How to Delete or Update Specific Training Data
Why You Need to Update Training Data
Training data is not something you set and forget. Business information changes constantly: pricing adjusts, policies update, products launch, staff changes, procedures evolve. Every change in your business that affects what your chatbot says requires a corresponding update to the training data.
If you do not update, two problems emerge. First, the chatbot gives outdated answers that may frustrate or mislead customers. Second, if both old and new information exist in the knowledge base simultaneously, the AI may blend contradictory information into confusingly wrong responses.
How to Delete Training Data
In the admin panel, open the AI Chatbot app and select the chatbot whose training data you want to modify. Navigate to the knowledge base or embeddings section.
Your training data is stored as individual chunks. Each chunk has a tag (the label you gave it when uploading) and a preview of its content. Browse or search for the chunks you need to remove.
Select the chunks and delete them. This removes both the text content and the vector embeddings from the database. The deletion is immediate, and the chatbot will no longer retrieve that information when answering questions.
How to Update Training Data
There is no "edit in place" function for training data chunks because changing the text requires regenerating the vector embedding. The update process is delete and re-upload:
Follow the deletion steps above to remove the old content.
Write or gather the new version of the information. Make sure it is current, accurate, and properly organized. See How to Organize Training Data for Best Results.
Upload the updated text as new training data. The platform will chunk it and generate new embeddings. Use the same tag as before if you want to keep your training data organized by the same labels. The cost is 3 credits per new chunk.
When to Update vs When to Add
Not every change requires deleting old data:
- Information changed: Delete the old version and upload the new one. Old pricing replaced by new pricing, updated return policy, changed contact information.
- Information expanded: If you are adding more detail to an existing topic without changing the facts, you can just add new chunks. The old chunks are still accurate, they are just less detailed.
- New topic entirely: Just upload the new content. No deletion needed since there is nothing to conflict with.
- Information removed: If a product, service, or policy no longer exists, delete the training data about it so the chatbot stops referencing it.
Bulk Updates
If you need to update a large portion of your training data (a full website recrawl, a new version of your documentation), it is often easier to delete all chunks with a specific tag and re-upload everything fresh. Tags make this manageable because you can delete by tag rather than hunting for individual chunks.
Verifying Your Updates
After updating training data, test the chatbot with questions that should now produce different answers. Ask the specific questions that prompted the update in the first place. If the chatbot still gives the old answer, check that the old chunks were actually deleted and the new ones were uploaded successfully. See How to Test If Your AI Learned the Right Information.
Keep your AI chatbot current with easy training data management. Update anytime, no coding needed.
Get Started Free