How we use AI
Our AI writes an opinion about a CV - a score, a summary, strengths and concerns, an expected salary, questions worth asking. It never advances, rejects or ranks anyone into or out of a process on its own. Every decision in this system is taken by a person, and recorded with their name against it.
| Where | What it produces |
|---|---|
| Reading a CV | A score out of 100 with reasoning, a two-line summary, up to five strengths and five concerns, a seniority level, a job-stability read with an explanation, extracted skills, and an expected salary range in the candidate's own market |
| Against a specific role | A match score with a written recommendation, so a long list can be ordered - by a person who can see why |
| Before an interview | Question sets drawn from that person's actual CV, each one referring to a real company, project, technology or gap |
| Writing to a candidate | A drafted outreach email for the specific role, which you edit and send |
| Opening a role | A drafted job description from a title and a few details |
All five produce text you can read, edit or ignore. None of them changes a candidate's stage, sends anything on its own, or hides anyone from you.
A score is a starting point for a human judgment, and it's shown with its reasoning so you can disagree with it. In our experience the concerns list changes more minds than the number does.
Every workspace runs on the same platform AI service (Google Gemini). You do not choose a provider, you do not paste a key, and nothing in Account settings turns this on or off.
Nothing is sent anywhere until you ask for it. There is no background analysis. Each of the five features runs when a person clicks it.
| When you run | What is sent to Google Gemini |
|---|---|
| A CV analysis | The text extracted from that CV, and - if you've asked for a role-specific score - that role's title, requirements and key skills |
| Interview questions | The same CV text and role details |
| A drafted email | The candidate's name, the role, and the context you're writing about |
| A job description | Only the details you typed in. No candidate data at all. |
Never sent: salary records, bank details, payslips, contracts, identity documents, medical or leave records, client commercial terms, or anything belonging to another candidate.
Results are stored on the candidate's record in your workspace, and they're covered by the same export and deletion rights as everything else.
If your policy does not allow candidate data to leave your environment, the AI features are not a fit. Everything else - hiring, onboarding, contracts, payroll, invoicing, offboarding - carries on as normal.
Some customers can't send candidate data to a third party, or won't until their own review is done. That's a supported way to use this product, not a degraded one.
Does the AI reject candidates?
No. It has no ability to. There's no automatic filter and no setting that creates one - every advance and every rejection is made by a person and recorded against their name.
How accurate are the scores?
A score is a reading and not a measurement, and it will sometimes be wrong, which is why we always show it alongside the reasoning, the concerns and the strengths. Treat it as a well-briefed colleague's first impression: useful for ordering your attention, not for making the call.
Is our candidate data used for training?
Not by us, ever. CV text is sent to Google Gemini to produce the analysis, under our agreement with Google, and is not used to train a model.
Can we see what was sent?
Yes. Each analysis is stored on the candidate's record with the model that produced it, and requests are logged.
We're required to audit tools used in hiring decisions. Where do we stand?
Our position is that this is a decision-support tool, not a decision-making one: it produces written recommendations, takes no action, and every outcome is attributed to a named person. What your obligations are depends on where you hire, and we'd rather help you answer that properly than tell you it doesn't apply. Ask us and we'll give you specifics about how the features work.
What does it cost?
AI is included on every plan. Token use is recorded per workspace so we can see what it costs to run. Going over a normal volume never turns the features off.