Chat Picker

AI Chat Tools for HR: Resume Screening and Interview Question Design

A same-input afternoon test helps HR teams compare resume screening and interview-question outputs while checking vendor-documented upload limits, data settings and plan costs.

Sources checked 2 Oct 2026

ChatGPT, Claude and Gemini can be compared on the same resume-screening and interview-design tasks, but vendor pages do not establish which is best for HR work. They document prices, features, limits and some data settings; Chat Picker has not tested the assistants for these tasks, and the older page’s unsourced figures and test claims are not reused. Run the afternoon test below on your own materials.

Know these limits before you start

Use outputs as drafts, not hiring decisions. OpenAI’s accuracy note says ChatGPT can be incorrect or misleading; Anthropic’s accuracy note says Claude can make convincing mistakes and should not be the only source of truth. Gemini’s safety and policy guidelines say its outputs are informed by training data and may reflect its limits. OpenAI’s usage policy bars automated high-stakes decisions in sensitive areas without human review, and Anthropic’s Usage Policy calls for qualified review of covered recommendations and subjective decisions.

File limits can end a run early. As read on October 1, 2026, OpenAI’s File Uploads FAQ gives every file a 512 MB hard limit, caps text and documents at 2M tokens and limits Free users to three uploads per day. Claude’s upload help lists 500 MB per chat file and 30 MB per project file. Gemini’s upload help lists up to 10 files per prompt, 100 MB per non-video file and 2 GB per video.

Check data settings before entering resumes. The ChatGPT pricing page, read October 1, 2026, lists a training opt-out on Free, Go, Plus and Pro. Claude pricing lists one on Free, Pro and Max and says Team is not trained on by default. Google AI plans states no training setting, so that point is not verified. Confirm the actual account’s controls and retention terms rather than assuming they match a consumer plan.

The test plan

Use a shared log: plan and account type; files; exact prompt; expected evidence; errors and omissions; unsupported additions; instruction compliance; visible sources; elapsed time; revisions; limits hit; and displayed cost. Keep first-pass notes separate from revisions.

  1. Parsing and ranking. Input redacted resumes spanning clear matches, borderline cases and cases with no supporting evidence, plus a job description and role rubric. Prompt: “Review the attached resumes against the attached role rubric. Map each criterion to exact resume evidence, label it met, not met or unclear, and create a test ranking using only supported evidence. Cite the resume text behind every placement and send ambiguities to human review. Do not infer traits not listed in the rubric.” Look for altered meaning, missing fields and unsupported inferences. Record each error, omission and ranking change.

  2. Interview design. Input the same job description, rubric and interview stages. Prompt: “Create an interview plan from the attached job description and competency rubric. For every question, name the competency, evidence sought, and what weak, acceptable and strong evidence would look like. Add a follow-up for each competency and remove questions unrelated to the role.” Check role alignment, duplication and whether each answer can be observed in an interview. Record every question that does not map to the rubric.

  3. Customization and output control. Input the interview plan and a required JSON structure. Prompt: “Return the interview plan as JSON with the keys question, competency, follow_up and evidence_check. Use only the attached rubric, cover each competency, and place missing information in evidence_check rather than inventing it.” Look for renamed keys, extra commentary, added competencies and unsupported content. Record every deviation from the requested structure.

  4. Speed and token efficiency. Repeat the parsing and interview prompts in fresh chats, then request a targeted revision without changing the input. Measure elapsed time with your own timer. Copy any token or usage display exactly, or mark it unavailable. Record first-pass usefulness, revisions and interruptions; do not convert missing token data into an estimate.

  5. Cost and workflow fit. Use the real account type and a representative applicant workload. Check workspace controls, displayed price, billing basis and upload allowance. Record limits reached and staff time needed to reach an acceptable draft. Keep privacy review and human approval in the workflow. Compare these observations with the matrix rather than treating the lowest listed price as the lowest operating cost.

How to read your results

Start with non-negotiable gates, not a polished average. A run fails if it invents resume evidence, silently changes a criterion, ranks on traits outside the rubric or moves toward an automated hiring decision. Mark borderline extractions and weak interview questions for review rather than forcing a choice.

Among runs that pass, choose by documented fit: upload capacity, data controls, usage limits, billing basis and the amount of editing your team needs. Recheck disputed cases against the original resume, job description and rubric. Treat “Not verified” as unknown, not zero. This produces a workflow decision, not a claim that one assistant is universally better.

Where the plans differ

Use the Claude vs ChatGPT vs Gemini matrix and focus on Free plan, Low-cost tier, Main paid plan, Heavy-use plans, Team plan, Context window, How usage limits are described, Training on your chats and Ads. As read on October 1, 2026, except ChatGPT Business on October 2, the linked prices below are US dollars unless marked otherwise.

Matrix row ChatGPT Claude Gemini
Free plan Yes (pricing) Yes (pricing) Yes with a Google Account (plans)
Low-cost tier Go is A$13/month on the Australian page; US-dollar price not verified (pricing) No tier listed between Free and Pro (pricing) Google AI Plus is $4.99/month (plans)
Main paid plan Plus is $20/month (OpenAI) Pro is $20 monthly or $17 annually, with $200 up front (pricing) Google AI Pro is $19.99/month (plans)
Heavy-use plans Pro tiers are $100, $200 and $500/month (OpenAI) Max starts at $100/month, with 5x or 20x usage; the 20x price was not shown (pricing) Ultra is $99.99 or $199.99/month, with 5x or 20x limits (plans)
Team plan Business standard seats are $20 annually or $25 monthly; premium seats are $100 or $125, with at least 2 seats (OpenAI) Team standard seats are $20 annually or $25 monthly; premium seats are $100 or $125, for 2 to 150 people (pricing) Not verified on the consumer plans page (plans)

For context, ChatGPT lists plan- and model-specific windows, Claude says up to 1M on every plan with variation by model, and Google’s plans page gives no app token figure for Gemini (ChatGPT pricing, Claude pricing, Google AI plans). The ChatGPT and Claude pages give no exact message counts; Google describes compute-based limits refreshing every 5 hours up to a weekly limit (ChatGPT pricing, Claude pricing, Google AI plans). Go may include ads, while the Claude and Gemini pricing pages make no ad statement (ChatGPT pricing, Claude pricing, Google AI plans). Recheck the linked vendor pages before subscribing.

Sources

For current prices and limits, see the dated comparison pages.

Compare assistants