Back in 2023 I tested whether ChatGPT could write Airtable formulas. The results were mixed. Sometimes it got the answer right, but often it produced something that looked correct while being completely invalid. Two years later, I wanted to revisit the question. Has ChatGPT improved, and can I trust it more now when I am working with Airtable?

To answer that, it helps to understand a concept that has become central to AI use: hallucinations.

What are AI hallucinations?

An AI hallucination is when a model generates information that sounds entirely convincing but is simply not true. The word comes from psychology, but in AI it means producing confident but incorrect outputs. Hallucinations are not the same as typos or small mistakes. They are polished, logical-looking answers that can mislead you into thinking they are accurate. They are confident lies.

In the context of Airtable, this might mean ChatGPT inventing a function that does not exist, such as suggesting Excel’s XMATCH or LET will work in a formula field. The formula reads well, the syntax looks plausible, and ChatGPT explains it in detail, but when you paste it into Airtable you're met with an error. That is a hallucination in action.

How far has ChatGPT come since 2023?

There's been real progress. ChatGPT is now much better at simple Airtable formulas. Date functions, conditional logic, and text handling are usually correct. If you are new to Airtable and ask for a basic formula, the output is far more likely to work than it was two years ago.

The challenge is that hallucinations have not gone away. ChatGPT still sometimes suggests functions that belong to Excel or Google Sheets, not Airtable. It does not warn you that these functions are unsupported. Instead, it delivers the answer with total confidence. Unless you already know Airtable’s formula library, it is easy to waste time trying to fix something that never stood a chance of running.

This is why hallucinations are so tricky. They exploit our trust. Because the answer looks right, we assume it is right.

Why Airtable AI is a safer choice

Since 2023 Airtable has released its own AI features, including a formula assistant. This has shifted the balance. Airtable AI does not hallucinate because it is built on Airtable’s own formula set. If you ask for a calculation, it can only give you what Airtable actually supports.

This makes it far more reliable for everyday use. With Airtable AI I do not have to wonder whether a function really exists. I can copy and paste the result straight into my base and know it will run. That reliability is what sets it apart from ChatGPT in this specific task.

Where ChatGPT still adds value

That said, you can still use ChatGPT alongside Airtable AI, just not as the primary tool for formula creation. Its strength is in exploration and explanation rather than precision.

• Breaking down formulas. If I have a long nested IF, ChatGPT is excellent at turning it into plain English.

• Exploring different approaches. It is useful for brainstorming possible ways to solve a problem, even if you need to adapt the details afterwards.

• Learning from other platforms. Sometimes I want to see how Excel or Sheets might handle a similar calculation. ChatGPT can give you that broader context, even if the formula itself is not portable.

As long as I treat its outputs as suggestions rather than truth, ChatGPT remains a helpful thinking partner.

Lessons for 2025

The theme here is simple: hallucinations are still with us. ChatGPT has improved, but it still makes things up in convincing ways. Airtable AI avoids this problem entirely, because it is built on a closed set of real functions.

So my advice is this:

• Use Airtable AI when you need a working formula you can rely on.

• Use ChatGPT when you want ideas, explanations, or comparisons, but always double-check the output.

  
  

This combination gives you the best of both worlds. You get creativity and inspiration from ChatGPT, and accuracy and reliability from Airtable AI.

Where to from here?

In 2023 I described ChatGPT as promising but flawed for Airtable formulas. In 2025 I would say the same, only with more emphasis on hallucinations. They remain the Achilles’ heel of using general-purpose AI for specialist tasks.

For me, the future lies not in choosing one tool but in knowing their strengths. ChatGPT is the idea generator. Airtable AI is the truth checker. Together, they make formula writing faster and less painful, as long as you keep one eye open for hallucinations along the way.