# How RemakeCV's AI formatting works

> RemakeCV extracts text from the CV, uses AI to identify roles, employers, dates and bullets as structured data, then renders that data into your agency template.

Source: https://www.remakecv.com/help/formatting-cvs/how-ai-formatting-works
Last updated: 2026-08-21

---
RemakeCV formats a CV in three stages. First it extracts raw text from the document, using OCR when there is no text layer. Then an AI model parses that text into structured fields — employers, job titles, date ranges, bullet points, education, skills. Finally those fields are rendered into your agency's template.

## Stage 1: text extraction

RemakeCV first has to turn a document into text.

- **DOCX and DOC** carry real structure — the paragraphs, tables and headings are all machine-readable. This is the most reliable input.
- **Text-based PDFs** have a text layer, but no structure. The text has to be reassembled from positioned fragments, which is why column order can be ambiguous.
- **Scanned PDFs** have no text at all. RemakeCV runs optical character recognition over the page images and works from the recognised text.

The CV records which path it took as `processing_method`: `text` or `image`.

## Stage 2: structured parsing

This is where the real work happens. The extracted text is a flat wall of words; the AI's job is to decide what each part *is*.

For each role it identifies the employer, the job title, the location, the date range, and the bullet points that belong to that role. It does the same for education, qualifications, certifications and skills.

Two problems make this genuinely hard:

- **Attribution.** When a candidate has held three roles at the same employer, the bullets under the second role must not be attached to the first. When a CV lists a client name next to an employer, the employer must win.
- **Reading order.** In a two-column layout, naive extraction interleaves the columns. The parser has to recover the order a human would read.

> **Note:** 
The parser preserves the candidate's wording. It decides where text goes, not what it says. If you want the language changed, that is a separate opt-in step — see [first-person neutralisation](https://www.remakecv.com/help/anonymisation-and-compliance/first-person-neutralisation.md).

## Stage 3: rendering into your template

The structured data is written into your agency template — your fonts, your colours, your section order, your headings, your date format. Because the data is structured rather than copied, the same candidate renders identically into any template you own.

This is also why switching template is cheap: the parse is already done, so regenerating into a different template costs nothing and does not consume a credit.

## Why does the output sometimes need correcting?

Because the input is genuinely ambiguous. Some recurring cases:

| Source problem | What you may see |
|---|---|
| Two-column layout | Content in an unexpected order |
| Scanned document | Occasional character errors from OCR |
| Same employer, several roles | Roles grouped or split differently than you would |
| Non-standard date formats | A date range read differently than intended |
| Sections in an unusual order | Content placed in a neighbouring section |

None of these require re-uploading. Correct them in the editor — see [reviewing and editing](https://www.remakecv.com/help/formatting-cvs/reviewing-and-editing.md) — or reprocess, which does not cost a credit.

## What does the AI never do?

- It does not invent employment history, dates or qualifications.
- It does not remove content from the CV unless you ask it to anonymise.
- It does not change the meaning of a bullet point.

If content is missing from the output, it was missing from the extraction — not deleted. See [the formatted CV looks wrong](https://www.remakecv.com/help/troubleshooting/formatting-looks-wrong.md).

## Frequently asked questions

### Does the AI rewrite the candidate's words?

No. The extraction step preserves the candidate's own wording. Bullet points are moved, not rewritten. The only features that generate new text are the candidate summary, the cover message and first-person neutralisation, and each is opt-in.

### Which AI model does RemakeCV use?

RemakeCV uses large language models from multiple providers and selects between them based on accuracy and availability. The model behind the pipeline changes as better ones ship; the output contract does not.

### Is my candidate data used to train AI models?

No. Candidate CVs are processed to produce your formatted document and are not used to train models. See security and sub-processors.