The English Test Hidden in Your Frontline Job Application

Most frontline hiring now runs through AI resume screening and a text-heavy online application, and nobody set up that combination to test written English. It tests it anyway. If you hire warehouse pickers, home care aides, line cooks or cleaners, a good share of the people who could do the job well are being scored on how they write about the job, which is a skill the job barely uses.

The numbers aren't small. Foreign-born workers made up 18.6% of the US labor force in 2023 [1], and about 8% of US residents over the age of five say they speak English less than "very well" [2]. They are overrepresented in the hourly, shift-based roles where agencies and employers hire in volume.

What the application actually measures

Think about what a candidate for a forklift job runs into. A career page in English. A form that asks them to describe their experience in a free-text box. Maybe a resume upload, which gets parsed and scored against a job description full of phrases like "demonstrated commitment to safety culture." Someone who has driven a reach truck for six years without an incident, and who explains things perfectly well out loud, types three short lines with a couple of grammar mistakes. The parser finds few keyword matches, and the score comes back low.

Nobody decided to reject that person for their English, and written English isn't on the job description. But the pipeline has one door, and the door is made of text.

Compare that with the work itself. A picker has to understand spoken safety briefings, read a handheld scanner, and tell a supervisor when something's wrong. A care aide has to talk with patients and families and follow a care plan. Some roles do need solid written English, so test for that directly. Most frontline roles need a working level of spoken English, which is a different skill, and a written form is a poor way to measure it.

ChatGPT made the filter stranger

You could argue generative AI has already fixed this. A candidate can paste their three rough lines into ChatGPT and get back a paragraph that sails through any parser. We've written about what happens when every application is well written, and the short version is that the text stops telling you much about anyone.

But look at who uses those tools. It skews younger and more comfortable with software. A 50-year-old cleaner with fifteen years at the same hospital, applying from their phone on a break, is less likely to run their answers through a chatbot first. So the written filter now measures something even odder than English: partly, whether you knew to get a machine to write for you. That's further from the job than before.

The long application costs you before any scoring happens, too. A 27-minute application mostly filters for patience, and a free-text section in your second language makes those 27 minutes a lot longer.

Separate the language question from the job question

The fix I'd push for is boring. For each role, decide what level of English the job needs and in what form, spoken or written. Write that down as its own requirement. Then assess it on its own, in the form the job uses, and stop letting it leak into every other judgement about the candidate.

For most frontline roles, that means the first real screen should be spoken. A ten-minute conversation tells you far more about whether someone can follow a safety briefing than any paragraph they type. It also gets you the information the form kept failing to collect: which equipment they've run, which shifts they can cover, how long the commute to the site would be.

The other half is keeping the written part as short as you can get away with. Name, contact details, the role, availability. Everything else can come out in the conversation.

How Asendia AI handles this

I'm biased here, since this problem is close to why we built Asendia as a voice-first AI recruiter and not another form. When someone applies, Asendia calls them, usually within hours and at any time of day, which matters when your applicants work nights. The call asks the questions a good recruiter would ask for that role, in the same structure every time, and follows up when an answer is vague. The candidate who typed three short lines gets to explain those six years on the reach truck properly.

Because the screen happens out loud, you hear the spoken English the job needs, and you can score it as its own item instead of letting it drag down everything else without anyone noticing. The answers and a summary land in your existing ATS next to the application, so whoever reviews the shortlist sees the conversation and not just the parsed resume.

The agencies using it tend to have this exact problem: hundreds of applicants a week for hourly roles, a small team, and no realistic way to phone all of them. Asendia lets them talk to every applicant without adding recruiters. As far as I can tell, that's the only way to stop the text filter from making the decision by default.

Final Word

This isn't an argument against written applications or against AI screening. It's an argument for knowing what your screen measures. If the first thing between a frontline candidate and a conversation is a block of text, you're testing writing, whatever the job description says.

Here's a quick check. Take one high-volume frontline role and pull the last 200 applicants. Sort them by resume score and read the bottom 30 yourself. Count how many show real, relevant experience written in short or awkward English. If it's more than a handful, your screen has a requirement nobody agreed to, and some of those people are working for your competitors by now.

Ready to transform your hiring strategy? Schedule a Demo with our founders today!: https://asendia.ai/talk-to-founders