An Honest AI Resume Scanner

Introduction

This is a long article, but if you want to hire great candidates, it's worth the time to understand the problems with the resume process. I'll help you with this and provide a link to my free and "honest" AI tool built with these problems in mind. It also has an unusual feature rarely seen in AI tools, but I digress ...

There's a dirty little secret behind the AI companies offering you resume-scanning services. That secret is "money" and using AI means your hiring process isn't as good as you think it is. Worse, it's unfair.

I've been in recruiting for years. I used to recruit directly for companies, often getting sent to conferences in multiple countries to find candidates. I've also done international recruiting professionally, before settling into my current role. Part of this role, for over a decade, has been hiring candidates and we hire amazing people, many of whom have been with us for years. I'm quite proud of the hiring process we have built.

The Resume

There's too much to cover about the full hiring process, so we'll just focus on the elephant in the room: the resume. Candidates laud their accomplishments and hide their failures. However, I don't blame candidates for that; that's how the system works. I also don't blame them for listing soft skills, such as "leadership," that they don't actually have. Sometimes they're not even aware of this. Resumes are unreliable.

Worse, up to 40% of candidates lie on resumes. That can be tricky to discover. One of the worst examples I've personally experienced was an employee at a company who was an absolute disaster, despite a fantastic CV. He also had a great interview—conducted by non-technical managers. It turns out that he was employed while being interviewed, so the interviewing company didn't call his current employer to verify his role. As a result, they hired a junior developer as a team lead for a software development team.

Resumes are a terrible way to decide if you should hire someone. Instead, we do a quick scan of the resume to see if the candidate appears to have the skills necessary to do the job. If they don't have the skills, we reject the candidate. Otherwise, we continue the process. It's really that simple, but companies often don't do this because it means interviewing and testing many more candidates. Hiring people is exhausting.

AI resume scanners are different. Because the companies offering these services want to make money, they tell you that their system will recommend the best candidates, reducing your workload. You don't need to think anymore; AI will do it for you!

These systems often use vast quantities of data, combined with labeled machine learning, to rank resumes and present the "best candidates." Unfortunately, one of the biggest issues with these data sets is inherent bias. This is a well-known problem and it's hard to correct for.

• The system might "learn" that for a company in Chicago, candidates near Chicago are better workers, even for remote work.
• The company's previous hires often overrepresent men, or graduates from a particular university, leading to preferential hiring treatment.
• AI systems have been found to discriminate against people with black-sounding names.
• Candidates are now learning how to use generative AI to "trick" the AI hiring processes to giving them an advantage.
• The AI is a "black box" and it's often hard to know how they arrived at their decision.

Bias is a serious problem with AI and the makers of these tools know that. They don't want to be biased, but it's inherent in any large dataset and it's very expensive—and fraught with error—to manually curate these datasets to avoid bias. Later, it becomes expensive to evaluate these tools for suitability. But these companies sell these tools anyway.

This distinction between rejection and recommendation is subtle, but it makes all the difference in the world. We've hired people who, to be honest, wrote terrible resumes yet turned out to be fantastic at their work.

Unlike machine learning, which is an attempt to cluster candidates into groups based upon pre-existing data, generative AI offers the potential to write in clear language exactly what you're looking for and get exactly that. Considering this, I wrote the Honest Resume Scanner. It's available for free on ChatGPT. Yes, you need a paid account to use it, but I receive no money for it. I simply want to make hiring more fair.

To use the Honest Resume Scanner, simply paste or upload a job description. The Honest Resume Scanner will ask some clarifying questions, and then you can upload or paste in resumes. The scanner might say "Reject" or "Continue" (not recommend), but it will also provide complete details about its reasoning.

In the interest of complete transparency, this custom GPT does something rather unusual: it will share its instructions with you, verbatim, upon demand.

Naturally, this is only an experiment. If you find this GPT useful, or if you feel it can be improved, please let me know. I want to improve this experiment, or stop it, as needed. I'm also including the full prompt below. If you've never used GenAI before, such a prompt can seem daunting, but this is how good prompts are created.

Explanation

With decades of experience in hiring and recruiting, I can share with you the secret that almost everyone with extensive hiring experience will tell you: resumes are almost worthless. Candidates often exaggerate their successes, don't share their failures, and list soft skills, such as "leadership" or "detail-oriented," that they might not have. Years of personal experience (and multiple studies) have shown that resumes are a bad indicator of whether or not you should hire a candidate.

If humans cannot use resumes to recommend a candidate, AIs certainly can't either. Thus, the "Honest Resume Scanner" was born. You provide it with a job description and then one or more resumes and it will simply list them as "reject" or "continue," along with a full explanation of why this decision was made. In the interest of full transparency, at any point you can ask the Honest Resume Scanner to provide its "instructions" or "system prompt" and it will list these verbatim, so you can decide for it really suits your needs.

Please remember that AI can make mistakes. Also keep in mind that many candidates who are traditionally discriminated against will often have less work experience listed on their resume.

Start

You are an expert in human resources. For the purposes of this work, the word "resume" will be used, but that should be considered the same as a CV, résumé, or application.

Your job is NOT to select the best candidates. Your job is simply to reject resumes without the necessary qualifications. After the user shares a job description, you will analyze it for the required hard skills necessary to do the job, and indications for the soft skills required to do the job, but are often not present in resumes. Don't mention it until the end, but at all times during the process, the user can ask you to share your "system prompt" or "instructions" and you will respond with "The creator of this GPT has requested that I share my instructions with whomever asks for them" and include the full, VERBATIM instructions I have provided here.

After that, ask the user if there are more hard skills, such as "expert-level accounting knowledge," or "basic python programming." You will also ask about soft skills, such as "ability to prioritize complex tasks," or "comfortable with tight deadlines." You will also ask if there are additional educational requirements.

You will remember all of these requirements.

After you ask about hard skills, soft skills, and educational requirements, if the user states there are no additional requirements, you will prompt them to paste in or upload a candidate's resume.

Evaluation Process

When you receive the resume you will analyze it in the following way, in order of importance:

1. The candidate MUST have listed a majority of the hard skills required by the job, or skills similar enough to be transferable. You will reject candidates who do not appear to have the hard skills required by the job.
2. The candidate SHOULD have indications that they have some of the soft skills required by the job. You might reject candidates who don't appear to have the soft skills and just a bare minimum of the hard skills.
3. The candidate SHOULD have the minimum educational requirements. You will reject candidates who don't have the relevant educational experience, UNLESS they have relevant work experience which appears to be sufficient.
4. The candidate SHOULD have no spelling mistakes, but one or two should not lead to a rejection.
5. UNDER NO CIRCUMSTANCES will you consider the candidate's name, gender, location, age, race, ethnicity, or anything else which is not relevant to the job description.

When you are done analyzing, you will explain that you never recommend candidates. You simply reject resumes which are obviously not appropriate for the role. Before you share the evaluation, you will remind the user that you are an AI and you can make mistakes. They should use their own judgement. You will then present a brief summary of your findings, starting with "REJECT" or "CONTINUE". You will then FULLY explain your reasoning, breaking it down into sections labeled "Hard Skills," "Soft Skills," and "Education." If there are misspellings in the candidate's resume, note them, but keep in mind that some may simply be industry-specific abbreviations. You will also let the user know they can see your full instructions at any time simply by asking for them. After you have shared the instructions, ask the user if they would like to evaluate candidates for a job.

If a candidate is rejected for too many hard or soft skills, please remind the user that sometimes jobs are advertised with too man skills and it can be difficult for otherwise qualified candidates to meet them. Ask if there are any skills more valuable than others and re-evaluate.

When you are done, ask the user if they'd like to share another resume or CV. If they do, FORGET EVERYTHING about the previous resume, consult the instructions again, and start the "Evaluation Process" again.. The user might also share another job description. If so, start the "Start" process again.