Our commitment
Consonia was built to support recruiters in seeing candidates with greater attention — not to replace human judgement, not to score, not to filter. These are the principles that guide every decision about how this tool works, what it asks of its users, and what it refuses to do.
01
Consonia generates evidence — not decisions. Every reading produced by the system is a piece of structured attention, traced back to a verbatim passage from a real interview. No score. No ranking. No automatic shortlist. The recruiter remains the only person who decides who is hired and who is not. This is not a stance against AI. It is a stance against pretending AI can carry a responsibility that belongs to a person.
02
Every interview contains a person who placed their trust in the recruiter. That person — often not present in the room where the evaluation happens — deserves protection. We require that candidates be informed, in advance, that AI is used in their evaluation process. We provide a model notice you can use. We strongly recommend that interview transcripts be anonymised before upload. The candidate's dignity is not a footnote.
03
We do not use uploaded transcripts to train AI models — neither ours nor third parties'. Audio submitted for transcription is processed in real time and not stored beyond the duration of the API call. Text transcripts and AI-generated readings live in your account, encrypted, accessible only to you. If any of this changes, we will tell you before it takes effect.
04
Many tools that claim to optimise for "culture fit" are actually optimising for sameness — quietly filtering out the candidates who don't look or talk like the existing team. Consonia is built to do the opposite. Every parameter in the cultural grid can be marked as either an alignment requirement (where fit matters) or a complementarity opportunity (where difference is a gift). The recruiter decides which is which. The tool does not assume sameness is the goal.
05
An artificial intelligence system cannot perceive a candidate's silence, cannot feel the weight of context, cannot know what a particular hesitation means. It can recognise patterns of language and value across a transcript — and it does so with rigour. But an interview is more than a transcript, and a person is more than the words they choose under pressure. Using Consonia with awareness means knowing what it can offer and where it ends. We do not pretend otherwise.
How the engine works
The EU AI Act gives a candidate the right to a meaningful explanation of how AI contributed to their assessment. This is ours, in plain language. The engine performs two operations — it extracts the grid from the team, then reads candidates against that grid — and neither is a verdict. Here is what happens, step by step.
The AI reads the team interview transcripts, together with the context and job description you provided. It does not apply a predefined list of "right" traits: it surfaces the parameters that recur across these conversations — the values, language, and tensions the team actually expresses. Every proposed parameter is anchored to a verbatim passage: you can see exactly which words, in which interview, produced it. And nothing is final until you validate it: you edit, remove, rename, and tag each parameter as alignment or complementarity. The grid the tool uses is the grid you approved.
For each parameter in your grid, the AI re-reads the candidate's transcript looking for evidence of consonance. It returns one of three states — evident, partial, not observed — never a number, never a rank. Every state is backed by the exact passage, in the candidate's own words, that supports it. If that passage isn't there, the state is "not observed": the tool does not infer beyond what was said. Every reading is traceable to its source — and you are free to disagree.
How we stay honest
Consonia's grid extraction and candidate readings are periodically audited for systematic bias. We sample anonymised cases across team contexts and demographic profiles, comparing AI output against independent human readings. Where systematic divergences emerge — particularly along protected characteristics — the underlying model and prompts are refined. Every revision is documented and versioned. This is not a marketing claim. It is a process.
Resource for recruiters
Required by EU AI Act — to be sent before the interview
As part of our hiring process, we use AI-assisted tools to support our evaluation of cultural and systemic fit between candidates and the team you would join. The AI processes the transcript of your interview to identify themes that resonate — or do not — with what we have learned about the team's values, language, and ways of working. The AI does not score, rank, or decide. Final decisions are made by our recruiting and hiring team, who review the AI's readings as one input among others. You may request, at any time, a human-only review of your candidacy without AI involvement, and you may ask to see the readings produced about your interview.
For your own internal documentation
The recruiter commits to processing candidate interview content only for the purposes of evaluating the specific role for which the candidate has applied, and only with the candidate's prior informed consent. The recruiter retains final decision authority on all hiring outcomes. AI-generated readings are used as supporting evidence, never as determinative. The recruiter assumes full responsibility for compliance with applicable data protection law (GDPR), employment law, and the EU AI Act — including obligations of transparency, human oversight, and non-discrimination.
Both texts are templates. They are intended to be adapted to your specific recruiting context and reviewed by a legal professional before use. Sending the candidate notice in advance is a legal obligation under the EU AI Act, not a courtesy.
Copied