Book a demo

For full terms & conditions, please read our privacy policy.
Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.
White plus

How to Use ChatGPT to Extract From PDFs

Miranda Hartley
July 14, 2023

3 steps to straightforward data extraction + summarisation from PDFs using ChatGPT

Initially released in November 2022, ChatGPT revolutionised the mainstream perception of AI. Leveraging natural language processing (NLP) technology, ChatGPT understands the meaning of words, enabling it to extract, organise, summarise and translate information.

Accessing its capabilities, however, requires certain insight. Due to its lack of export function, it’s not immediately obvious how to extract information and data from PDFs.

When asking ChatGPT reflexively how to extract data from PDFs, the response is somewhat convoluted. The bot suggests writing Python code to extract and search for patterns in the data.

Do you have to be proficient in Python to extract information from .pdf files in ChatGPT?

However, there’s an easier way to achieve the same function. We’ll break down these steps into a simple three-step process.

1. Copy and paste the text into ChatGPT.

If your PDF is a scan, you can use a service like Evolution AI's Transcribe platform to convert the image into raw data to copy-and-paste into ChatGPT. However, if you're using GPT-4 (not 3.5), you can directly upload the PDF file to ChatGPT's interface.

2. Prompt ChatGPT with a specific request.

For optimal results, attach a clear and unambiguous prompt. For instance, you can request the data to be outputted in a specific JSON format. Other effective ChatGPT prompts might be:

  • ‘Where on this document can I find x?’
  • ‘What formula in Excel can I use to analyse y?’
  • ‘Can you convert the currency in this document to z?’

ChatGPT will read the PDF data and identify the relevant information.

3. Quality-check the data & amend where necessary

ChatGPT doesn’t have a perfect batting record when it comes to the accuracy of requests. Consequently, some organisations, medical firms, for example, have regulations against involving ChatGPT in their data processing. Therefore, manual validation is necessary to prevent anomalous errors. Once the information’s accuracy has been confirmed, output the data the same way it was inputted: copy and paste.

Summarising information from PDFs

Asking ChatGPT to produce summaries from PDFs is simpler than direct extraction. Condensing data is helpful for long documents containing a variety of information and concepts. Simply paste the data and submit a prompt.

Example applications of this feature might include:

  • Identifying key trends in reports, e.g.‘Sort the transactions on these bank statements into different categories.’
  • Summarising client feedback forms, e.g. ‘Please identify common themes, positive and negative sentiments, and any specific areas of improvement mentioned by the clients.’
  • Highlighting anomalous data, e.g. ‘Please highlight any values in the dataset that deviate significantly from the norm and provide any insights or patterns you observe in the anomalous data points.’

Bonus: Automating extraction or summarisation from ChatGPT

Using an automation tool like Zapier could remove some of the friction from this process.

Example automation flows could be:

  • Importing from Google Sheets and exporting the PDF data to ChatGPT.
  • Extract key information from labelled emails, use ChatGPT to identify important information and then add to a Google Sheet.
  • Using ChatGPT to summarise PDFs and send the summaries in an email, Slack or Mattermost message.

Despite being a landmark achievement, ChatGPT cannot guarantee complete accuracy. Data extracted from documents with complex tables, like financial statements, should undergo quality checks.

Users have reported instances where ChatGPT read PDFs incorrectly: making inaccurate connections between data, fixing non-existent typos, and introducing small errors into datasets.

For a cleaner and user-friendly AI-based data extraction solution, consider Transcribe. Our Evolution Transcribe platform can extract data from images (such as scans of PDFs) without any training data; outputting structured, completely accurate data in a single step. And, like ChatGPT, you can try it for free.

Thoughts? Questions? Get in touch with us at or Tweet us.

Share to LinkedIn