What candidates do
Real work, not an exam, set for the kind of role you are filling (office and administration, sales, customer service, marketing, finance, operations, managing people, or general). Part 1: write a short piece of work from a brief using an AI assistant built into the page, as they would at work. Part 2: tick the mistakes in another AI draft, and answer four quick questions about data and judgement. About 10 minutes, no time limit, nothing to install, no AI account needed.
What you get
A one-page report: a level and a mark in five areas, each backed by what the candidate actually did; strengths and gaps; three to five questions to ask at interview; how long they took and how they used the assistant; and their full conversation, finished version and answers.
How marking works
AI marks how they worked with the assistant, and a mark only counts when their own messages back it up. Plain rules check the facts in their finished version against the brief, the sentences they ticked, and their four answers. Every candidate for a kind of role gets the same task. A job description, if you add one, only tailors the notes and interview questions: it never changes the tasks or the marks.
Fair use
There is no pass or fail and no automatic rejection: a person decides. Candidates see a notice before they start explaining what you receive and how it is used, can ask you for adjustments by replying to their invitation, and have no time limit. Timing information is a signal, not proof. Results are a guide, not a qualification. We recommend you note your use of the check in your data protection impact assessment.
Data
You are the controller and we are your processor. We keep each candidate's email address, their conversation with the assistant and other answers, timing, and the report, and delete them 180 days after the invitation (or as soon as you delete them). Nothing is used to train AI. We use OVHcloud (hosting), Anthropic (the assistant and AI marking), our email provider, and Stripe (your payments only).
Price
£25 per candidate, or 10 for £200, 50 for £750. Paid once by card. A check is held when you invite someone and comes back if they decline, do not start within 14 days, or their answers cannot be marked. Prices exclude VAT while we are not VAT registered.
Start free with your own check, then invite candidates from your page.
The standard
Every check, for teams and for candidates, is marked 0 to 4 in five areas. Levels: under 1.5 on average is Starting, under 2.5 Developing, under 3.3 Capable, and above that Strong.
- Asking AI clearly: gives the AI assistant the facts it needs, says who the work is for and the tone, length or format, and improves the draft with clear follow-up (pointing out mistakes, asking for changes) or tells it to stick to the brief.
- Checking what AI gives you: spots the sentences in an AI draft that are wrong or not supported by the brief: wrong dates and figures, promises the brief does not make, and invented claims, without flagging correct ones. Scored by mistakes found, less correct sentences flagged.
- Keeping data safe: keeps personal and confidential information out of AI tools the organisation has not approved, and still finds a safe way to get the job done. Scored from two short scenario questions.
- Knowing when to use AI, and when not to: keeps decisions about people, and anything with legal or financial consequences, with a person who checks and can explain them. Scored from two short scenario questions.
- Getting a good result: ends up with a finished piece of work that has every fact from the brief right, adds nothing invented and suits its audience.