The short answer is that you are fighting a format war with a hammer when you need a different tool entirely. Power Query is excellent for semi-structured data, but PDFs are not data, they are documents designed for human eyes, not machine parsing. The jumbled text you describe is not a failure of your technique; it is the inherent limitation of treating a presentation format as a data source. Every vendor that sends you a PDF with a different layout is effectively speaking a different dialect, and no single Excel feature can translate them all without a custom schema for each one.
What you are actually describing is the gap between document intelligence and spreadsheet logic. AI tools that promise tabular output often fall short because they treat the problem as a text extraction task rather than a structural inference problem. The real breakthrough happens when you stop asking Excel to read the PDF and instead ask an AI model to understand the invoice *as a human would*, by recognizing the semantic roles of each field, not just its position on the page. This is not about building a custom solution for each vendor; it is about training or selecting a model that can generalize across layouts by learning what a "total due" or "invoice number" looks like in context, regardless of where it appears.
The workflow that actually scales, based on what practitioners have found, is a layered approach. First, use a tool that converts PDF to machine-readable text with layout awareness, something like OCR with coordinate mapping. Then, pass that text to an AI model fine-tuned on invoice documents, not generic language. The model should output structured JSON or a table, which you then load directly into Excel or a modern spreadsheet environment. This eliminates the manual copying and the need for per-vendor parsing rules. The key is that the AI handles the variation; your spreadsheet remains the clean, consistent endpoint.
Our opinion is that you should stop trying to make Excel read PDFs and start making PDFs talk to Excel. The tools exist now, and they are more accessible than building a custom parser for every file type. The next time you open a mismatched invoice, consider whether the problem is the format or the approach. You do not need to learn to speak every vendor's layout language, you need a translator.