Everyone points AI at the CV. Why does nobody point it at the decision to apply?
I logged 47 of my own applications and checked them afterwards. Where I met the core requirement of the role I converted at 71%. Where I was stretching, 8%. My confidence score at the point of applying predicted nothing at all.
That was not a writing problem. I had been using AI at the last step, polishing the CV, when the expensive mistake happened one step earlier, choosing the role. And nothing I was paying for could have told me, because none of it held my history.
3 things I would like to hear from this room, whether you have job hunted recently or you are on the hiring side:
What stayed manual for you? Not what was annoying, what you actually did by hand every single time.
Where did you give up? Most people I ask say the tracker, around week 3.
Have you ever pointed AI at the decision rather than the output? Triage before you apply, or a retro afterwards on what came back.
I launched WorkstationAI here today. It tailors CVs like everything else does, but the part I built it for is the log, the triage and the retrospective, so I am genuinely trying to find out whether that loop matters to anyone but me.
Featurehunter
Capture user feedback from your most valued customers.
Comments (5)
I have used AI before the CV-writing stage: I give it my background and a job description and ask whether applying makes sense. The problem is that a one-off conversation has no memory of what I applied for previously, which CV version I used, or what produced a response. It can generate a confident match score, but it cannot learn from my actual outcomes.
What remains manual for me is deciding which requirements are genuinely essential rather than standard wording, documenting why I chose to apply, and retaining the context needed for a useful retrospective. The tracker becomes difficult to maintain when most applications produce no explicit outcome—there is simply another silence to record.
That is also my main question about WorkstationAI. I would hesitate if the triage produced only a single match score. I would trust it more if it showed:
- which core requirements are supported by evidence from my experience;
- which signals are missing;
- which previous applications are genuinely comparable;
- how uncertain the recommendation is.
If the product can make the retrospective useful even when most outcomes are silence, that feels much more differentiated than CV tailoring.
How do you currently classify applications that receive no response: as unsuccessful, unknown, or a separate outcome?
One additional piece of context I should have mentioned: I’m also a co-founder of HRGrowbit, a product that approaches the hiring decision from the employer’s side: hrgrowbit[dot]com
Asem is the founder and leads the HR methodology and assessment direction. I joined as co-founder and am responsible for building the technical platform.
HRGrowbit helps employers evaluate candidates before the first interview through professional, cognitive, and psychological assessments, resume analysis, and comparable reports. This is why WorkstationAI is particularly interesting to me: your product helps candidates decide whether to apply, while ours helps employers decide whom to move forward.
Both products face a similar trust problem from opposite sides: an automated recommendation is only useful when people can understand the evidence behind it—not just see a score.
Short answer to your last question first: no response is its own outcome, not a decline. In my tracker anything with no reply after 21 days moves to "No response", kept separate from "Declined". Both count as resolved when I run the retro. Silence is data as long as you don't pretend it's nothing.
On how the retro actually works. Every application leaves a record at the time I made it: the job ad, the score, the gap I flagged, which CV version went out, the date. When something resolves I run a decline triage on it, and the triage only has 3 verdicts: fit gap, writing lesson, or noise. Most declines are noise, and saying so out loud is what stops you rewriting a CV that was never the problem.
Two weeks ago I ran it across all 47 resolved applications at once. The AI found a pattern I could not see from inside any single application. Job ads soften their own must-haves: payroll filed as "nice to have" for a Time Off PM, payments listed as optional by a company whose entire product is payments infrastructure. I read those labels and scored myself a 5. I now call it the spine gap. No spine gap: 7 applications, 5 interviews, 71%. Spine gap: 39 applications, 3 interviews, 8%. My CV was never the root cause. The targeting was.
But here is the part that matters most. The score reviews the job. It does not decide. My only interview from a "2, weak match" came from overruling my own score and applying anyway, because I wanted that role. So the rule changed: a 2 is worth applying to when the gap is named plainly in the cover letter and answered with proof.
AI is a suggestion, never an instruction. It sees what I cannot see. It cannot see what I want. That is the human part, and I do not want to automate it away.
Your read on trust is right, and it is why the live tool showing only a score is the thing I am fixing next. Given HRGrowbit sits on the other side, what does an employer do with an honest "we are not sure about this candidate"?
↳ Replying to You Li
That is a very good question. My view is that “we are not sure” should change the employer’s next action, not silently lower the candidate’s rank.
Uncertainty can mean at least two different things:
- the available evidence is incomplete;
- the evidence points in conflicting directions.
Neither should automatically become a rejection. A useful assessment should state what remains unknown, why it matters for the role, and what could resolve it: a targeted interview question, a work sample, clarification of previous experience, or an additional assessment.
The candidate should remain under consideration until that evidence exists. Otherwise, the system turns missing information into negative evidence and may disproportionately filter out people with unusual career paths.
Asem leads HRGrowbit’s methodology, so this is my product and technical perspective rather than a formal methodological answer. But the principle I would like us to preserve is very close to yours: AI can structure the evidence, expose gaps, and suggest what to investigate next. It should not disguise uncertainty as a precise score or make the final decision.
In hiring, an honest “we do not know yet” is often safer and more useful than false certainty.
That distinction between optimizing the application and optimizing the decision to apply is really interesting. In my experience, the most manual part is still figuring out whether a role is actually worth the time before tailoring anything — reading the requirements, comparing them against my experience, and deciding whether a gap is realistic or just wishful thinking. I also think the retrospective piece could be especially useful because most job seekers collect outcomes but rarely turn them into a feedback loop. If WorkstationAI can show someone patterns like “you consistently get responses from roles with X but not Y,” that feels more valuable than another layer of CV polishing.
Sign in to comment or upvote.