Recommended Free Tools
A resume can parse incorrectly because information is lost or misread at any of several steps: the file may not be accepted, text may not be extracted, layout may scramble the reading order, or extracted words may be assigned to the wrong profile fields. Parsing is structured data extraction—not the later decision about whether a candidate is suitable. A parsing problem can leave a profile incomplete or require manual correction; the documented workflows here do not establish that a parsing error automatically rejects an application.
How resume parsing works, from file to candidate profile
“The parser” is shorthand for a chain of separate operations. The boundaries matter: a file can pass one stage and still fail at the next. The examples below draw on Greenhouse Recruiting support guidance, Roche candidate guidance, and Apache Tika documentation. Their product-specific behavior should not be assumed to describe every applicant tracking system (ATS).
As an Amazon Associate I earn from qualifying purchases.
| Stage | What happens | What can break |
|---|---|---|
| 1. Intake and type detection | The service accepts a file, checks it, and routes it according to its format. | A file can exceed a product’s size limit, be malformed, or be detected as a type for which the installed parser has no handler. Apache Tika distinguishes detecting a type from having a parser available for it. |
| 2. Text acquisition | A format-specific parser extracts a document’s text. OCR may be needed to turn text visible only in a scan into machine-readable characters. | A scanned page may have no ordinary text layer. If OCR is not configured or does not recognize the image accurately, the visible words may never reach later stages. |
| 3. Layout and reading order | The system turns positioned text into a sequence and attempts to identify sections. | Columns, tables, headers, footers, or text boxes can make the sequence ambiguous, so text may be reordered or omitted. |
| 4. Field mapping | Extracted content is assigned to profile fields such as contact details, work experience, and education. | Unfamiliar section headings, incomplete job titles, or inconsistent formatting can leave fields empty or put information in the wrong place. |
| 5. Output and storage | Structured fields and related content are emitted and saved or displayed as a candidate profile. | An output format may discard detail, or an exception may be recorded in metadata rather than raised in a way the calling application displays. Tika documents these possibilities; this is an engineering example, not evidence that a particular ATS uses Tika. |
| 6. Validation and recovery | The system reports an outcome, and a person or workflow checks the resulting record. | A successful parse can still contain incorrect values. If an error is not visible or the profile is not reviewed, bad or missing data can persist. |
Four different ways extraction can fail
These categories are a practical way to diagnose the symptom, rather than a published taxonomy from one vendor. Start by asking whether the information is absent, misplaced, misclassified, or blocked by an operational error.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
No text was acquired
If a resume is a scan or image, its words may be pixels rather than embedded text. OCR is a separate capability, not an automatic property of every file parser. Apache Tika’s image parsers do not read image pixels by default; its documentation describes OCR options including Tesseract and vision-language parser options. Its PDF configuration also provides different strategies—AUTO, OCR_ONLY, and OCR_AND_TEXT_EXTRACTION—along with page limits and thresholds. The result therefore depends on the tool and its configuration, not merely on whether a person can read the page.
#1 Best Overall
Text exists, but the sequence is misleading
A two-column resume may look orderly to a person while its text is extracted in an unexpected sequence. Roche warns that some ATSs may read columns straight across rather than top-to-bottom and may drop header or footer information. Greenhouse likewise lists columns, complex tables, graphics, and contact details placed in headers, footers, or text boxes among formatting risks. These are documented risks, not proof that every parser fails on every such document.
Text was acquired, but the fields are wrong or empty
Finding words is not the same as understanding their role. Greenhouse cites unclear sections, inconsistent formatting, company names without identifying terms, and incomplete job titles as causes of incorrect or partial interpretation. A parser may also skip content it judges implausible: Greenhouse’s examples include fake names or company names. Roche recommends conventional section labels, since an unusual heading can be harder for a system to classify.
Rank #2
The process failed operationally
A parser can stop before producing a usable record, for example because of an unsupported format, an exception, or a resource limit. Apache Tika distinguishes an exception associated with an individual document from Tika Server fork-process failures such as a timeout, memory exhaustion, or crash. Its documentation also notes that a container-level exception can be stored in metadata rather than thrown; a caller that checks only for a thrown exception could miss it. These details illustrate why systems need to inspect and surface errors, but they do not describe the internals of Greenhouse or Roche.
What Greenhouse and Roche tell applicants about formatting
Greenhouse Recruiting’s documented limits and failure examples
Greenhouse Support’s troubleshooting article, last updated March 2, 2026, says Greenhouse Recruiting cannot parse resumes larger than 2.5 MB. That is a Greenhouse-specific limit, not a general ATS file-size rule. The same article groups causes under file size, fake resumes, and formatting issues. It lists spaces between letters; graphics, photos, and word art; image uploads instead of DOCX or PDF; complex tables; contact details in headers, footers, or text boxes; column layouts; unclear sections or inconsistent formatting; company names without identifying terms; and incomplete job titles. This is vendor troubleshooting guidance, not a ranked measure of how often failures occur across the industry.
Roche’s employer-specific recommendations
Roche Careers advises candidates to avoid tables, text boxes, logos, images, graphics, columns, headers and footers, uncommon section headings, and important words embedded in hyperlinks. Roche recommends DOCX over PDF for parsing accuracy in its candidate guidance, while noting that PDF better preserves visual layout. Treat this as Roche’s recommendation, not a universal rule: file-format behavior depends on the receiving system and the document.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.How to reduce errors and correct a bad parse
For a candidate, the goal is not to make every resume look identical; it is to keep important information easy to extract and then verify the profile the system created. A simple, readable file also makes errors easier to spot.
- Use a text-based document. Prefer a DOCX or a PDF with selectable text when the application accepts it. If a PDF contains only scanned page images, OCR may be necessary, and the receiving system may not have OCR enabled.
- Keep the reading order straightforward. Use a single main text flow where practical. Avoid putting essential contact information in headers, footers, floating text boxes, graphics, or tables when applying through a system known to have parsing trouble.
- Label sections plainly. Use recognizable headings for experience, education, and skills, and give job titles and employer names enough context to be identifiable.
- Check the file before uploading. Confirm it opens, contains selectable text if it is a PDF, and is within any stated size or format limit. Do not treat Greenhouse’s 2.5 MB limit as a limit for other systems.
- Review the populated application fields. Compare names, contact details, dates, employers, titles, and education against the original document. A parse that completes is not proof that every field is correct.
- Use the correction path if parsing fails. Greenhouse says a failed parse leaves the resume attached and requires the candidate’s details to be entered manually. Its instruction is: “If a resume fails to parse, you’ll need to manually input the candidate’s details into the fields.” Roche also tells applicants to review the application fields.
What accuracy claims do—and do not—show
There is no universal accuracy figure established here for ATS resume parsing. A result is meaningful only when the parser or model, input types, languages, fields, dataset, and metric are identified. Text extraction, section classification, field extraction, and candidate-job ranking are different tasks; performance on one does not establish performance on the others.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →- ResumeBench (EMNLP 2025): Zijian Ling and coauthors evaluated 24 language models on 2,500 synthetic resumes spanning 50 templates, 30 career fields, and five languages. Results varied across models, and the paper notes challenges in cross-lingual structural alignment. The resumes are synthetic, so this benchmark should not be treated as an exhaustive sample of real applicant resumes. Its authors write: “JSON outputs enhance schema compliance but fail to address semantic ambiguities.”
- A 2019 BERT paper: Bhatia, Rawat, Kumar, and Shah report tests involving 715 LinkedIn-format resumes and 1,000 non-LinkedIn PDF resumes. They report 100% accuracy for distinguishing LinkedIn from non-LinkedIn formats on a test of 100 of each, and 100% classification into subcategories on a 100-resume LinkedIn test set. Those percentages are bounded to the paper’s small, specific test sets and tasks; they do not show that a general-purpose parser or ATS is 100% accurate. The paper also evaluates candidate-job suitability, a downstream task distinct from extraction.
When comparing tools, look for the same test corpus and field definitions, then check whether evaluation covers selectable-text PDFs and scanned PDFs, DOCX, complex layouts, multiple languages, OCR, field-level completeness and correctness, malformed input, traceable errors, and human correction. Without comparable conditions, headline percentages can describe different problems.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
