Trouble Passing Uploaded PDFs into ChatGPT Conversation Step – Unexpected TXT Files & Failures

Hi ​@lwhathaway 

Help us have true context by posting screenshots showing:

• how your Zap steps are outlined
• how your Zap steps are configured in EDIT mode in the CONFIGURE tab with the field mappings visible
• any encountered errors from testing Zap steps

Thanks, Troy! Screenshots attached and the code from the code step pasted below: 

 

// Extract attachments from input

const attachments = inputData;

// Helper function to normalize inputs

function toArray(...items) {

return items.filter(i => i !== undefined && i !== null);

}

// Collect matching PDFs

let pdfUrls = [];

let pdfNames = [];

for (let i = 1; i <= 5; i++) {

const mime = attachments[`mime${i}`];

const name = attachments[`name${i}`];

const url = attachments[`url${i}`];

if (mime && mime.toLowerCase() === "application/pdf" && url) {

pdfUrls.push(url);

pdfNames.push(name || `attachment-${i}.pdf`);

}

}

// Return up to 5 URLs and names individually

return {

foundPdf: pdfUrls.length > 0,

fileUrl1: pdfUrls[0] || null,

fileUrl2: pdfUrls[1] || null,

fileUrl3: pdfUrls[2] || null,

fileUrl4: pdfUrls[3] || null,

fileUrl5: pdfUrls[4] || null,

fileName1: pdfNames[0] || null,

fileName2: pdfNames[1] || null,

fileName3: pdfNames[2] || null,

fileName4: pdfNames[3] || null,

fileName5: pdfNames[4] || null

};

The downloaded_file.txt issue happens because Zapier sometimes sends a text wrapper instead of the actual PDF binary when pulling Gmail attachments. Even if you filter by MIME type, Gmail often just gives Zapier a link or base64 data, so the OpenAI Upload step treats it as text. The fix is to normalize the file before upload: either use Zapier’s “Formatter → Utilities → File” step on the Gmail attachment, or in your Code step explicitly return the file with .pdf as filename and application/pdf as MIME type. That way the OpenAI step sees a true PDF and won’t generate .txt placeholders.

Hey ​@lwhathaway 👋 Just came across this and wanted to check how you’re getting on, were you able to get this solved?

If not, it looks like you’ve got file object rather than file URLs mapped to the url fields on the Code action: 

Since the Gmail trigger provides the data for each of the attachments in differently numbered fields could you maybe get rid of the Code step and map those fields directly from the Gmail trigger into the relevant ChatGPT actions?

Looking forward to hearing from you!