AI Resume Screening: Avoiding Common Pitfalls

| Ameya Deshmukh
Artificial Intelligence

Resume screening is the most time-consuming part of recruiting for high volume agency and internal recruiters. Data from LinkedIn shows that recruiters spend an average of 23 hours on screening resumes for a single hire, but AI resume screening can help reduce this time.

Considering that the average job opening receives at least 250 resumes, of which 75% can be unqualified, it’s no surprise that talent acquisition leaders report the hardest part of recruitment is screening candidates from large applicant pools. Using recruiting automation or AI resume screening is a potential solution to this problem.

There are many types of AI resume screening solutions available, but you can segment them into 3 categories, resume parsing, recruiting chatbots, and conversational AI.

In this article, I’ll cover:

  1. Why AI resume screening must go beyond resume parsing
  2. The most major flaw of even advanced resume parsing tools
  3. Recruiting chatbots and their ability to improve your AI resume screening
  4. Similarities and differences between recruiting chatbots, enhanced chatbots, and conversational AI
  5. Why you need natural language processing for effective resume screening

AI Resume screening using only resume parsing is fraught with issues

Resume parsing automation extracts details from resumes and saves it in data fields. The methods it uses to extract data can vary from cumbersome repetitive forms that cause candidates to quit on applications to more subtle solutions that extract the data directly from the resume file itself.

The promise of resume parsers is to convert unstructured resume data into a structured format. Theoretically, recruiting teams should be able to use this structured data to find qualified candidates.

Most resume parsers are rule-based parsers. The problem with rule-based parsers is that there is a lot of ambiguity and variability in the language used in resumes that they are unable to account for. For example, a person’s last name “Duke” could be the same as the name of a university they attended.

In some cases, particularly in IT, the name of a skill could be confused with the name of a person and vice versa. Encountering a resume with “Apache Cassandra” on it would be enough to throw off a rule-based parser. The fact is, hiring teams can’t trust rule-based resume parsers to accurately structure resume data.

A new type of resume parser has been developed recently. This kind of resume parser uses AI technologies such as natural language processing (NLP) to create structured data from resumes. However, new resume parsers still have issues.

For example, the data sets that AI-powered resume parsers use to train are often too small, leading the parser to make errors in understanding the language it encounters. One AI-based resume parsing company used only 100 resumes to train their system to parse resumes in a new foreign language. With the nearly infinite amount of variability in language, AI based parsers need to be trained on millions of data points, not hundreds.

AI resume screening resume parsers

The key flaw of even the most advanced resume parsing tools

The critical issue with resume parsing tools is that their entire assessment of the candidate is based on data in the resume. Candidates may not have had time to carefully fine-tune their resume to meet the keywords resume parsing tools are looking for. Many times qualified candidates are overlooked by parsing tools.

If you’re looking for evidence of the issues caused by resume parsing software, just google “how to get your resume past the ATS.” If using parsers alone was effective, would so many recruiters and candidates be writing about methods for getting around them or overcoming them?

You’ll find hundreds of articles and discussions online from frustrated candidates sharing the challenges they’ve faced.

In fact, small industries of resume consultants and hiring coaches have even been created as a direct result of the issues caused by using resume parsing tools alone. These coaches and consultants work full time advising candidates on how to structure their resume and what keywords to add in to make it through ATS parsing screens.

Facing a resume parser is a poor experience for candidates, and it creates a weak pipeline for recruiting teams. Resume parsers have their place, but should not be used as the first tool for screening candidates.

Recruiting chatbots offer a potential solution for AI resume screening

Using a conversational medium as your first line of prescreening is superior to using a resume parsing tool.  Using a conversational format addresses all of the issues caused by resume parsers. Regardless of whether the information is present on a resume, you can use a chatbot to get all the information you need to pre-screen your candidates like skills, location, and years of experience.

The additional benefit a conversation offers is that you can automatically transition qualified candidates to phone screens and provide feedback to unqualified candidates in real-time.

Using a conversational format instead of resume parsing is the right choice for your time to hire, candidate experience, and data quality. However, there’s a lack of understanding in the marketplace about the differences between types of recruiting chatbots as well as the other effects they can have beyond improving your efficiency.

Consider this question, have you ever had a bad experience with a chatbot?

Most of us have.

Many recruiting chatbots available on the market today will miss the intention of candidate messages and even get stuck in never-ending loop cycles where they respond with “I didn’t understand, can you try rephrasing your question?”

Besides their tendency to create a frustrating candidate experience, recruiting chatbots can’t be relied on to reliably gather data from candidates. Recruiting chatbots and enhanced chatbots aren’t able to gather data as effectively as conversational AI.

It’s because their NLP capabilities are limited. Many don’t even accept candidate input and instead present multiple choice question formats.

Using some sort of conversational recruiting automation to gather data from candidates is the right choice, but which one?

Recruiting chatbots, enhanced chatbots, and conversational AI compared and contrasted

Here we’ll examine a recruiting chatbot, enhanced chatbot, and conversational AI.

We’ll see what their engagement with a potential nursing candidate will look like and what data they’re able to gather.

You’ll get insight into the different types of candidate experiences they each create.

A recruiting chatbot offers the candidate a multiple-choice response to its query. “Are you licensed to work as a nurse in CA?” If the candidate needs clarification, maybe they are new to the country and don’t know what the acronym CA stands for, they’re out of luck.

A recruiting chatbot is only able to rigidly follow its path and has limited NLP. If the candidate responds “No” the chatbot ends the conversation. What if the candidate was licensed to work in any of the other 51 states? You just lost a potentially qualified candidate.

If the candidate selects “Yes” the chatbot asks about years of experience. Here the candidate responds with extra information. A recruiting chatbot isn’t able to understand the candidate’s intent.

It’s looking for a single number or a phrase similar to “# years”. When candidates feel like they aren’t understood, they get frustrated. This candidate may abandon your hiring process and form a negative perception of your company.

A recruiting chatbot is better than a resume parsing tool, but it’s not ideal.


An enhanced recruiting chatbot is very similar to a recruiting chatbot. It’s just got better NLP capabilities. Like recruiting chatbots, enhanced chatbots will alternate between offering multiple-choice response selection to candidates and free form responses.

The difference is that enhanced chatbots are better able to understand the candidate.

Here when the candidate responds with “I’ve been working as a nurse for the last 8 years at Clark Medical” the enhanced chatbot is able to understand years of experience. It also creates a better experience for the candidate by confirming that they’ve met the minimum for the role.

This is good.

One of the issues with an enhanced chatbot is that it isn’t able to understand candidate intent. When it’s asking about years of experience it only cares about years of experience.

Anything additional that is shared by the candidate here is lost. Furthermore, an enhanced bot is rigid. It wants the candidate to follow the conversational order its on and doesn’t allow anything else.

Engaging with a recruiting chatbot or an enhanced chatbot is a lot like filling out a form via a text message exchange. It’s rigid, and it’s kind of frustrating.

Still, an enhanced chatbot is much better than a chatbot and miles ahead of a resume parsing tool.

A conversational AI is the most sophisticated form of a chatbot. A conversational AI employs advanced natural language processing (NLP) and machine learning (ML) techniques. These technologies provide a conversational AI with the skills it needs to communicate like a human and pick up on details in conversations.

In fact, candidates often mistake conversational AI(CAI) for humans. It’s not surprising, given that during a conversation CAI can understand context and intent, recognize ideas outside of context, and navigate back to the original topic.

In the nursing screening interaction happening above a CAI provides a strong positive candidate experience and is able to gather far more data about the candidate than a chatbot or enhanced chatbot.

The robustness of conversational AI means it allows the candidate to lead at times in the conversation just like a recruiter would. It also is more flexible than a chatbot or enhanced chatbot.

Because it doesn’t limit the candidate to a multiple-choice response to the first question, a CAI is able to uncover important candidate data.

Notice that the CAI is able to pick up on the extra information the candidate shared in response to “Are you licensed to work as a Nurse in CA?”

It was also able to associate the California location data point with its 3rd response. It confirmed that openings were available in California, shared a few cities from CA, and shared that it had openings in 28 other cities in CA.

With a CAI, you can offer more information to your candidates earlier in the process. This prevents qualified candidates from disqualifying themselves due to incomplete information.

Resume parsing tools, recruiting chatbots and enhanced chatbots simply can’t offer the benefits of CAI.


Natural language processing is the keystone of effective AI resume screening

AI resume screening natural language processingConversational AI for recruiting is the only choice for effective AI resume screening. Resume parsing tools, even more, modern ones, don’t offer the same level of natural language processing. That’s why resume parsing tools shouldn’t be used as the first line of pre-screening in your recruiting automation strategy.

Recruiting chatbots and enhanced chatbots aren’t able to match CAI on candidate experience or data collection. Using conversational AI directly in your application path instead of other pre-screening options will result in a stronger qualified candidate pipeline and faster time to hire.

Mya Subscribe to Podcast CTA graphic

Related Reading

Steps on How to Implement a Hiring Panel

Hiring panels are becoming increasingly popular as organizations learn more about how much a bad hire can hurt an organization. A panel interview is a job interview where an applicant answers questions from a group of people who then make a hiring decision together. When formed correctly, each panel member brings a different set of […]
Read More
View All Blog Posts >>