AI transparency
Last updated : 15 September 2026
Onbord uses artificial intelligence at several points. This page says which ones, what the machine decides, what it does not decide, and what you can require if an assessment strikes you as unfair.
Working draft. The text below describes the service accurately, but it has not yet been reviewed by a lawyer and some details are still missing.
1. Where AI is involved
Reading the job posting: a language model extracts the expected skills, the level required and the context of the role.
Building the assessment: the same kind of model writes the scenarios, the instructions and the scoring criteria that follow from it.
Correction: candidate answers are compared against those criteria and produce a score with written reasoning, tied to specific extracts of the answer.
2. What AI does not decide
No candidate is ever automatically rejected by Onbord. Onbord only builds a shortlist: the recruiter always makes the final call. The assessment produces the score and the supporting evidence behind that decision, nothing more.
Scoring criteria are reviewed and approved by the recruiter before the first candidate takes the assessment. An assessment never goes live on the model's choices alone.
3. Candidates using AI
Simulations allow the use of AI tools, because real work allows them. Trying to ban them would mean assessing a situation that no longer exists.
How well those tools are used is part of what is assessed: knowing what to ask, checking what comes back and fixing what is wrong is a professional skill, treated as one.
4. Known limits
A language model can get things wrong: misread an answer written in an unusual style, mistranscribe an audio recording, or apply a criterion too literally.
That is exactly why every score comes with its reasoning and the extract it rests on: a score you cannot trace back to its source should never weigh on a decision.
5. Requesting human review
Any candidate may ask for a person to review their assessment, obtain the reasons behind the score, and put their point of view.
The request goes to the company running the hiring process, or to hello@onbord.be, which will pass it on.
6. Model providers
The models used are Anthropic's (the Claude family), called through their business interface.
Content submitted is not used to train those models. See the privacy policy for the full list of processors.