How JobApplier.site Handles Custom Application Questions

Every job application has questions beyond "upload your resume." Work authorization, salary expectations, "why do you want to work here?" — each one matters, and getting any of them wrong can sink an otherwise strong application. Here's how JobApplier.site handles every type.

The Two-Track System

We learned early that using GPT for every form field is a recipe for disaster. GPT hallucinates. When it hallucinates on "Do you require visa sponsorship?" and answers "No" when you actually do, you've created a legal problem — for yourself and the company.

Our solution: a two-track system.

• Track 1: Deterministic handlers — For questions with a known correct answer (work authorization, EEO, binary yes/no questions). These read from your config file and never touch GPT.
• Track 2: GPT with guardrails — For freeform questions where creativity matters ("Why this company?", "Describe a technical challenge"). GPT generates the answer, but with strict anti-cliche prompting and length constraints.

Track 1: Deterministic Questions

Work Authorization

The most critical question on any application. Your answer is configured once in your YAML file:

work_authorization:
  authorized_us: true
  requires_sponsorship: false
  authorized_eu: false

The agent reads this config and selects the correct radio button or dropdown option. No GPT involvement. No hallucination risk.

Salary Expectations

Configured as a range in your preferences:

salary:
  min: 180000
  max: 220000
  currency: USD
  default_answer: "Open to discussion based on total compensation"

When a salary field appears, the agent uses your configured range or default text. For dropdown fields, it selects the bucket that contains your target range.

EEO / Demographic Questions

These are legally voluntary. The agent selects "Decline to self-identify" for all demographic questions by default. You can override this in your config if you prefer specific answers.

Yes/No Pattern Questions

Common binary questions and their deterministic handling:

• "Are you willing to relocate?" → Reads from relocation: true/false in config
• "Are you 18 or older?" → Always "Yes" (must be true to work legally)
• "Have you worked at this company before?" → Always "No" (with optional override)
• "Are you currently employed?" → Reads from currently_employed: true/false

Track 2: GPT-Generated Answers

The Problem with Default GPT

Ask ChatGPT to answer "Why do you want to work at [Company]?" and you'll get:

"I am excited about the opportunity to join [Company] because of your innovative approach to [industry] and your commitment to [value]. I believe my skills in [technology] would be a great addition to your talented team."

This is what every other AI tool generates. Recruiters have seen it thousands of times. It's an instant tell.

Our Anti-Cliche Prompting

JobApplier.site uses a structured prompt that forces specificity:

1. Company-specific detail injection — The prompt includes the company name, a brief description from the JD, and any recent news we can find. GPT must reference at least one specific detail.
2. Banned phrases list — "I am excited," "great opportunity," "passionate about," "hit the ground running," "I believe" — all banned. GPT must find different language.
3. Length cap — Maximum 3 sentences for short-answer fields. This forces conciseness.
4. Tone calibration — "Write as a senior engineer speaking to a peer, not a student writing a cover letter."

The result reads like a real person wrote it — because the constraints eliminate all the AI cliches.

Example Output

For "Why do you want to work at Stripe?":

"Stripe's developer API is the reason I got into payments infrastructure — I've integrated it into three production systems. The Atlas team's work on international business formation solves a problem I've seen firsthand at two previous startups. I'd bring 6 years of Python/PostgreSQL backend experience to the billing platform challenges you're hiring for."

Specific. Grounded. No cliches. References real Stripe products.

Edge Cases We Handle

• Multi-select checkboxes — When a question lists technologies with checkboxes, the agent checks the ones that appear in your master resume
• File upload prompts — "Upload your portfolio" or "Upload a code sample" — the agent skips these with a note in the dashboard that manual follow-up is needed
• Calendar/date pickers — "What is your earliest start date?" — reads from your config (start_date: 2026-09-01)
• Text areas with word limits — GPT is instructed to stay under the character/word limit shown in the form field

Quality Control

Every GPT-generated answer is logged to the dashboard so you can review what was submitted. If an answer is poor, you can add it to a "corrections" file that the agent uses as few-shot examples for future applications.

Over time, the answer quality improves because the agent learns from your corrections — not by modifying its weights, but by appending your preferred answers as in-context examples.

Try It

Configure your deterministic answers in 2 minutes, then let the agent handle the rest. Start free — 3 applications/month, every question handled.