OpenAI Chat Model return 'I don't know' even if there is results

Describe the problem/error/question

I trigger the flow with a chat question : I am looking for a light wood table
In the input of OpenAI Chat model we can see that some results have been found. (check input below)
but the output returns : I don’t know
It seems that the input is not well processed.

I don’t really understand why results are not well processed.

Appreciate you help.

What is the error message (if any)?

No error message just return I don’t know

Please share your workflow

Share the output returned by the last node

Input in OpenAI Chat Model :
System: Use the following pieces of context to answer the users question.
If you don’t know the answer, just say that you don’t know, don’t try to make up an answer.

Simplicity, versatility and functionality are three of the characteristics that define the Royal Square Beech Wood and MDF Dining Table (80x80 cm). This designer table is made up of a beech wood structure and an MDF top. In addition, the lower part of the legs is equipped with non-slip and anti-scratch studs that will allow you to use it on any type of floor. What are you waiting for to fill your home or establishment with good vibes? Let yourself be carried away by the Nordic style and fill your living rooms, dining rooms, kitchens or any other establishment with good taste and refinement. It is recommended to clean it with a damp cloth and avoid the use of chemicals.

A true icon of Scandi style. It perfectly combines design and comfort. Classic and very elegant. Composed of beech wood legs and an MDF top. Ideal for dining rooms and living rooms. Indoor use only.

If you are looking for a spacious and functional table, the Rectangular Folding Dining Table Iron in Mango Wood and Metal (200x100 cm) will look great in your dining room or living room. Made of 100% natural mango wood, a robust material, together with its resistant iron legs, this combination of quality materials guarantees greater durability for daily use. It has a folding system that allows you to keep your spaces clean and tidy when not in use. Includes some cleats to prevent damage to the floor and ensure greater stability. Mango wood can have a wide variety of shades due to its characteristics. We can find everything from brown or golden tones, through yellow and red, and you can even find darker parts with gray or black tones. This range of colors is part of the natural beauty of this type of wood and makes each piece of furniture unique, and there may always be differences in tone and color from one to another. For greater durability of the product, follow the specific maintenance instructions for each type of wood.

The Gudda coffee table is a piece of furniture with a very elegant and original design that is committed to mixing textures and natural materials, in order to obtain an innovative and elegant piece, in addition to being functional. It is made of ash wood, a 100% natural material, very resistant with an oval MDF board. Give freshness and color to the most insignificant corners of your home such as the living room, dining room, office or bedroom. Given its great versatility, you can integrate it into various styles of decoration such as Nordic or natural. For perfect care and maintenance of the material, at SKLUM we recommend cleaning it with a damp cloth, thus avoiding the use of any chemical product.
Human: I am looking for a light wood table

Output of OpenAI Chat model :
I don’t know

Information on your n8n setup

  • n8n version: 1.71.3 Cloud Version
  • Database (default: SQLite): Postgres
  • n8n EXECUTIONS_PROCESS setting (default: own, main): default
  • Running n8n via (Docker, npm, n8n cloud, desktop app): n8n cloud
  • Operating system: MacOS

Hi @jboo2412

Thanks for posting here and welcome to the community! :cupcake:

The Question & Answer Chain node is built on simple text completion generative AI functionality. This means that it is not designed to be capable of reasoning.
The instructions you give are "Use the following pieces of context to answer the users question." but then the user doesn’t ask a question: "I am looking for a light wood table."

It’s not “intelligent” enough to infer what the user wants or ask questions back, so it will simply do what it has been instructed, when it doesn’t know: "If you don’t know the answer, just say that you don’t know, don’t try to make up an answer."

A few things you can do:
Either you build your workflow in a way that only presents questions to the Q&A chain (like it’s designed) or try in the system message to instruct it to something like: "Ask the user what they want to know if you don't understand their question."

Or you could use the AI Agent node instead, which is uses different generative AI functionality that has more reasoning and can process inputs better by creating a more comprehensive context.

I would highly recommend our tutorial on AI Agents here as a learning resource :slight_smile:

Hope this helps! :raised_hands:

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.