The same test as for engineers, without the code. Give a marketer, a finance hire or a recruiter a real task from their job in ChatGPT or Claude, and see whether they catch it when the AI is wrong.
Why test AI skills outside engineering?
Most AI use at work is not code. In Microsoft and LinkedIn’s 2024 Work Trend Index, 75% of knowledge workers said they use AI at work, and 78% of them bring their own AI tools (Microsoft). So your next marketing or finance hire will likely do much of the job with AI, whether you test for it or not.
The risk is quiet
When AI gets code wrong, something often breaks. When it gets a budget, a contract summary or a customer email wrong, nothing does. It reads just as well, and someone acts on it.
The skill is the same
Understand the task, brief the AI, check what comes back, fix it, and hand over work that is right. Only the task and the tool change.
How an AI skills test differs for non-technical roles
Technical roles
Non-technical roles
The AI tool
Claude Code or Codex, working in the code
ChatGPT or Claude Cowork, working with documents, spreadsheets and the web
When AI is wrong
Often something breaks: a test fails, the code does not run
Usually nothing breaks. A made-up number or a wrong claim reads as well as a right one
What checking means
Running the tests
Tracing each number and claim back to its source
What they hand over
A change someone reviews
An email, a plan or a summary someone acts on, often without checking it again
The six skills
The same
The same
Four rules for a non-technical AI skills test
1Use a task from the job, with real files.Documents, spreadsheets, notes, a policy. Not a quiz about AI, and nothing that needs code.
2Let them use the AI tool they already use.ChatGPT, Claude or Claude Cowork. Forcing a new tool tests the tool, not the person.
3Say that AI is expected.People who are unsure whether AI is allowed hide how they work, or do it by hand.
4Read the conversation and the work.Score how they briefed, checked and fixed, not how polished the writing is.
AI skills tests for marketing, finance, sales, people and operations
Lemonsignal has a real task for each of these roles, done in the candidate’s own AI tool, with a first read from Lemonsignal that your team checks.
For roles we have no ready task for, these are ideas to start from: a real task, and the kind of place where AI tends to go wrong in it. Use your own files.
Role
The task
Where AI tends to go wrong
Sales
Prepare for a call with a named account, from its website, the last call notes and the CRM history.
The AI fills gaps about the account with things that sound right and are not.
Recruiting and people
Write a job ad and a screening plan from a hiring manager’s notes and the salary band.
A requirement that should not be there, or wording the AI adds that narrows who applies.
Customer support
Answer five tickets from the help centre and the refund policy.
The AI promises something the policy does not allow.
Operations
Turn a messy process doc and a month of requests into a checklist the team can follow.
A step that only appears in one request, or a rule the AI makes up to fill a gap.
Legal and compliance
Summarise what changes in a new supplier contract against the old one.
A change in one clause that the AI’s summary leaves out.
An AI skills assessment result, for a finance hire
The brief: “Check last quarter’s 400 supplier invoices against six contracts and find where you are being overbilled.” Lemonsignal read six skills from the candidate’s conversation with ChatGPT, and every level links to the turn it came from.
LemonsignalAlex Morgan
Alex MorganFinance operations analyst · Task: Find the overbilling · ChatGPT · 2 h 15 min
Checked by Lemonsignal
72%Good match, one open question
Had the AI list the contract terms before matching invoices.
Recomputed the total by hand instead of trusting the AI’s.
Left confidential details in the summary.
UnderstandSolidRead the six contracts before asking for anything.Turn 2
BriefStrongHad it list every price term from the contracts first, then match one supplier at a time.Turn 3
CheckStrongRecomputed two suppliers by hand against the contracts, and fixed the total.Turns 14, 17
CorrectSolidSaid which contract term the total was missing.Turn 15
Sign offPartialA clear summary for the CFO, with the suppliers’ bank details still in it.
ReuseSolidRan this quarter’s invoices with the same prompt.
Skills for the job
Reading contract termsAbove the bar
Reconciling accuratelyAt the bar
Handling confidential dataBelow the bar
What they handed in, as they handed it in. Each mark is a skill from their result.
Sign offPartial
To the CFOWe were overbilled €48,210 last quarter
23 of 400 invoices do not match their contract.
Three suppliers make up €45,300 of it: Nordfrakt, Brekke Pack and Solvik IT.
I suggest we ask each for a credit note. The invoice list per supplier is attached.
This quarter’s invoices get the same check before we pay them.
Questions about AI skills assessments for non-technical roles
Can you test AI skills for non-technical roles?
Yes. Give the candidate a real task from the job, such as planning a budget, reconciling invoices or answering support tickets, in ChatGPT or Claude. Then read their conversation with the AI and what they hand in. The six skills are the same as for engineers; the task changes.
Do candidates need technical skills for an AI skills assessment?
No. For non-technical roles the task is documents, spreadsheets and the web, done in a chat tool such as ChatGPT or Claude Cowork. Nothing needs code.
Which AI skills matter most for non-technical roles?
Checking, most of all. In non-technical work, nothing breaks when AI is wrong: a made-up number reads as well as a right one. Then signing off: what they send is often acted on without anyone checking it again.
Is an AI literacy quiz enough for non-technical staff?
For training, it can be a start. For hiring, it tells you what someone knows about AI, not whether they catch it when it is wrong on real work.
Does writing quality or English fluency affect the score?
It should not, and in Lemonsignal it never does. Polished writing is easy to get from AI. What matters is whether what they hand over is right and ready for its reader.
Which roles does Lemonsignal have ready tasks for?
Marketing, finance, product, design and data, as well as engineering. For other roles, bring a piece of real work and we turn it into a task with you.
