Overview

You ask AI to compare three candidates for a senior engineering role. It produces four paragraphs of prose, one per candidate, describing their technical skills, leadership potential, cultural fit, and growth areas.

Now you need to compare them. You manually create a spreadsheet. Candidate A vs. Candidate B on dimension X. How do they compare? You scan the prose, extract information, build a table. You spend 30 minutes doing work that shouldn't have been necessary.

You've created work. AI output should save you time, not create more work.

Better approach: Ask AI to output a comparison table directly. Candidates in rows. Dimensions in columns. You can compare at a glance. Two minutes of output review vs. 30 minutes of manual extraction. You're not processing information. You're using it.

Structured output means: asking AI to produce information in a format you can actually use. Tables, checklists, scorecards, matrices. Not prose. Not narrative. Structured.

This lesson teaches you how to get AI to produce structured output for HR work. You'll learn the formats that work best for different tasks. You'll learn how to specify structure in your prompt. You'll learn to request formats that save time and improve clarity. By the end, you'll have a toolkit of structured formats you can deploy immediately in your HR role.

Why This Matters for HR Professionals

Unstructured output (prose) requires you to extract information. Structured output gives you information ready to use.

Tables: You can scan and compare in seconds instead of reading four paragraphs. When you're evaluating five candidates or comparing three benefit plans, a side-by-side table eliminates the cognitive load of holding multiple narratives in your head. You're not mentally juggling descriptions. You're reading numbers and ratings aligned perfectly for comparison.

Checklists: You know if you've done everything. No more wondering if you missed a step. No more firefighting last-minute issues. A new hire onboarding checklist prevents the 3pm Friday discovery that equipment hasn't been ordered. It prevents the Monday morning call from a new employee who can't access the network. Checklists convert knowledge from the heads of experienced HR people into a reusable system.

Scorecards: You can rate and compare consistently across candidates or evaluations. You have data, not impressions. This matters enormously when you're defending a hiring decision to leadership or documenting performance issues for a termination. "Candidate A scored 9/10 on technical skills, 4/10 on leadership, 7/10 on communication" is defensible. "They seemed pretty good" is not.

Matrices: You can see relationships, tradeoffs, and priorities at a glance. You can make decisions faster. A prioritization matrix showing effort vs. impact immediately answers: "Should I fix the benefits enrollment form or overhaul our entire compensation philosophy?" (The form is quick; the philosophy takes months. Both matter, but timing matters more.)

Asking for structured output is a small change in how you prompt. But it saves significant time and improves clarity. The difference between "here's a paragraph about candidate A" and "here's how candidate A scores on technical skills: 4/5, leadership: 3/5, communication: 5/5." One is narrative. One is decision-ready.

Pro Tip: Structured output also creates an audit trail. When you need to defend a hiring decision, you have the scorecard showing how you evaluated all candidates. You have objective data, not subjective gut feel. This matters legally and ethically.

Why HR Professionals Resist Structured Output (And Why They Shouldn't)

Many HR professionals worry that structured formats reduce nuance. "A table can't capture the full picture of a candidate," they say. "I need the narrative to understand the person."

This is partly true. But it's backwards thinking. The narrative comes first (the interview, the reference calls, the work samples). The structure comes second. You extract and systematize what you learned. You're not losing information. You're organizing it for decision-making.

Think of it like this: your interview notes are prose. Your scorecard is the synthesis. The interview is where you discover. The scorecard is where you decide.

A VP of People Ops at a 2,000-person fintech company discovered this the hard way. They used prose feedback for 18 months: hiring managers submitted written evaluations of candidates. Nothing was standardized. One evaluation said "great communication," another said "struggled in group conversation." Two interviewers rated the same candidate with completely different conclusions. When a lawsuit came up challenging their hiring practices, they had no defense. The documentation was inconsistent, subjective, and indefensible.

They switched to a structured scorecard: five dimensions, 1-5 scale for each, clear anchors describing what 1, 3, and 5 look like. Same hiring managers, same interviews. Now they had consistent, defensible documentation. Candidates were rated on the same criteria. The data was comparable.

The process didn't lose nuance. It added discipline.

HR-Specific Formats and When to Use Them

1. Comparison Tables

When to use: Comparing candidates, comparing plans, comparing options, side-by-side analysis, evaluating vendors, assessing compensation packages.

Format:
- Rows: Items being compared (candidates, benefits plans, vendors)
- Columns: Dimensions of comparison (skills, cost, support, timeline)
- Cells: Brief descriptions, ratings, or key facts

Why this works: The human brain excels at comparing things arranged side-by-side. A table lets you scan left-to-right, top-to-bottom. You see patterns immediately. "Candidate A has strong technical skills but weak communication. Candidate B is the opposite."

Example Candidate Comparison:

Dimension
Candidate A
Candidate B
Candidate C

Technical Skills
9/10. Expert in Python, system design, databases. Built three major systems at last company.
7/10. Solid Python, learning system design through online courses. No database experience.
8/10. Expert in Go and system design. Limited Python (could learn quickly).

Leadership
Led team of 3, positive feedback from reports. Mentored two junior developers.
No direct management experience. Good at helping peers.
Led team of 5, strong mentorship reputation. Feedback mentions excellent delegation.

Communication
Clear and concise in interviews. Good at explaining technical concepts.
Good with technical peers. Struggled explaining complex ideas to non-technical stakeholder in final round.
Excellent communicator. Very clear in interviews. Strong at bridging technical and business.

Growth Area
Could strengthen frontend knowledge and user experience design.
Needs real-world system design experience beyond coursework. Has theoretical knowledge but no production responsibility.
Could deepen Python knowledge. Other skills are solid.

Fit for Role
Strong fit. Has most required skills. Would start contributing immediately.
Good fit. Will need mentorship on system design. Expects 2-3 month onboarding for core projects.
Good fit. Language learning curve is real but manageable. Her system design expertise is valuable.

This beats prose. You see the comparison immediately. You see patterns: one has leadership, one doesn't. One has weak communication, one has strong. You can discuss tradeoffs: "A is strongest overall but C might be better long-term if we invest in teaching her Python. C is the better communicator." You can see risk: "B will need significant mentorship, do we have capacity?"

Prompt pattern: "Create a comparison table for [items]. Rows: [what you're comparing]. Columns: [dimensions]. Keep cells to 1-2 sentences max. Use ratings (X/10) where helpful. Show data, not narrative prose. Be specific (e.g., 'led team of 3, feedback was positive' not just 'good leader')."

2. Checklists

When to use: Onboarding, preparation, project planning, verification, implementation tracking, transition management.

Format:
- Task description (specific and measurable)
- Owner/Responsible party (who is accountable)
- Timeline or deadline (when does it need to happen)
- Status (not started / in progress / done / blocked)
- Notes (any blockers, dependencies, or context)

Why this works: Checklists prevent two problems: paralysis and omission. You know what to do (it's on the list) and you know when you're done (everything is checked). A good checklist saves the person who wrote it the mental load of remembering every step.

Example Onboarding Checklist:

Task
Owner
Due
Status
Notes

Order equipment (laptop, monitor, keyboard, mouse, docking station)
HR
Day -5 (before start)
Done
All ordered from vendor. Delivery confirmed for Day 0 at 10am. Backup equipment available if any item is damaged.

Set up workspace (desk, chair, shelving)
Facilities
Day 1 morning
In Progress
Desk assigned in Team A area. Chair delivered. Waiting for monitor delivery to set up dual monitors. Will have full setup by EOD Day 1.

IT onboarding (email, VPN, access to tools, authentication setup)
IT
Day 1 morning (before 9am)
In Progress
Email account created and working. VPN access pending (usually 2-4 hours). Cloud storage, Slack, and project management tools all configured. Will test all access by 9am.

Manager 1:1 to discuss role expectations, team, first week priorities
Manager
Day 1 afternoon
Not Started
Scheduled for 2pm. Topics: role expectations, team dynamics, first week goals, reporting structure, how we work. Duration 60 minutes.

Team intro meeting (meet all team members, brief on current projects)
Manager
Day 3
Not Started
Calendar invite sent to all 7 team members. 30 minutes, includes casual intro and current project overview.

Benefits enrollment meeting (health, dental, 401k, HSA, FSA, life insurance)
HR
Day 5
Not Started
Materials sent via email. 30-minute call scheduled. Enrollment window is 5 business days. Deadline is EOD Day 5 (Friday).

Security training (password policy, data protection, security protocols)
Security
Day 3
Not Started
Online course (20 minutes). Required before accessing sensitive systems. Course link sent, self-paced.

Direct deposit setup (submit banking information)
Payroll
Day 2
Not Started
Form sent via secure portal. Deadline EOD Day 2 so first paycheck processes correctly.

This keeps everyone on track. You can see what's done (equipment ordered, email set up), what's in progress (VPN pending, workspace being set up), and what's not started (team intro, benefits enrollment). You can see where you're at risk (benefits enrollment deadline is Day 5, don't forget). You can see who owns what (facilities owns workspace, IT owns VPN, HR owns benefits). When someone asks "Is the new hire ready?" you can glance at this checklist and say "mostly yes, VPN still pending but should be done by 9am."

Prompt pattern: "Create an onboarding checklist for [role]. Include these elements: [equipment needed], [IT access], [meetings], [training]. Format as table: task, owner, due date, status, notes. Be specific about timelines. Include a brief explanation of why each task matters (1 sentence per task). Show realistic dependencies (e.g., some things can only be done after equipment arrives)."

3. Scorecards

When to use: Interview evaluation, performance rating, capability assessment, skill evaluation, offer decision, potential evaluation.

Format:
- Rows: Candidates/people being evaluated
- Columns: Dimensions to score (skills, competencies, fit)
- Scale: 1-5 with clear anchors for what 1, 3, 5 look like
- Include a summary or recommendation row

Why this works: Scorecards force consistency. You rate everyone on the same criteria using the same scale. This eliminates the bias that comes from comparing people sequentially ("A seems great" vs. "B seems great compared to A"). It creates defensible documentation. It allows discussion: "Let's debate whether candidate A's leadership potential is really a 4 or a 3."

Example Interview Scorecard with Clear Anchors:

Candidate
Technical Skills (1-5)
Communication (1-5)
Leadership Potential (1-5)
Cultural Fit (1-5)
Overall Score
Recommendation

Candidate A
5 - Exceeds all requirements. Expert-level knowledge in all needed areas. Can architect and execute complex systems alone. Mentors others.
4 - Clear and concise. Explains technical concepts very well. Minor room for improvement in complex multi-audience explanations.
4 - Led team of 3 for 18 months. Good mentorship feedback. Delegates well. Could handle larger team.
4 - Strong alignment with our values. Similar work style to team. Would integrate well.
17/20
Strong Hire. Top technical skills + proven leadership. Hire immediately.

Candidate B
3 - Meets core requirements. Can execute most tasks independently. Some knowledge gaps in advanced areas. Would need guidance on novel problems.
3 - Adequate communication. Sometimes unclear when explaining complex technical details to stakeholders. Gets to the right answer but communication could be clearer.
2 - No direct management experience. Shows potential to manage in the future but no track record. Would need mentorship and coaching.
4 - Good alignment. Different working style from some team members but would complement team.
12/20
Conditional Hire. Good technical fit but needs growth in communication and leadership. Hire only if leadership development is acceptable or if non-leadership role.

Candidate C
4 - Strong in most required areas. One meaningful technical gap (no experience in database optimization). Could learn this in 3-4 weeks.
5 - Excellent communicator. Very articulate in interviews. Explains complex concepts clearly. Bridges technical and business thinking.
4 - Led larger team (5 people) for 2+ years. Strong mentorship reputation. Good at delegation and developing others.
3 - OK alignment overall. Different working style from some team members. Not a culture fit issue but would need intentional integration.
16/20
Strong Hire. Excellent communication + proven leadership. Technical gap is learnable. Leadership skills are valuable. Hire second choice after A.

Now you can compare across candidates objectively. Scoring is consistent: everyone is rated on the same 1-5 scale with the same anchors. You can discuss trade-offs: "A is strongest overall (17/20). C has better communication (5/5) and proven leadership with larger teams. B is weakest overall but could work if we're willing to invest in development." You're not relying on memory or gut feel. You have data.

This also creates a decision paper. When the hiring team debates "Should we hire Candidate B?" you can point to the scorecard and say, "B scored 12/20 primarily due to communication gaps and no leadership experience. A scored 17/20. C scored 16/20 with stronger communication. The data suggests A is the strongest hire."

Prompt pattern: "Create an interview scorecard for [role]. Candidates in rows. Columns: [dimensions like technical skills, communication, leadership, cultural fit]. For each dimension, provide a 1-5 scale with clear definitions: 1 = does not meet requirement (explain what this looks like), 3 = meets requirement (explain what this looks like), 5 = exceeds requirement (explain what this looks like). Show the scale definition once at the top, then use it to rate each candidate. Include a final column with overall score and hiring recommendation (hire/conditional/no hire)."

4. Matrices

When to use: Prioritization, risk assessment, decision-making, resource allocation, vendor evaluation, capability planning.

Format:
- 2x2 or 3x3 grid
- X-axis and Y-axis represent two important dimensions
- Items placed in grid based on their scores on both dimensions
- Use the grid to guide decision-making

Why this works: Matrices show relationships and tradeoffs. A 2x2 matrix (effort vs. impact) immediately tells you what to do first. High impact, low effort? Do that now. High impact, high effort? Plan carefully. Low impact, high effort? Don't do it.

Example Prioritization Matrix:

LOW EFFORT
HIGH EFFORT

HIGH IMPACT
Do First (Quick Wins)
Plan Carefully (Major Projects)

LOW IMPACT
Low Priority (Time Fillers)
Avoid (Don't Do)

Applied to HR Projects:

LOW EFFORT (< 2 weeks)
HIGH EFFORT (> 1 month)

HIGH IMPACT
Improve job description templates (quick win - do immediately)
Overhaul performance review system (major project - plan for Q3)

LOW IMPACT
Update phone contacts list
Rebrand all HR materials (expensive, low ROI)

In this example:
- "Improve job description templates" is low effort, high impact. Better JDs lead to better candidates. Do this now. You could finish in one week.
- "Overhaul performance review system" is high effort, high impact. It matters but takes months. Plan this for Q3 when you have capacity.
- "Update phone contacts list" is low effort, low impact. Do it if you have spare time, but don't prioritize it.
- "Rebrand all HR materials" is high effort, low impact. Don't do this. It wastes resources on something that doesn't improve outcomes.

Example Risk Matrix (Likelihood vs. Impact):

LOW IMPACT
HIGH IMPACT

HIGH LIKELIHOOD
Monitor closely
Mitigate immediately

LOW LIKELIHOOD
Accept and monitor
Prepare contingency

Applied to HR Risks:

LOW IMPACT
HIGH IMPACT

HIGH LIKELIHOOD
Turnover in entry-level roles (monitor closely - normal churn)
Key leader departures (mitigate - critical risk)

LOW LIKELIHOOD
Office internet outage (prepare contingency)
Compliance lawsuit (prepare contingency + insurance)

The matrix shows: "Key leader departures are likely and would have huge impact. This is our biggest risk. We should invest heavily in retention." By contrast, "Entry-level turnover is normal and expected. Monitor it but don't over-invest in fixing it."

Prompt pattern: "Create a prioritization matrix for [items]. X-axis: [dimension 1, e.g., effort from low to high]. Y-axis: [dimension 2, e.g., impact from low to high]. Place each item in the appropriate quadrant. For each item, explain briefly why it belongs there. Add an action recommendation for each quadrant (e.g., 'do first,' 'plan carefully,' 'avoid')."

The Biggest Mistake: Not Specifying Structure

When you want structured output, be explicit about format. Vague prompts produce vague results.

Bad prompt: "Compare these three candidates."

This is open-ended. AI might produce prose, a table, or something unstructured. You'll get back whatever feels natural to the model, which is often narrative prose.

Good prompt: "Create a comparison table for these three candidates for the Senior Engineer role. Rows: candidate names. Columns: relevant experience (1-2 sentences), technical skills (rating 1-10 + brief explanation), leadership experience (yes/no + detail), communication quality (1-5 rating), key concerns or gaps, fit for this role (strong/good/fair + reasoning). Format as markdown table. Keep cells to 1-3 sentences max. Use ratings and numbers where specified, not prose."

The good prompt specifies exactly what format and structure you want. There's no ambiguity. AI knows what you need.

Components of a strong structured output prompt:

• Format type: Table, checklist, scorecard, matrix, list
  - Rows: What are we organizing on rows? (People, tasks, items, time periods)
  - Columns: What are we organizing on columns? (Dimensions, criteria, time periods)
  - Cell content: What goes in each cell? (Ratings with scale, brief text, yes/no, numbers)
  - Length constraints: How much text per cell? (1 sentence, 2-3 sentences, brief bullet)
  - Output format: Markdown table, HTML, plain text table, CSV
  - Any additional guidance: Sort order, highlight certain cells, include totals

Example of a detailed structured output prompt:

"Create a comparison scorecard for three benefit plans we're evaluating. Format: markdown table. Rows: plan names (BlueCross PPO, Aetna HMO, United PPO Plus). Columns: monthly premium (include employer and employee cost), annual deductible, doctor visit copay, emergency room copay, prescription drug coverage (separate rows for generic/brand/specialty tiers), out-of-pocket maximum, best use case (describe ideal candidate in one sentence), and overall value rating (1-5 scale where 1 = poor value, 5 = excellent value).

For costs and deductibles, use actual numbers. For coverage, use brief descriptions (e.g., 'generic $10, brand $30, specialty $50' not 'good coverage'). For out-of-pocket max, include the number and briefly explain what this means. In the 'best use case' column, describe the type of employee this plan suits best (e.g., 'younger employees with minimal healthcare needs' or 'families with frequent specialist visits'). In the final row, add 'Recommendation' with brief reasoning (e.g., 'Best overall value for diverse workforce')."

This is clear. AI knows exactly what you want: row labels, column labels, types of data in each cell, and how much detail. The output will be usable and organized.

Important Warning: Vague prompts often produce beautiful prose that looks authoritative but isn't structured. "Compare these candidates" might produce four paragraphs that read like executive summaries but don't help you decide. You'll think you have good information, but you're actually buried in prose. Structure forces clarity.

Building Your Structured Output Toolkit by HR Function

Recruiting and Sourcing

Candidate Comparison Table: Rows are candidates, columns are job criteria (skills, experience, requirements). Use 1-5 ratings or yes/no. Helps you compare at a glance.

Interview Scorecard: Rows are candidates or interview rounds, columns are evaluation criteria (technical skills, communication, fit). Consistent scoring across all candidates. Defensible documentation.

Sourcing Channel Matrix: Rows are channels (LinkedIn, referrals, job boards, recruiters). Columns are quality of candidates (high/medium/low) and effort (time and cost). Helps you decide where to focus recruiting efforts. "LinkedIn gives us good candidates but takes high effort. Referrals give us great candidates with low effort."

Offer Recommendation Scorecard: Rows are candidates receiving offers. Columns are offer components (salary, bonus, equity, benefits, start date, location). Side-by-side comparison of what you're offering each candidate ensures internal consistency and helps negotiate confidently.

Performance Management and Development

Performance Review Summary Table: Rows are employees or goals. Columns are accomplishments, growth areas, rating, and recommended next steps. Consistent format across all reviews. Defensible documentation if needed.

360 Feedback Summary: Rows are feedback themes or competencies. Columns are feedback from manager, peers, reports, and self. Shows where there's alignment and where there are gaps. "Everyone says this person communicates well. Self-rating is much lower."

Goal Tracking Scorecard: Rows are goals. Columns are goal description, status (not started / in progress / complete), accomplishment percentage, priority, and notes. Helps you track and discuss progress in 1:1s.

PIP (Performance Improvement Plan) Checklist: Rows are expectations, milestones, check-ins. Columns are what was expected, evidence of performance, progress (met / in progress / not met), next steps. Required for legal protection in termination cases.

Learning and Development

Training Needs Matrix: Rows are skills. Columns are people needing the skill, priority, and current proficiency. Shows where to invest in training. "Five people need leadership training. Two need advanced Excel. One person needs language training."

Course Outline Table: Rows are modules. Columns are learning objectives (what people will learn), content type (video, reading, exercise), time required, and assessment method. Used for course design and communication.

Learning Path Progression: Rows are courses or certifications. Columns are prerequisites (what must be completed first), duration, level (beginner / intermediate / advanced), and assessment. Shows the journey from novice to expert.

Assessment Rubric: Rows are skills or competencies. Columns are proficiency levels (1-5 scale) with descriptions of what each level looks like. Used for evaluating skills and identifying gaps.

Compensation and Total Rewards

Salary Range Table: Rows are jobs or roles. Columns are minimum salary, midpoint (50th percentile), maximum salary. Shows your pay bands and helps ensure consistent pay.

Compensation Analysis Matrix: Rows are roles. Columns are internal pay, external market benchmark (from salary surveys), difference/gap, and recommended action (adjust up / keep / decrease). Shows where you're competitive and where you need to adjust pay.

Total Rewards Statement: Rows are benefit categories (health, retirement, paid time off, wellness, perks). Columns are value or amount. Shows the total value of compensation beyond salary. "Your total rewards package is worth $175,000 (salary $120,000 + benefits $55,000)."

Equity Analysis: Rows are roles. Columns are current equity level, market level for that role, gap, and recommendation. Shows who is under-equitied and needs adjustment.

Employee Relations and Safety

Investigation Timeline: Rows are dates or events. Columns are date, what happened, who said what, evidence, follow-up needed. Creates a factual record of an investigation (harassment, discrimination, misconduct).

Conflict Resolution Tracking: Rows are parties involved. Columns are core issue, stage of resolution (initial conversation / mediation / follow-up), next steps, deadline. Keeps everyone aligned on next actions.

Accommodation Checklist: Rows are accommodation requests or steps. Columns are what was requested, conversations that happened, decision (approved / approved with modification / denied + reasoning), implementation plan, follow-up date. Shows you followed the legal process correctly.

Try This Now: Two Hands-On Exercises

Exercise 1: Convert One Piece of Prose to Structured Output

Think of something you normally output as prose (a job description analysis, a performance review summary, a compensation analysis, a candidate evaluation).

Write your usual output in prose (narrative form). Now convert it to a structured format.

Example: You normally write: "Candidate A has excellent technical skills. They're strong in Python and system design. Their communication is good. They haven't led a team before, but they show promise."

Restructured: Create a table with these candidates, and columns for technical skills (rating 1-5), leadership experience (yes/no + detail), communication (1-5), and overall recommendation. Rate Candidate A as: Technical Skills 5/5, Leadership No, Communication 4/5, Recommendation: Strong hire for IC role, consider for future leadership.

Compare the two versions. Which makes it easier to:
- Understand where Candidate A is strong? (Structured version is clearer)
- Compare to other candidates? (Structured version wins)
- Make a decision? (Structured version is faster)
- Document your reasoning? (Structured version is more defensible)

Exercise 2: Create a Reusable Template

Pick one structured format you'll use repeatedly (a candidate comparison table, an onboarding checklist, a performance review scorecard, a compensation analysis matrix).

Prompt: "Create a [format] template for [HR task]. Include: [what dimensions/rows/columns]. Show me 2-3 example rows with realistic data. This template should be usable for every [task type] going forward. Make it detailed enough to be useful, simple enough to reuse."

Save the template in Google Docs, Notion, or a folder on your computer. Label it "Template: [Name]."

Next time you do that task, start with the template. Fill in the data. You've saved yourself the setup time and ensured consistency.

Example template prompt: "Create a candidate comparison template for engineering roles. Rows should be candidates' names. Columns should be: technical skills (1-10 rating + skill list), system design experience (yes/no + detail), leadership experience (yes/no + years + team size), communication strength (1-5 + brief note), cultural fit (yes/no + brief reasoning), and hiring recommendation (strong hire / hire / hold / no hire). Format as markdown table. Show me a completed example with three realistic candidates."

Templates compound over time. You build a library of reusable formats that speed up your work and ensure consistency.

Practical Application - Monday Morning Action Steps

• 
  Identify your three most-used formats: What do you create most often? Job descriptions? Candidate comparisons? Onboarding materials? Performance reviews? List the three things you output most frequently.
• 
  For each format, ask yourself: "Is this structured?" Are you outputting a table, checklist, or scorecard? Or prose? If it's prose and you're using it to make decisions or comparisons, it should be structured.
• 
  Build a template for each format: For this week, spend 30 minutes creating one template. "Candidate Comparison Table for Engineering Roles" or "New Hire Onboarding Checklist" or "Performance Review Scorecard." Use AI to help build it. Save it.
• 
  Test the template on your next task: When you next hire, onboard, or review performance, use the template instead of your usual approach. Track: How long did it take? How clear was the output? How easy was the decision?
• 
  Refine the template: After using it once, improve it. Remove columns that didn't matter. Add columns that would have helped. Make it better for next time.
• 
  Build your library: Over the next quarter, create templates for the five most common HR tasks you do. By end of Q2, you have a library of reusable formats that save you 5+ hours per month.
• 
  Share templates with your team: If you have an HR team, share these templates. Everyone uses the same candidate scorecard. Everyone uses the same onboarding checklist. Consistency compounds across the team.

Common Mistakes and How to Fix Them

Mistake 1: Tables with too many columns. You create a candidate comparison table with 15 columns (skill 1, skill 2, skill 3, communication, leadership, cultural fit, learning ability, growth potential, salary expectations, location flexibility, start date, etc.). By column 8, the table is illegible. You can't see all columns without scrolling. The whole point of structured output is that it's immediately readable.

Fix: Limit tables to 7-8 columns. If you need more dimensions, create multiple tables (one for skills, one for experience, one for cultural fit) or switch to a checklist format.

Mistake 2: Cells with too much text. You create a scorecard but each cell is a paragraph. You've just created elaborate structured prose. It's not faster to read than a narrative, and it's harder to compare.

Fix: Limit cells to 1-3 sentences. Use ratings (1-5 scale) instead of detailed explanation. "5/5 - Expert in Python and system design" not "This candidate has deep expertise in Python and system design. They've led the architecture of three major systems. Their knowledge is extensive and they can mentor others."

Mistake 3: No consistent scale. You create a scorecard where one dimension is 1-5, another is yes/no, another is high/medium/low. You can't compare across dimensions and the scale is confusing.

Fix: Use one consistent scale across all ratings. 1-5 is most common: 1 = does not meet requirement, 2 = below requirement, 3 = meets requirement, 4 = exceeds requirement, 5 = far exceeds requirement. Apply this scale consistently across all dimensions.

Mistake 4: Ambiguous rows or columns. Your checklists say "Task: Onboarding" (too vague), "Deadline: Next week" (too vague), "Status: On track" (too vague). Someone else reads it and doesn't know what's actually supposed to happen, when it's due, or whether it's truly on track.

Fix: Be specific. "Task: Order laptop and monitor for new hire" (clear), "Deadline: EOD Thursday, Day -3" (specific), "Status: In Progress - laptop ordered, monitor arriving Friday" (concrete detail).

Mistake 5: No decision guidance. You create a scorecard showing scores but no recommendation. "Candidate A: 17/20. Candidate B: 12/20. Candidate C: 16/20." What do I do with this? Hire A? What's 17/20?

Fix: Include guidance. Add a "Recommendation" column: "Strong hire. Top candidate," "Conditional hire. Needs development in X," "No hire. Significant gaps."

Advanced Technique: Combining Multiple Formats

Sometimes one format isn't enough. You might need a matrix to prioritize, then a checklist to execute.

Example: "Create a prioritization matrix for my HR projects. Then, for the items in the 'do first' quadrant, create an execution checklist with owners and deadlines."

First output: A 2x2 matrix showing effort vs. impact. The "high impact, low effort" quadrant has: improve job descriptions, update benefits website, create offer template, audit compensation gaps.

Second output: A checklist for those four projects: who owns each, when it's due, dependencies, status.

You're using structure strategically. First you prioritize (matrix). Then you execute (checklist).

When NOT to Use Structured Output

Not everything should be structured. Save prose for strategy, narratives, conceptual thinking.

Use prose for:
- Strategy and planning (explaining your long-term people strategy)
- Narrative and context (telling the story of why something matters)
- Conceptual development (brainstorming how to approach a problem)
- Communication and persuasion (explaining decisions to executives or employees)

Use structured output for:
- Comparison and selection (which candidate to hire, which plan to choose)
- Evaluation and rating (scoring people or ideas)
- Tracking and execution (checklists, project plans)
- Data and decisions (anything you need to reference repeatedly or compare)

A strong HR document often combines both: an executive summary in prose (why we're doing this, what it means), followed by a structured table or scorecard (here's the data).

Key Takeaways

• Structured output saves time: Tables beat prose for comparison and analysis. Checklists beat free-form notes. Scorecards beat qualitative descriptions.
  - Specify format explicitly in prompts: Be clear about what you want. "Create a table with rows [X] and columns [Y]" not "Compare these candidates."
  - Use HR-specific formats: Checklists for tasks, tables for comparisons, scorecards for evaluation, matrices for prioritization.
  - Create reusable templates: Save and adapt them. Templates make structured output faster and more consistent.
  - Consistency matters: Same format every time makes patterns visible and comparison easier. Data accumulates and becomes more useful.
  - Balance structure with nuance: The interview is where you discover. The scorecard is where you decide. Structure doesn't lose information; it organizes it.
  - Scale matters: Use consistent scales (1-5 across all dimensions). Use clear definitions (what does 1 mean? What does 5 mean?). This allows real comparison.

FAQ

Q: What if AI won't format my output as a table?
A: Be very explicit: "Format as a markdown table with | symbols. Rows: [items]. Columns: [dimensions]. Do NOT output as prose. Do NOT output as bullet points. Use table format only." Sometimes you need to be blunt about the format.

Q: Can I use Excel format?
A: AI can't create .xlsx files directly, but it can output markdown tables or HTML tables that paste easily into Excel. Output in markdown or plain text table format, then paste into Excel if you prefer to work there.

Q: Should ALL my HR output be structured?
A: No. Strategy documents, narratives, and conceptual thinking work better as prose. Use structure for decisions, comparisons, and tracking. Use judgment based on how you'll use the output.

Q: What if my table is too wide and doesn't fit on one screen?
A: Your table has too many columns. Reduce to only the most important dimensions. Or break into multiple tables. Usability (can a person read it without scrolling?) matters more than comprehensiveness (fitting everything in one view).

Q: How many rows or columns before a table gets unwieldy?
A: More than 10 columns or 15-20 rows gets hard to read and compare. If you're that complex, either break into multiple tables or switch to a different format. A table should fit on one screen.

Q: Can I combine formats?
A: Absolutely. "Create a prioritization matrix for these projects. Then, for the high-priority items, create a checklist of execution steps with owners and deadlines."

Q: What if I need to track changes over time?
A: Add a date or period column. "Candidate Score (Week 1)" and "Candidate Score (Week 4)" to show how evaluation changes. Or create a timeline format: dates in rows, progress or metrics in columns.

Q: Are there tools to create these formats?
A: Markdown tables (write them in any text editor), Excel/Google Sheets, Notion, or even Word can create these formats. AI can output in any of these. The format matters less than the structure; the structure is what saves time.

Q: What's the difference between a scorecard and a matrix?
A: A scorecard rates people or items on multiple dimensions (usually 1-5 scale on each dimension). A matrix shows relationships between two dimensions (effort vs. impact). Use scorecards for evaluation. Use matrices for prioritization or relationship-mapping.

Q: How do I know what columns to include?
A: Ask yourself: "To make a decision, what information do I need to compare?" Those are your columns. If a column doesn't help with the decision, remove it. "Nice to know" information should be a note, not a column.

Q: Should I share these structured formats with candidates or employees?
A: Sometimes. Scorecards should remain internal (used to document your evaluation). Checklists can be shared (helps the person know what's expected and when). Decide what's confidential and what's transparency that builds trust.

What's Next

You've now learned how to request and use structured outputs. You understand tables, checklists, scorecards, and matrices. You know when to use each format and how to specify structure in your prompts.

Lesson 8.4 is the final lesson: building your personal verification checklist. Everything you've learned across all chapters, consolidated into one system that catches problems before they happen. You'll learn to verify your own work systematically, reducing errors and risk. The lesson before this (8.2) covered guardrails and safety. This lesson (8.3) covered structured output. The final lesson will show you how to put it all together into a bulletproof verification system that catches AI mistakes and HR mistakes before anyone sees them.