What a cover-letter score can tell you
A scan can return an AI-text estimate for a letter long enough to process. It cannot establish who wrote the letter, whether its claims are true or how well the candidate would do the job. A hiring decision needs evidence about the candidate's work and experience.
We have not published separate false-positive measurements for cover letters or business writing. Our published evaluation includes a non-native English writing slice, but that result does not establish accuracy for job applications by non-native speakers. The methodology explains the score's limits before you use one in a decision.
If you are doing the hiring
- State your rules about AI assistance in the application instructions. Explain what help is allowed and what candidates should disclose.
- Assess the letter against the role. Identify a claim about experience or a project that you can ask the candidate to explain, and check it against the material they supplied.
- Use consistent interview questions and a relevant work sample where appropriate. Decide what you are assessing before you review the answers.
- If authorship becomes a concern, explain the concern and give the candidate a chance to respond. Do not use an AI score to automatically reject an application or to label someone dishonest.
Formal English and polished sentences are not grounds for rejecting a candidate. Liang and colleagues' 2023 study found that the detectors they tested misclassified non-native English writing. It studied those tools and samples; it does not measure this product's performance on today's applicants.
If you are applying
Make the letter specific enough that you can discuss what it says. Check every project, qualification and claim about the employer, then read the application rules for any disclosure requirement. Keep the drafts you already have and a copy of what you submitted.
If you choose to use the free text scan, submit one letter of at least 150 English words. A low result cannot predict how an employer's tool will score it, and a high result does not establish that you used AI. Do not rewrite accurate, useful details just to lower a score.
If your letter is challenged, ask what evidence raised the concern and offer the drafts or work examples that help explain it. The false-accusation guide gives you a way to prepare that response without treating another scan as proof.
For freelancers and clients
Agree on permitted AI assistance before the work starts, along with the deliverables and the review process. In a dispute, compare the work with those terms and discuss the drafts or revision history available. A detector result alone cannot establish that a contract was breached. The email guide covers related questions about workplace messages.
Sources
- Liang et al., GPT detectors are biased against non-native English writers (2023). Research on the tools and writing samples studied, not a measurement of cover-letter accuracy for Cobalynx.
Common questions
Could a recruiter really tell if I wrote my letter with AI?
A recruiter may suspect AI assistance, but writing style and a detector score do not establish authorship. Cobalynx has not published separate accuracy measurements for cover letters. Ask about the work and the application claims rather than treating a percentage as proof.
Should I reject a candidate whose letter was flagged?
No. Do not reject a candidate on an AI-detector score. Check the application against the stated requirements and give the applicant a chance to explain disputed claims. Our published evaluation rates are not validated error rates for a hiring population.
Was it a bad idea to write my cover letter with AI?
Follow the employer's rules on AI assistance and disclosure. Check every factual claim, remove invented achievements and make sure you can explain the experience described. A detector score does not decide whether your use complied with those rules.
Could I see what a recruiter would see if they scanned my letter?
You can see the result Cobalynx produces for your text, but cannot predict another detector or an employer’s decision. A likely-human result is not proof of authorship. Keep drafts and revision history; the false-accusation guide explains how to respond to a disputed score.