Paste your resume text, get an Authenticity Score on a 0 to 100 scale, and see which specific bullets carry the AI signal that Workday, Greenhouse, Lever, and iCIMS now surface on the recruiter view. The score is the summary; the per-bullet colour map is what you actually fix. Resumes concentrate AI signal more than any other professional document because each bullet is short, structurally rigid, and surrounded by other short bullets. That is why a STEM master and a recent-grad with a strong project section can both score in the 50 to 70 band on their first scan even after writing every line themselves. This page is the pre-application check job seekers run before they apply: scan the resume, read the highlights bullet by bullet, rewrite the flagged ones with quantified specifics, re-scan to verify the score moved above 80, then submit.
This is the routine job seekers actually follow before submitting a tailored application. The score is the entry point. The per-bullet highlights are where the work happens. Re-scanning is what closes the loop before the application leaves your laptop.
Open the TextSight detector and paste the full text of your resume, header through Education. The free tier covers 1,500 words per month, which is enough for two to four typical resume versions or several revisions of one master document. Bracketed placeholders such as [Company], [Result], or [Percentage] should be replaced with the real values before pasting because every major ATS classifier flags bracketed tokens as template indicators and pulls the score down by 10 to 20 points on their own. The scan returns in a few seconds.
The Authenticity Score runs from 0 to 100 where 100 reads fully human to the classifier and 0 reads fully AI. Treat the number as a summary. The colour map underneath highlights every bullet that tripped one or more signals. Green bullets passed every check. Yellow bullets tripped one or two. Red bullets tripped three or more. Most resumes have five to twelve bullets carrying most of the AI signal. Those are what you act on, in that order: red first, yellow second, green left alone.
Open your resume alongside the highlights and rewrite the flagged bullets before reaching for any tool. The single biggest practical lever is quantified-impact specificity. A bullet that names the tool, the team size, the time frame, and the percentage move will almost always score human-likely, because the specificity is the signal that LLMs hallucinate worst. Replace cycled action verbs with the verb that actually describes the work. Replace vague impact phrasing with the real number, even if the number is small. Replace skills-list bullets with one concrete project that demonstrates the skill.
Paste the revised resume back into the detector and re-scan. Aim for 80 or higher on resumes; the band is tighter than essays because each bullet concentrates signal. If a single bullet still flags red, go back to step 3 for that one bullet; do not run a Maximum-mode rewrite pass over the whole resume because that flattens the specifics that make the resume work in the interview. Then submit through your normal channel. TextSight does not interact with any specific ATS provider and we make no promises about specific application outcomes; we score honestly so you can decide whether the resume is ready.
A number on its own does not tell you whether to apply. These five bands describe what the classifier is seeing, what major ATS providers tend to do with the same resume, and what the right next move is at each band.
Almost certainly clears the AI-detection layer Workday, Greenhouse, Lever, and iCIMS added through 2025. Recruiters reading the bullets will not sense AI patterns either. This is the target band for senior-role, FAANG-tier, and competitive analyst applications where an AI flag meaningfully reduces interview rates. Submit through your normal channel and move on to the cover letter.
Most ATS providers will not flag in this range. Recruiters reading the resume usually do not notice AI patterns on the first sweep. For non-competitive roles or roles where applicant volume is moderate, this band is fine. If you have time, scan the remaining yellow bullets in the colour map and run one editing pass on the bullets in your most recent role; those are the bullets recruiters weigh hardest, so they are the two-minute edits with the highest leverage for moving from 75 to 85.
Roughly 30 to 50 percent of ATS classifiers in this range will surface an AI-suspect flag on the recruiter view. Recruiters who use AI-detection browser extensions may see the flag too. Edit the red bullets using the four resume-specific fixes (quantified impact, specific tool, named project, real verb) before submitting to a role you actually want. Two or three targeted rewrites usually move a 60 into the high 70s. This is the band most STEM and recent-grad resumes start in even when written entirely by the candidate.
Most ATS providers will flag in this range. Recruiters sorting by AI-flag status will see your application lower in the queue, and the ones who read it will read it with the AI frame already in place. The fix is structural rather than cosmetic. Restructure the bullets (drop cycled action verbs, replace vague impacts with real numbers, swap skills lists for specific stories) before resubmitting. A single rewrite pass will not move a score in this band into safe territory.
Almost certainly raw or lightly-edited LLM output, probably with the same prompt across every bullet. ATS detection will flag this consistently. Recruiters who get the application at all will read it last, if at all. The fix is a complete rewrite from your own memory of the actual work you did, not a quick edit. Use the AI rewriter only on individual hardened sentences after you have rewritten the structure from your own notes.
Resumes use the same five base signals as the essay scorer, but with weighting calibrated for short bullet-driven content. Action-verb cycling and vague-impact phrasing carry the heaviest weight because they are the strongest resume-specific tells. Quantified-specificity detection runs as a positive offset that lifts the score back up when you do the work.
"Spearheaded", "Orchestrated", "Leveraged", "Pioneered", "Championed", "Drove", "Engineered", "Architected" all appearing on the same resume in rotation. LLMs cycle through the same fifteen high-flash verbs because they were trained on resume guides that recommend exactly that list. Real candidates use the verb that actually describes the work, often a simpler one (Built, Wrote, Shipped, Reduced, Fixed). The scorer flags resumes where more than five flashy verbs appear on a single page.
"Resulting in increased efficiency", "Leading to significant improvements", "Contributing to enhanced productivity". Real impact comes with a number. LLMs avoid numbers because they hallucinate the wrong ones, so they default to abstract impact phrasing. The scorer counts the density of impact phrases that lack any number or named outcome; resumes above three vague-impact phrases per page land in the yellow-to-red band on this signal alone. Replace each instance with the specific result, even if the result is "by 4%" or "from 11 minutes to 8".
"Proficient in Python, Java, C++, SQL, AWS, Docker, Kubernetes, and Terraform." This comma-separated skills cram is a classic recent-grad resume pattern and one of the patterns ATS classifiers learned to flag earliest, because it carries zero evidence the candidate has actually used any of those tools. The fix is replacing the list with one or two bullets that name a single tool and the specific thing you built with it. Keep the skills cram in a separate Skills section if the format demands it; do not put it inside an experience bullet.
Resumes where every bullet runs 14 to 18 words trip a structural flag because real careers do not produce uniform-length stories. LLMs default to uniform length because they were trained on advice that says bullets should be "consistent and parallel". Real bullets vary: a strong quantified bullet might run 22 words, a clean one-line achievement runs 9. The scorer treats variance in bullet length as a positive signal and uniformity as a negative one.
This signal pulls the score up rather than down. The scorer detects bullets that name a specific tool, a specific team size, a specific time frame, or a specific percentage or dollar move. Each named specific is worth roughly two points of score lift on resumes, capped at twelve points. This is the cheapest score gain on a flagged resume and the single highest-leverage thing to add before re-scanning: turn one round of edits into a quantified-specifics pass, and the score typically moves 8 to 15 points on its own.
All three AI rewriter modes available on every paid plan. Free covers two to four resume versions. Active job seekers running tailored resumes per application usually move to Pro for the full search. Full details on the pricing page.
Billed $89.88/year, Save $30
Billed $179.88/year, Save $60
Billed $359.88/year, Save $120
Yearly billing saves 25%. View full pricing →
The detection layer in major applicant tracking systems caught up through 2025, and resume parsers got the upgrade before the cover-letter parsers did. A flagged resume can route your application to a lower review tier before a recruiter ever opens it. Here is what changed and what the score predicts.
Workday surfaces a writing-quality flag on the recruiter view that fires on resumes and cover letters alike. Greenhouse and Lever both expose third-party AI scores through integration partners; at high-volume employers the flag appears next to the candidate name in the dashboard. iCIMS rolled out similar at enterprise tier. The flag does not always block the application, but it influences ranking inside the recruiter dashboard, and at companies receiving more than 500 applications per role it routinely decides whether your resume gets a first-pass read.
Internal evaluation against a 100-resume benchmark spanning recent-grad, mid-career, STEM, sales, and analyst categories shows TextSight Authenticity Score correlates within 10 to 15 percentage points of the ATS AI flag for major providers. It is a directional pre-flight check, not a mirror of any specific classifier. A TextSight score above 80 typically means the ATS will not flag the resume. A score below 50 typically means it will. The 50 to 80 middle band is where the prediction loosens and the per-bullet highlights matter most for deciding what to revise. Calibration leans deliberately toward a low false-positive rate; the most recent run on the 100-resume set showed a 4.1% FPR overall, with the highest FPR sitting on STEM resumes at 6.2%.
A resume bullet is 12 to 22 words. A cover-letter sentence is 18 to 32. An essay sentence runs longer still. The signal density per word is roughly four times higher on a resume bullet than on an essay sentence, which means every AI tell carries dramatically more weight per line. A single cycled action verb at the start of a bullet can drop the score by 5 points where the same verb buried mid-paragraph on a long essay would drop it by less than 1. The scorer is calibrated for that density and the bands above reflect it.
The ATS flag is one risk; the recruiter is the other. After reading a few hundred AI-generated resumes, the pattern recognition is automatic and arrives in the first five seconds of scanning the page. The score helps you catch what the recruiter would catch on the bullet sweep.
Senior recruiters reviewing fifty resumes a day make the assessment on the first bullet under your most recent job title. By the second bullet the frame is set, and the rest of the experience section reads through that frame. Specific, quantified work in the top bullet is the single highest-leverage edit you can make on a resume. The scorer weights action-verb cycling and vague-impact phrasing accordingly: a quantified opener that names the tool, the team, and the result moves the score more than any other single change.
Spearheaded, Orchestrated, Leveraged, Pioneered all on the same page signal a template to the recruiter the same way they signal one to the classifier. Cut them. Use the simpler verb that actually describes the work (Built, Wrote, Shipped, Reduced, Fixed, Trained, Hired). The score moves accordingly when you make this edit and the recruiter response moves with it, because a plain verb followed by a specific result is the rhythm a real career produces and no LLM defaults to.
"Proficient in Python, Java, C++, SQL, AWS, Docker, Kubernetes" inside an experience bullet is the recruiter equivalent of a sign saying "this candidate has used none of these in production". Move the skills cram to a separate Skills section if you must, then replace the experience bullet with one project that actually used one of those tools. A score that lands in the 80 to 90 band after this edit usually does so because the skills bullet was carrying half the AI signal on the page.
An honest pre-flight check is closer to a careful proofread than to anything else. We want to be explicit about which side of the hiring-trust line this scorer sits on so you can decide whether it fits your situation.
Resumes you wrote yourself, including ones where you used ChatGPT to draft bullets from your own notes and then edited them down. The career is yours, the tools are real, the numbers are real. The scorer catches bullets where assistant register leaked into the prose so the submitted resume reads in your own voice and the interview conversation lines up with the page. We score honestly so you can decide what the resume needs.
It is not a tool for fabricating qualifications or pretending you wrote something you did not. The AI rewriter cannot put authentic experience into a bullet about work you did not do. If your resume sounds AI because the underlying experience is borrowed wholesale from a job description, the scorer will tell you that and no rewrite pass will magically fix it. The most useful thing TextSight can do for that case is point you back to writing about the work you actually did, even if the work feels less impressive than the job description sounds.
The output of a good revision pass should pass a simple test: if the recruiter asked you in an interview to walk through any specific bullet on the resume for two minutes, you should be able to do it confidently. If you cannot, the revision added voice but not substance, and the resume will fail anyway when you reach the screen. The score is a draft check, not a substance check, and we are honest about that limit because the alternative is a candidate who passes the bullet sweep and then stalls on the first behavioural question.
The upstream detector page with the full benchmark methodology and ATS calibration details.
Read the methodology →The sister scorer for cover letters, with template-opener and enthusiasm-cluster calibration.
Open the scorer →Why STEM and ESL writing get flagged at higher rates, and what we calibrate for to keep FPR low.
Read the guide →How we measure precision, recall, and false-positive rate across document types and writing styles.
Read the methodology →The full four-fix rewrite workflow for ChatGPT cover letters with before/after examples.
Open the guide →How the 0 to 100 score is computed and what threshold to aim for across formats and genres.
Read the guide →Authenticity Score, per-bullet highlights, ATS-aware calibration. Free for two to four resume versions per month.