Working with Spreadsheets & ChatGPT: What I’ve Learned (So You Don’t Have To)
Also Deep Research is here!
Spreadsheets, data cleanup, and AI—sounds like a dream team, right? Well, not always. While ChatGPT can be an incredible tool for automating and organizing data, it has its quirks, especially when it comes to order of operations, formatting, and large datasets. This week, I’m sharing some key lessons I’ve learned from using AI with spreadsheets, plus a look at ChatGPT’s recent weird behavior, and an exciting update: Deep Research is here!
If you’ve ever tried working with spreadsheets, data, and ChatGPT, you’ve probably run into a few unexpected surprises. While AI can make data cleanup, analysis, and automation easier, it also has some quirks. Here are a few tips & insights I’ve picked up along the way:
Order Matters (Ask in the Right Sequence!)
ChatGPT follows a logical order of operations, so the way you structure your request changes the output. If you ask it to sort data before cleaning it, you might end up with a mess. Instead, break it down:
Step 1: Normalize or clean the data (fix errors, standardize formats).
Step 2: Organize and categorize it (sort, filter, group).
Step 3: Run calculations, summarize, or extract insights.
Example: If you’re working with messy names & emails, clean duplicates FIRST before sorting or extracting key data.
Normalize Your Data (It’s Not Just an IT Term)
Before you start analyzing or automating, you need consistency—this is called data normalization. ChatGPT can help, but you have to guide it!
Addresses? Convert "123 Main St." and "123 Main Street" into a standardized format.
Dates? Make sure everything follows one date format (YYYY-MM-DD is safest).
Case Issues? Ensure all names are Title Case or UPPERCASE for uniformity.
Pro Tip: ChatGPT can convert mixed data formats into a clean, structured format if you give it a few clear examples first!
Be Specific About Data Types
ChatGPT doesn’t always automatically recognize data types the way Excel or Google Sheets does. If you need:
Numbers, specify integer or decimal (so it doesn’t turn "$100.00" into “100”).
Dates, clarify MM/DD/YYYY vs. DD/MM/YYYY (because ambiguity = chaos).
Categorical data, tell it to group similar items instead of just listing them.
Example: If you ask ChatGPT to “analyze sales trends,” it might just describe the numbers. Instead, say “Calculate the percentage growth by month and highlight the highest/lowest sales.”
Watch Out for Hidden Formatting Issues
ChatGPT doesn’t always “see” invisible issues like extra spaces, missing commas, or inconsistent line breaks—but these can break formulas when you import data back into Excel.
Ask ChatGPT to trim whitespace & remove extra spaces before processing.
If numbers aren’t working in formulas, they might be stored as text—convert them before running calculations!
Pro Tip: If a spreadsheet formula isn’t working, ask ChatGPT to check for hidden characters or format mismatches.
Large Datasets? Break It Down
ChatGPT has token limits, so it can only process so much data at once. If you’re working with thousands of rows, try:
Asking for a summary first before diving into specifics.
Breaking the dataset into smaller chunks and processing in parts.
Using Python, Excel formulas, or Google Sheets scripts for heavy lifting instead.
Example: Instead of “Analyze this 10,000-row dataset,” try “Summarize key patterns in the first 500 rows” and iterate from there.
What’s the weirdest spreadsheet challenge you’ve faced? Reply and let me know, I love a good AI data puzzle!
Notice Any Weirdness with ChatGPT Last Week?
If you've noticed that sometimes ChatGPT is just plain weird? it's not just you. Last week there was an update that changed things but no one knows exactly what.
Some custom GPTs are broken or like me, people noticed things. not finishing, being offered a file to download (with the expected outcome) but no file link, or an uploaded file would go missing, can you upload it again?
Many specialized custom GPTs (like AI illustrators) were completely broken.
No official documentation or warning was provided about the changes.
Sometimes it goes off the rails. Reminds me of Google when the algo would change, is this the ChatGPT version of a Google dance?
Deep Research: AI That Actually Thinks
Deep Research is here! Previously for Pro users, Plus users can now use this tool. Ever wished you could hand off a complex research task and get back a clear, synthesized answer without spending hours digging through sources? That’s exactly what Deep Research, powered by OpenAI’s upcoming o3 model, is designed to do. Unlike standard AI responses, this agent uses advanced reasoning to analyze large amounts of online information, connect insights, and complete multi-step research tasks—all while you focus on more important things.
Deep Research isn’t instant—it works more like an AI research assistant, taking anywhere from 5 to 30 minutes to compile and refine its findings. And since this kind of processing power is resource-intensive, usage is currently limited to 120 runs per month for Pro users and 10 for Plus users, with a rolling count that resets based on your first use.
I'm trying it as I finish this newsletter and can't wait to see what it comes back with. My first query was met with some clarifying questions which I answered and now it's chugging away. It's fun to see the Activity and Sources update as it works. It’s looking good!
Until next time!
AI is evolving fast, and sometimes that means unexpected changes, hidden quirks, and brand-new tools. Whether it’s getting ChatGPT to play nicely with spreadsheets, dealing with surprise updates, or testing out Deep Research, there’s always something new to explore. Have you noticed anything strange in ChatGPT lately? Or maybe you have a go-to spreadsheet trick that AI has helped with? Hit reply—I’d love to hear your thoughts!
Lisa