Panels are beginning to use tools like ChatGPT to shape their questions. Not lazily, but deliberately. The phrasing becomes tighter. The competencies clearer. The bias, at least in theory, reduced.
Then the answers arrive.
Measured. Structured. Calm.
And someone, somewhere, pauses.
“Do we think this was written by AI?”
It is a reasonable question. It is also the wrong one.
A room we might recognise
Picture a panel. Not a specific one, but a familiar one.
There is quiet diligence in the room. People reading closely, comparing carefully. They are not hunting for perfect prose. They are trying to answer something more difficult:
Who has actually carried responsibility when things were uncertain?
Who has made decisions that changed outcomes?
Who understands what happens when systems, people and timing do not quite align?
The writing, though, is good. Consistently good.
And that consistency introduces doubt.
Not about honesty, but about substance.
When the signal drifts
The moment “AI” enters the conversation, even silently, attention shifts.
From:
what the candidate did
To:
how the answer feels
It is subtle, but consequential.
Because the task of a panel is not literary critique. It is interpretation. Reading between lines, looking for evidence of judgement.
You can feel when that starts to slip. Answers that are technically correct, but somehow leave no trace of a decision being made.
The thing we are actually searching for
Strip it back and the question is simple:
What changed because this person was there?
That is rarely found in tidy summaries.
It appears in edges.
The moment something nearly went wrong
The point where assumptions stopped holding
The decision that cut across a comfortable plan
Those are not always elegant to describe. They are often slightly awkward. Occasionally contradictory.
And that is precisely why they matter.
The smoothing effect
AI-assisted writing does something quite helpful. It brings order. It removes repetition. It makes things readable.
But in doing so, it can smooth away the very texture panels are looking for.
Not deliberately. Just as a consequence of optimisation.
You end up with answers that say:
“I worked with stakeholders to ensure successful delivery”
When what the panel needs to see is:
where the friction was
what the risk looked like
and who chose to act
Without that, the answer floats.
A familiar pattern
There is a pattern I keep coming back to, though I rarely name it directly.
First, something changes.
Then, people try to make sense of it.
Then, they act.
It sounds obvious when written like that. Almost trivial.
Yet most answers skip the middle step.
They describe the change. They describe the action. But the interpretation, the moment of judgement, is missing.
That is where seniority lives.
The quiet test
Panels develop a kind of instinctive test.
They read a paragraph and ask themselves:
“Could I probe this?”
If the answer is yes, there is something there. A thread to pull.
If the answer is no, the answer may be too general, however well written it is.
This is why slightly uneven answers often outperform perfectly polished ones. Not because they are better written, but because they reveal where something real happened.
A wider echo
This is not just about recruitment.
It touches a theme that keeps resurfacing across different conversations about AI.
We are becoming very good at producing answers. Clean, structured, plausible answers.
But the harder task remains:
deciding what is worth asking in the first place
and then recognising when an answer actually contains something meaningful.
In other words, the machine helps us see more clearly. It does not decide what matters.
The paradox, again
So we find ourselves in a curious position.
We use AI to:
shape better questions
create consistency
remove some of the noise
And candidates use it to:
express themselves more clearly
structure their experience
present coherent arguments
Yet neither side has solved the central problem.
Because the central problem was never writing.
It was always interpretation.
A lighter ending
If our hypothetical panel did try to formally “detect AI”, they would quickly run into difficulty.
They would have to explain:
how they know
why it matters
and how it relates to the role
That is not a comfortable place to be.
It is far easier, and far more useful, to return to something simpler.
What did this person notice that others didn’t?
Where did they challenge the direction of travel?
What changed because they chose to act?
If those answers are present, it does not matter how the words were assembled.
If they are not, no amount of polish will compensate.
And that, for now, still feels like a human judgement.
Comments
Post a Comment