When should you use ChatGPT and when should you use the API version?
Ray Poynter, 9 January 2025
This post focuses on ChatGPT, but the same argument applies to any pairing of a Generative AI platform and its API complement.
The great thing about web-based ChatGPT is that it does not require much preparation, assuming you have sufficient skill and knowledge to drive it well. If you have a task that would benefit from ChatGPT’s assistance, you can log in and get working. However, this flexibility comes at the expense of two key features 1) it can require considerable topic knowledge from the user (limiting who can use it for a specific task), and 2) it does not have any guardrails, it can make big, plausible mistakes.
An alternative to the ‘regular’ ChatGPT is to use the ChatGPT API. An API is a programming interface. You can create your own programs that, as part of their function, can call ChatGPT. This is slower to set up and requires some programming skills, but it creates solutions that can be more efficient, easier to use, and safer.
In this post, I look at when to favour one over the other and when to look for a third alternative.
When is ‘regular’ ChatGPT best?
By regular ChatGPT, I mean ChatGPT accessed via a browser, probably through a paid-for licence. The key clues to its suitability are:
- Ad hoc problems. If you are dealing with one-off, unique situations, then the flexibility of ChatGPT is what you need. It would be uneconomic and time-consuming to create your own routines to do tasks that are one-off or occasional.
- Skilled users. If the task being undertaken is within the skillset of the user, for example an experienced researcher writing a customer satisfaction questionnaire, then the speed and flexibility of ChatGPT can boost productivity and can reduce errors.
- Lack of coding skills. If there are no available skills to create routines calling the API, the API route becomes even slower and more expensive.
When is the ChatGPT API best?
The key clues to when the API route is preferred are:
- High volume and automation. The API route will help you automate tasks and widen the pool of people who can perform them.
- Adding AI to existing processes. If you already have tools that perform key tasks, adding AI (via the API) will often be better, rather than trying to create a whole new system.
- Technical expertise. If you have coders then the API route is more accessible to you.
What about other solutions?
There are (at least) two other options that may be better than traditional ChatGPT or API routes. The first is existing, non-AI solutions, and the second is Custom GPTs.
For advanced analytics, it may be better to use tools like Displayr, SPSS or R. For visualisation you might be better off using Tableau, for Conjoint Analysis try Sawtooth Software. These tools are powerful, and they have guard rails. When using these tools, you will probably find it helpful to have ChatGPT open to ask it to remind you how to run various options and what alternatives to consider.
Custom GPTs are a halfway house between regular ChatGPT and using the API. You don’t need to be a coder to create Custom GPTs (although a coding mindset is useful). A Custom GPT is created and trained to boost ChatGPT’s ability and, at the same time, constrain it. For example, you can teach it your style preferences for a questionnaire and constrain it so that it does not use other options. Custom GPTs can be really useful in automating tasks such as proposal writing, analysis, and client-facing tools.
Want to learn more?
With ResearchWiseAI, I am running several courses this year, including