Getting better results from the AI
Practical steps that improve extraction accuracy: prefer DOCX over scanned PDFs, reprocess ambiguous layouts, and configure template rules once rather than editing every CV.
RemakeCV's extraction accuracy tracks input quality. A clean single-column DOCX gives near-perfect results; a scanned two-column PDF is where errors appear. Ask candidates for Word versions where you can, reprocess ambiguous CVs (it costs no credit), and encode recurring preferences in your template rather than fixing them by hand each time.
What input gives the best results?
Ranked, best to worst:
| Input | Reliability |
|---|---|
| Single-column DOCX | Highest — real document structure |
| Single-column text-based PDF | High — text is extractable, structure inferred |
| Multi-column DOCX | Good — structure present, reading order inferred |
| Multi-column text-based PDF | Variable — reading order is genuinely ambiguous |
| Scanned PDF, clean scan | Variable — OCR quality dominates |
| Scanned PDF, photographed or skewed | Lowest |
The highest-leverage change is not a setting. It is asking candidates for the Word version. One email moves a CV from the bottom of that table to the top.
What should I do when a CV parses badly?
Reprocess it
Free, and on ambiguous layouts a second pass sometimes resolves the ambiguity differently.
If it comes back the same, stop reprocessing
Two identical results mean the document is being read as well as it is going to be. Further attempts waste time.
Correct in the editor
Fix the specific fields. See reviewing and editing.
For a bad scan, get a better source
A re-scan or a Word version beats any amount of correcting.
If a pattern recurs, report it
A layout that consistently parses wrong is something we can fix at the pipeline level. Email support@remakecv.com with an example.
Configure once instead of editing every time
If you find yourself making the same correction on every CV, it is a configuration problem, not an AI problem.
| Recurring edit | Fix once |
|---|---|
| Reordering sections | Set section order in the template |
| Renaming a heading | Set the heading in the template |
| Reformatting dates | Set the date format |
| Deleting a section you never use | Hide it at template level |
| Typing the same field | Add it to your template |
| Extracting the same detail manually | Set up a custom extraction field |
See date formats, section order and headings.
What can't be fixed by configuration?
Genuine source ambiguity. If a CV lists a client name in the position where an employer normally goes, no configuration can reliably resolve that — the document itself is ambiguous, and a human is the right arbiter. That is what the editor is for.
Frequently asked questions
- Does reprocessing the same CV give a different result?
- It can, on genuinely ambiguous layouts, and it does not cost a credit. If two passes agree, the layout is being read as well as it will be — correct it in the editor instead.
- Which input format gives the best results?
- A single-column DOCX. It carries real structural information rather than requiring reconstruction from positioned text.
- Can I give the AI instructions for my agency?
- Recurring rules belong in your template configuration rather than per-CV instructions. Email support@remakecv.com with the rule.
Related articles
Last updated . Still stuck? Email support@remakecv.com or book a call.