I upload a PDF with questions but the system does not recognize the questions well

Recognition gets stuck on formatting, not content

A PDF that is not recognized properly usually comes down to one of three things: the PDF is a scan (an image of text, not actual text), the questionnaire is in a table with merged cells, or the numbering differs from what the system recognizes as a question (for example "3.2a" instead of "Question 12"). In all three cases, the system sees text, but cannot reliably determine where one question ends and the next begins. The system prefers to fill in nothing rather than place an answer with the wrong question.

The fastest solution is usually to consult your customer's own portal: many portals offer an Excel export or a web form of the same questionnaire alongside the PDF, and these are usually recognized in full. If the system recognizes part of the questions, you can add the rest manually — the linked answers still come from the same supply chain file, so you don't type the same thing twice. If it concerns a question about fixed information, such as the company's registration number or the number of employees, that answer is always available separately from recognition to be transferred manually.

Why we don't "cleverly" force PDFs

The system recognizes questions based on text structure, not meaning: it does not read a list, it only recognizes patterns such as numbering, question marks and table layout. When these patterns are missing — such as with a scan without underlying text layer, or with a PDF formatted with images instead of text — the system cannot establish question-answer pairs. That is a deliberate limit: better to leave part of the list unfilled than to link an answer to a question it cannot read with certainty. You remain the person who confirms each suggested answer and submits it to your customer's portal.