Screen candidates
Goal
Get a first, consistent read of ten CVs against one job ad, so you know which to read closely first, without letting a machine decide who gets an interview.
The scenario. A small design studio advertised a junior designer role and received twelve CVs as PDFs and Word files. The founder has one evening. She wants an ordered reading list and a reason for each CV's position, and she wants to be sure no good candidate was lost to the tool's blind spots.
Read this first: what this recipe is and is not
- Resume Analyzer is AI-generated: scores and advice are a model's opinions, not measurements. The same CV can score a little differently on a second run, and a language model can invent or misread details.
- Use scores to order your reading, never to reject. The chapter says explicitly not to use it as the only basis for hiring decisions.
- Language models can carry bias (names, schools, gaps, formats). Nothing in the chapters says this tool corrects for it. See Trust and verification.
- Other people's CVs go to an AI service. Read Privacy, and check your own rules and local law on automated screening of applicants before you start.
- There is no PDF Redactor in Documentize, so you cannot blank out names or photos with a built-in app.
Starting state
- One file per candidate (PDF or DOC; scans are accepted). If one candidate sent several files, join them with Merger.
- A written job description (the real ad) and a job title.
- Protected CVs unlocked (Unlock).
Chain
(Merger) → Resume Analyzer, one CV at a time, same settings → your own score table → read the top AND a sample of the rest
Step by step
1. Fix your criteria before you look
- Write 4 or 5 things that matter (for example: portfolio quality, tool skills, communication) and who judges them. The score will not know these. This protects you from letting the tool set the criteria.
2. Resume Analyzer: one CV per run, identical settings
- App: Resume Analyzer.
- Upload one CV per run. The chapter says that with two or more CVs the app always switches to a comparison report, whatever operation you chose. For comparable individual scores, run them one by one.
- Operation type: Compare with Job Description (a job-match analysis; needs the job text) or ATS Resume Extraction (match score and missing keywords). Pick one and use it for every CV.
- Job title / Job description: paste the same text every time. A vague ad gives vague results.
- Theme: Professional. Leave the report parts on.
- Download each PDF report.
3. Build your own table
- In a spreadsheet, one row per candidate: file name, score, missing keywords, red flags, your own notes. This is the comparison step: you do it yourself.
- The "several CVs at once" comparison mode exists and you can try it as a second opinion, but treat its ranking the same way: as a prompt to read, not a verdict.
4. Check the machine
- For the top 3 and for 3 from the bottom, read the CV yourself against your criteria.
- Re-run one CV a second time with the same settings. If the score moves a lot, give less weight to small differences.
- Check facts in the extraction: dates, employers, names (scans especially).
Expected result
A table of scores and flags, plus your own judgement on a short list. The deliverable is your reading order, not a decision.
Check it worked
- Did a candidate you expected near the top land low? Open that CV and ask why. If the reason is format (a designer's visual CV that an ATS reads badly) or missing keywords for synonyms, the tool misled you.
- Is every score based on the same job text? A different ad paste gives different scores.
Variations
- Your own CV: Resume Analyzer with ATS Resume Extraction, then Improve Resume (needs the job description).
- Industry change: Industry Fit Analysis (fill in the Industry field).
- Facts only: Extract Resume Data turns each CV into structured data for your own table.
- Poor scans: Searchable first (the app also recognises scans itself).
What can go wrong
- You uploaded five CVs and got one comparison. That is by design; upload one at a time.
- "Improve" complains about the job description. It needs the text.
- Generic advice. Paste the real ad and a specific title.
- Scores cluster together. Then they tell you little; read more CVs yourself.
- Bias goes unseen. Sample the bottom of the list on purpose.
Why this order
The criteria come first so the tool does not define them for you. Identical settings make the scores comparable. The table is yours because the app compares only when you hand it several CVs at once, in a mode you cannot tune. Human checking comes last on purpose: it is the step that makes the result safe to use.