headings below say. If they contradict each other, the text applies and not the strip.
Text on this site
Nearly every page on theseo.nl is written with the help of AI and then read by a person and changed where needed. Nothing appears online without somebody here having checked it. That is why there is no label per page: the text is not raw model output, it is editorial work with AI as a tool.
Whether we nonetheless add a visible note, and what it would look like, has not been decided. Form and placement of the note: not yet established.
Output from the free generators
This is different. What the generators on this site give back is written directly by a language model and does not pass any of us before you see it. That is exactly the point of such a tool, but it also means you have to check the text before you use it. Names, prices, promises and anything about your own business: check it.
If you are in the United Kingdom there is a second reason to check, and it is not ours. Since 6 April 2025 fake reviews and unmarked paid reviews are a banned practice there under the Digital Markets, Competition and Consumers Act 2024, enforced by the Competition and Markets Authority. Our review reply generator writes a reply to a review you received; it does not write reviews, and it must not be used to. Source: CMA208, fake reviews guidance.
At present the results of the free tools carry no visible note that the text was written by AI and has not been checked. That note belongs there and is being added.
Images, sound and video
The imagery on this site is largely built from photographs of real people and from screen renderings drawn with code. For the older imagery it has not yet been established per file whether an image model was involved; that inventory is running. Realistic synthetic image, sound or video that could pass for real, we mark recognisably where confusion is possible, except where a statutory exception applies.
There is no complete inventory yet of which existing image files on this site were made with AI. Inventory of imagery: not yet established.
Conversations
If you end up in conversation with an AI rather than a person on this site or in one of our products, we say so. At this moment there is no chatbot on theseo.nl that could pass itself off as a person.
Scores from the AI Index
A score from the AI Index always carries the measurement date, the sample, the method, the limitations of the source and the version number. Without those five a score is meaningless. It is a snapshot with methodological limitations, not an official market standard, not a certification and not a promise about your visibility.
What AI output is and is not
- It is a probability calculation, not a fact. AI output can be wrong, out of date, incomplete or unoriginal. Including when it sounds convincing.
- You remain responsible for the use. For the context, for checking sources, for the rights in what you publish and for what you finally do with the output.
- A person judges what materially matters. Publications, advice to clients and far-reaching decisions go past a person here before they take effect.
- What we do not do. We do not use AI for hidden influence, not to give people a social score, and not to take automated decisions with legal effect on our own without a separate legal basis and appropriate safeguards.
- What you should not put in. Do not enter special category data, criminal offence data, data about children, payment card details or secrets, unless the product documentation expressly supports it and arrangements have been made about it.
We do not promise that AI improves your turnover, your positions or your visibility. Our services are obligations to make best efforts. An outcome becomes an obligation of result only where a signed engagement says so in as many words and in a measurable way.
What happens to your input
We do not use client content to train general models without separate, demonstrable consent for that. Only approved business connections may process production data, so no loose free accounts or personal subscriptions belonging to a member of staff.
Beyond that we hold to four things that we also enforce technically:
- we put in as little as possible, and strip out what is not needed to do the work;
- we remove unnecessary identifiers before anything goes to a model;
- we do not log full prompts or full output without it being written down why that is necessary;
- an external AI supplier is only added once the purpose, the data storage location, the retention period, the processing agreement and the basis for transfer have been approved.
For the work we do on a client's behalf, the client is the controller and we are the processor. For our own contracts, invoicing, fraud prevention and security logs, we are the controller ourselves.
The full overview of suppliers, processing locations and transfer mechanisms is on the sub-processors page. The parties that receive data through the use of this website are also in the privacy statement.
What you may ask for
If AI played a part in a decision that affects you, you can ask for a person to look at it after all. Email sales@theseo.nl and describe which decision it is about.
See output that is wrong, hurtful or otherwise should not be there? Report it at the same address. Say where you saw it and what you had entered, so that we can trace it.
Your rights over your personal data, such as access, correction and erasure, are in the privacy statement. If you discover a security problem, email sales@theseo.nl with the words data breach in the subject line; that route is also described in the privacy statement and on the responsible disclosure page.
How we keep this up to date
We keep an AI register. Per application it records what it is for, who owns it, which data goes in, which supplier sits behind it, what risk attaches to it, where the human check sits, how we measure whether it works and when we reassess it.
A new model or a new function only goes to production after privacy, security, rights in the content used and the obligations arising from the EU AI Act have been looked at. Which obligations those are differs per application and we assess it case by case. People here who choose, configure, assess or deploy AI have to know what they are doing; we maintain that actively and record it.
We are not certified and we do not lean on the name of a standard on this point either, because naming a standard proves nothing. What does exist is the AI register above: per application the purpose, the owner, the data, the supplier, the risk, the human check and the moment of reassessment. That is not a legal guarantee.
Does this page change?
Yes. If a supplier is added, if what we use AI for changes, or if how we mark output changes, we amend this page and raise the version number. What was here before stays available on request.
See also our AI Policy for the other direction, the privacy statement, the terms and conditions and our accessibility statement.
Version 1.0 · in force from 11 August 2026 · English edition published on 25 August 2026