The applications are not landing. You have sent forty and heard back on two, and somewhere around the thirtieth you started wondering whether a person has read any of them. So you open one of the free AI resume tools, paste in your history, and let the machine tell you what is wrong with it.

The instinct is sound. Something automated probably is reading your application before a human does, and a tool that speaks the same language can help. What these tools cannot do is understand why your history looks the way it does. For a job seeker whose record includes medical leave, supported employment, part-time stretches, or a vocational rehabilitation placement, the distance between what the tool optimizes for and what your career contains is where the trouble starts.

The Three Kinds of Tools You Will Run Into

They get marketed as one product category and they are not one thing.

ATS checkers compare your resume against a specific job posting and score the overlap. Jobscan is the best known of them. You feed it your resume and the job description, and it returns a match rate plus a list of terms the posting uses that your resume does not.

AI resume builders and rewriters generate or rework the content itself. You describe a job you held, and the tool produces bullet points. General chat assistants do this on request, and dedicated products wrap the same capability in a template editor.

Autofill and cover letter generators handle volume. They populate application forms from a stored profile and draft a letter for each posting.

The first category is mechanical and low risk. The second and third are where judgment gets handed to something that has none.

What ATS Checkers Are Genuinely Good For

Applicant tracking systems are parsers before they are anything else. They read a document and try to sort it into fields: employer, title, dates, skills. Anything that interferes with that reading costs you. Two-column layouts, text inside images or graphics, contact details tucked into the header or footer, and tables all parse unreliably. A checker that flags those problems is doing real work on your behalf.

Keyword extraction is the other honest use. A posting that says "care coordination" eleven times is telling you the phrase the system is filtering on. If your resume says "case management" throughout, you are describing the same work in vocabulary the filter does not recognize. Matching the posting's language is not gaming anything, it is answering the question that was asked.

Neither of those functions requires the tool to know a single thing about you, which is exactly why they are the safe part.

The Part That Gets Smoothed Away

An AI rewriter returns the most statistically ordinary version of your career it can produce. That is what the underlying model is built to do. It has read millions of resumes and it hands back the shape of the average one.

The average one is continuous. Full-time, ascending, no interruptions. When your history is not that, the tool treats the difference as an error to correct. Fourteen months of medical leave becomes a vague "career transition" that invites the exact question you did not want asked. A supported employment placement gets inflated into something you did not do. Three years of part-time work while managing a condition get compressed into a date range that hides the fact that you were working the entire time.

Some rewriters go further and offer to explain the gap for you. The suggested language tends to edge toward health information: recovering from, managing a personal matter, returning after treatment. You are under no obligation to volunteer any of that, and a tool with no concept of disclosure will hand it over on your behalf. Our guide to explaining employment gaps without disclosing disability works through language that answers the question honestly without opening your medical history, and the companion piece on where gaps belong on the page covers the formatting side.

The fabrication problem applies to every job seeker, and it lands harder when your history already invites scrutiny. Language models invent numbers. Ask for a stronger bullet point and you may get "increased team efficiency by 30 percent" attached to a job where nobody measured anything. Every figure the tool produces is yours to verify or delete before the resume leaves your hands.

What You Type In Does Not Necessarily Stay With You

Consumer AI products vary widely in what they retain and whether your input trains the next version of the model. Free tiers are the most permissive. Paid and enterprise tiers frequently are not, and most chat products have a setting that controls this, usually switched on by default.

That setting matters more than usual when the thing you are pasting is a diagnosis, an accommodation request, a treatment schedule, or your reason for leaving a job. Resume-hosting sites add a second layer. Some of them sell profile data or make hosted resumes searchable by recruiters, which is fine for a work history and much less fine for anything you did not intend to publish.

The workable rule is to describe the job, not the condition. A tool can write a serviceable bullet point about coordinating a caseload of sixty without knowing anything about your health, and it will not do the job better if you tell it.

When the AI Is Sitting on the Employer's Side

The tools you choose are one half of this. The other half runs after you hit submit: video interviews scored by software, timed game-based assessments, one-way recorded questions with a countdown clock.

Those are the systems that can screen you out for the disability rather than the qualifications. Speech differences, facial differences, motor differences, processing speed, and eye contact patterns all affect the signals such systems measure, and none of them relate to whether you can do the job.

Under the ADA, you can request a reasonable accommodation during the application process, not only after you are hired, and that covers assessments. A live interview in place of a recorded one, extra time, questions in writing, or an alternative format are all requests an employer is obligated to consider. Make the request before the assessment link expires rather than after. Our articles on video interview accommodations and requesting extra time for interview responses cover how to word it.

Some states and cities require the employer to tell you an automated tool is in play. Illinois requires notice and consent before an AI system analyzes a recorded interview. New York City requires employers to notify candidates about automated employment decision tools and to have those tools independently audited for bias. A notice like that is your cue to ask what accommodation is available.

A Division of Labor That Holds Up

Give the machine the mechanical work: parsing checks, formatting cleanup, keyword extraction from the posting, and first-draft phrasing you were going to rewrite anyway.

Everything else stays with you. How your gaps get described, whether and when you disclose, which accommodation to request and how to word it, and whether any given number on the page is true. Those are decisions about your life, and the consequences of getting them wrong land on you rather than on the software.

The forty applications that went nowhere were probably not a phrasing problem. They were a volume problem, a keyword problem, or a parsing problem, and a tool will find those faster than you will. Run the next posting through a checker, fix what it catches in the formatting and the keywords, then close the tab before it starts rewriting the parts only you can explain.