When I perform a search on ChatGPT with the following prompt:
“Make me an SEO optimized title (128 characters max) from the following EAN:
5024763160257
Search for and accurately identify the corresponding product. Include key features, category, and brand (at the end of the title, for example: - Diamwood). Check multiple sources to confirm accuracy.”
It finds the correct information.
But when I try the same thing with a node and Newton, it responds with nonsense. I’ve tried using just the Me model or even the Agent, and it’s the same result—I’m losing my mind…
(same with multiples EAN)
ps: I promise I haven’t started celebrating New Year’s yet!
So, when you use OpenAI you’re using a full app that has more capabilities than the gpt models alone.
When you use the n8n OpenAI node or the AI Agent as you’ve shown you’re using the model alone. And because it doesn’t “know” this information it hallucinates an answer.
In order to get the same result from your workflow you’ll need to also connect a way for it to search the information. The SERP API, for example, which accesses and processes search engines’ information. This way the model will be able to give you a correct and consistent answer, like the OpenAI app does.
Another option would be to use AI assistants. For this, you would need to check your OpenAI account and set up an assistant - https://platform.openai.com/docs/assistants/overview.
After that, in your OpenAI node you’d need to select Assistant as the resource and Message Assistant as the operation.
