If the writing test stated Do Not Use AI ...

[Anonymous]

... and half the candidates used AI anyway (according to three different checkers), how would you approach this? We have a very strong set of candidates, for a writing-heavy role, so it's no issue for us to decline those candidates who used AI on the writing test. But should I tell them that's why they won't be moving forward in the process? I feel like they should know they're getting caught and not do it again.

[Anonymous]

Checkers can still be wrong--and I love the em dash.

[Anonymous]

If they’re otherwise strong candidates, I think you should ask them if they used it and see how they respond. I don’t trust the checkers.

[Anonymous]

How confidant are you in your AI detection? I have seen lots of reports about how inaccurate they can be in the educational setting.

[Anonymous]

I wouldn’t eliminate them based on the checkers. Did the writing sound like AI?

If the writing was good and fit the assignment and didn’t sound like AI, does it actually matter if they used AI?

[Anonymous]

I also think you open yourself up to issues if you’re wrong.

[Anonymous]

If you're confident they used AI, don't hire them, but I am really skeptical of those AI checkers. I think you need a better system other than the honor system and AI checkers. Can you do an in-person written exam on a company laptop (internet disabled)?

You should check with HR, but we're always told to never give specifics as to why we don't hire someone. "We opted for another candidate" is basically all we say.

[Anonymous]

Anonymous wrote:Checkers can still be wrong--and I love the em dash.

I love an em dash, but I don't space like AI does.

[Anonymous]

Is the issue that it's LLM-generated or that it's generic slop? If it's generic slop, "you got caught" doesn't matter; the writing quality wasn't what you were looking for, and presumably you wouldn't normally give that feedback this early in the process. (I assume it's early because you seem to have a lot of candidates.)

[Anonymous]

I don't think you need to tell them why if it's a disqualifying factor and they're not moving on.

[Anonymous]

This is not a job for someone who's good at prompts. Sure, there are jobs for prompt-writers, but this is a job for an actual writer. We need to stand out in a field where all our competition sounds the same.

Yes, the checkers can be wrong, but that isn't a reason to stop using them, it's a reason to do multiple checks and to assess the results with a critical eye. So if Candidate A was 100% AI across all tools, only an idiot would say "but but but the checkers could be wrong!" Same with Candidate B, who was 0% AI across all tools. Why on earth would I look at that and say "hmmm this reads like a human to me and to the tools, but the checkers could be wrong!" Candidate C, now... Candidate C might have 10% on one tool, 0% on another tool, and 15% on the last tool. I'll assume that candidate C used an AI for formatting or similar, and if their other criteria were strong, not dismiss them automatically.

[Anonymous]

Anonymous wrote:Checkers can still be wrong--and I love the em dash.

I love the Oxford comma. That's another marker.

Why, OP, did you prohibit AI use? What were you trying to see?

[Anonymous]

I think you should evaluate the writing as it came through and down select any candidates that you thought presented good writing samples. We simply do not have the tools right now to identify AI with a high degree of accuracy.

[Anonymous]

Maybe they don't think grammar and spellcheck are AI?

[Anonymous]

Anonymous wrote:This is not a job for someone who's good at prompts. Sure, there are jobs for prompt-writers, but this is a job for an actual writer. We need to stand out in a field where all our competition sounds the same.

Yes, the checkers can be wrong, but that isn't a reason to stop using them, it's a reason to do multiple checks and to assess the results with a critical eye. So if Candidate A was 100% AI across all tools, only an idiot would say "but but but the checkers could be wrong!" Same with Candidate B, who was 0% AI across all tools. Why on earth would I look at that and say "hmmm this reads like a human to me and to the tools, but the checkers could be wrong!" Candidate C, now... Candidate C might have 10% on one tool, 0% on another tool, and 15% on the last tool. I'll assume that candidate C used an AI for formatting or similar, and if their other criteria were strong, not dismiss them automatically.

Couple of quick comments:

1. They have copied and pasted excerpts from the Bible and famous novels (can't remember which) and all the AI checkers said they were 100% AI; and
2. Are your candidates literally f**king idiots? Any 15 year old knows to run their work through all the prominent AI checkers and you can literally change 1 or 2 words and go from 100% AI to 0% AI.

I suppose #2 is the reason to reject them right there...however, just understand that what I wrote in #2 is true. You may end up hiring someone who in fact used AI, but knew the tricks so it came out 0% AI from the checkers.