The corpus: 10 public forms, 6 systems, 8 weeks.
Between 8 August and 16 September 2026 we captured and dissected 10 public job application forms across 6 applicant-tracking systems: five forms on Greenhouse (one generic, four live employer applications at large US technology companies), two on Lever, and one each on Breezy, Freshteam, and Zoho. These are the same captures our field-recognition engine is tested against — real DOMs, not screenshots or secondhand descriptions.
For each form we counted the visible interactive controls a person must deal with — text inputs, selects, radio groups, textareas, and the resume upload. Hidden inputs, option rows inside selects, and the voluntary self-ID block are excluded from the field count and reported separately.
| Capture | System | Visible fields | Employer-written questions |
|---|---|---|---|
| GH generic | Greenhouse | 14 | 3 |
| GH employer A | Greenhouse | 18 | 10 |
| GH employer B | Greenhouse | 12 | 3 |
| GH employer C | Greenhouse | 18 | 6 |
| GH employer D | Greenhouse | 15 | 8 |
| Lever A | Lever | 22 | 1 |
| Lever B | Lever | 16 | 11 |
| Breezy | Breezy | 7 → 17 | 0 |
| Freshteam | Freshteam | 12 → 28 | 5 |
| Zoho | Zoho | 8 | 0 |
Download the per-form table (CSV) · Download the verbatim question corpus (CSV)
How many fields a form asks: median ~16, range 7–28.
The shortest form asked 7 fields before the resume was parsed; the longest asked 28 after it. The median form sits around 16 visible controls — before the self-ID block, which every Greenhouse and Lever form appends.
Two forms deserve a second look because they grow. On Breezy and Freshteam, uploading a resume does not shorten the form — it expands it. Breezy goes from 7 fields to 17 after parsing, Freshteam from 12 to 28. The parse pre-fills some fields and exposes the rest: education rows, employer rows, dates — all still needing human review. “Upload your resume to autofill” buys you more fields to check, not fewer to fill.
These numbers count controls, not effort. A 16-field form with three open questions (“why do you want to work here?”) costs more than a 28-field form of dropdowns. Our field notes measured the effort side on live applications; this census measures the form side.
One question, seven wordings.
Every employer form in the corpus asks about work authorization or visa sponsorship — it is the single most universal employer question we found. What changes is the phrasing. These are all the wordings in the corpus for what is, underneath, one question:
| # | Verbatim wording |
|---|---|
| 1 | Are you legally authorized to work in the country where this position is located? |
| 2 | Are you legally authorised to work full-time in the country where this job is based? |
| 3 | Will you require sponsorship for employment visa status now or in the future? |
| 4 | Will you now or in the future require sponsorship or a visa transfer for employment visa status (ie: H-1B visa)? |
| 5 | Will you require the company to sponsor you for a work permit now or in the future for the location(s) you selected? |
| 6 | Please share your work authorization. |
| 7 | Are you authorized to work in the location(s) you selected in your previous response? |
British authorised next to American authorized, “sponsorship” next to “work permit” next to “visa transfer,” one employer naming the visa type (H-1B) — a person answering seven applications reads seven versions of the same ask. (Company names in quoted text are normalized to “the company”; everything else is exact.)
Measured at scale, the duplication is bigger than it looks. The product's field map currently records 425 label spellings that collapse to 56 canonical facts — roughly eight different phrasings per fact before counting languages. “First name” alone appears under at least 19 spellings across the forms and languages we have mapped. The application problem is not that forms ask too much; it is that they ask the same things and never agree on how.
What employer forms actually ask: ten clusters.
Grouping the ~40 employer-written questions in the corpus produces ten recurring clusters. None of these exist to learn who you are — your resume already says that. They exist for legal cover, routing, and knockout filters.
| Cluster | Presence | Verbatim examples |
|---|---|---|
| Work authorization & visa sponsorship | 5 of 5 employer forms | Seven different wordings — see the table below. Always required, always a dropdown or yes/no, never pre-filled from anything. |
| Location & mobility | Every form | “What city and state are you currently located in?” · “In what cities are you available to work?” · “Please provide the zip code of your primary residence.” |
| Referral source | 3 of 5 employer forms | “How did you hear about this job?” · “How did you hear about this opportunity?” · “How did you hear about this role? (Please select all that apply)” |
| Prior employment with this company | 3 of 5 employer forms | “Have you previously worked for us?” — asked three ways, including “in any capacity” and “or an affiliate.” |
| Resume repeats | Recurring | “Who is your current or previous employer?” and “What is your current or previous job title?” — asked as custom questions in a form that already collects the resume. |
| Knockout qualifiers | 4 of 5 employer forms | “Are you at least 18 years of age?” · “Do you have 4+ years of B2B SaaS sales experience?” · “Please upload proof of certification or licensure.” |
| Compliance & integrity | Regulated employers | Government-official and conflict-of-interest questions (3 in one form) · “I certify that the information provided in this application is true and correct.” |
| Consent & marketing opt-ins | 3 of 5 employer forms | “Please email me about future job openings” · “Do you opt-in to receive WhatsApp messages from recruiting?” |
| Availability & preferences | Recurring | “Please let us know the soonest date you can start.” · “If this role offers the option to work from a remote location, do you plan to work remotely?” |
| AI-use disclosure (new in 2026) | 1 of 5 employer forms | “Which of the following best describes how you use AI tools today?” · “I understand that the company may use AI tools to assist in the application and interview process.” |
The resume-repeat cluster is worth isolating: a form can upload your resume, parse it, and then ask — as a required custom question — “Who is your current or previous employer?” This is not an edge case in the corpus; it is routine. The resume communicates to humans; the form fields communicate to the system. You are the translator between them.
The 2026 signals in the corpus.
Two patterns in the captured questions belong to 2026 specifically, and we expect both to spread:
Employers now ask how you use AI. One employer form asks “Which of the following best describes how you use AI tools today?” as a required select, and makes you acknowledge that the company may itself use AI tools in the application and interview process. A year ago neither question existed on this form. Disclosure is moving both directions.
Forms are localizing the money questions. The India-market form asks “Expected CTC (in INR)” and “Notice Period (in Days)” — compensation and notice conventions that US-centric advice pages answer badly. If you apply across markets, the same question hides behind different labels again.
How this shows up in search behavior.
We can corroborate one part of this census from the other direction. In our own search-console exports, the queries that bring people to this site include these employer questions pasted whole — punctuation, required-field asterisks, “(ie: H-1B visa)” and all. A person mid-form hits a question they do not fully understand, copies it, and searches the exact string.
That behavior explains why the wording problem matters beyond inconvenience: when a form phrases a question unusually, the applicant's search for help is unusually specific too — and often finds nothing that matches. The guides linked below exist because the questions above are what people actually paste.
Method and limits.
What a “field” is here: a visible interactive control a person must inspect, fill, or check — text inputs, selects, radio groups, textareas, and the resume upload. Hidden inputs, select option rows, and the voluntary self-ID block do not count toward the total; self-ID is reported as one line item because its length is set by regulation, not by the employer's curiosity.
Sample: 10 public form captures across 6 ATS systems, 8 August – 16 September 2026, skewed to Greenhouse (the captures exist because our packs are tested against them), US, and English. It is too small to estimate market averages — read the field range and the clusters, not a mean.
Employer anonymity: the four live employer forms were public pages when captured; names are withheld in the dataset because the value here is the pattern, not the call-out. Question text is verbatim so claims can be spot-checked. No user data appears anywhere in this census — forms, not applicants.
Staleness: employers edit forms. Each capture is a snapshot with a recorded window; treat the corpus as “what these forms looked like in Aug–Sep 2026.” We will re-run the census as the capture set grows.
Questions.
Because we only publish what we actually captured and parsed — these are the forms the engineering team recorded while building and testing field recognition, not a survey designed for a headline. A larger census needs a larger capture set; the method below is built so we can add forms over time.
No. The sample leans Greenhouse (because that is where our test applications ran), US, and English. Workday and iCIMS forms are known to run much longer — our field notes measured a Workday average of 73 fields — but those captures are not in this corpus yet.
As a block, not per option. Every Greenhouse and Lever capture ended with the voluntary self-identification section (gender, race/ethnicity, veteran status, disability — each with a decline option). Counting every option row as a field would inflate totals without measuring effort, so it is one line item here.
The public application forms themselves, captured 8 August – 16 September 2026. Employer names are withheld in the published dataset; the question text is quoted verbatim so any claim can be spot-checked against the live forms.
Disclosure: Jobsistant builds a job-application autofill and has a commercial interest in application forms being repetitive. The corpus is real captures with the counting rules stated above; we have reported the short forms alongside the long ones and the sample's lean toward Greenhouse rather than presenting the median as a market estimate.