Sorry I also get an error when I try to use my own api key. Tried a couple of different ones
options-general.php?page=ai-search&embeddings_generated=true&tab=embeddings:1063 Uncaught TypeError: $(…).pointer is not a function
at options-general.php?page=ai-search&embeddings_generated=true&tab=embeddings:1063:43
Hey @jamesc91 , thank you for your words, I’m glad you’re enjoying the plugin.
At this moment, we don’t have a specific limit so sometimes it’s challenging to find the exact accuracy level. Besides the title, do you also have description and content related to each page? Have you also checked the Custom Fields tab? In there you can choose which fields you can include to generate the similarities. When you have more information, the results will be more accurate and you’ll be able to increase the accuracy limit.
I don’t think that error in specific is related to AI Search, but if you want me to take a look, please share some screenshots and I’m going to try to reproduce from my side.
Thank you!
Samuel
Thanks Samuel I managed to get it working. When you switch from the plugins free plan to your own api you need to fully clear all embedding data for it to work.
Do you know how I can check to see what the embedding data looks like and make sure it includes all the attributes that have been selected? I have selected to include descriptions, all attributes etc but it still comes back with poor results mostly related to the title of the product only, not taking into account attribute data.
We have a lot of data per product so it should be accurate. For example
“wall light with a glass shade” = no results. Even though we have category = wall lights + attribute material = glass.
Glass wall light = lots of results as glass wall light is in the title of the product.
I have been testing different weights of thresholds but that doesn’t change it either.
Hey, thank you for the feedback. I’ve just released version 1.16.0 with improvements that directly address what you’re experiencing (I think most of them will help you.
What’s new:
- Case-insensitive search – All content and queries are now lowercase, eliminating case-matching issues
- Embedding preview – Edit any product and check the “AI Search” meta box → “View Embedding Content” to see exactly what’s being indexed (title, categories, attributes, etc.)
- Better threshold handling – More consistent filtering across all search types
- Similarity Score Display – If you’re logged in as an admin, you’ll now see similarity percentage badges on each search result, helping you understand how well each result matches the query.
have a try and let me know the results!
Hey Samuel,
Thank you very much.
I see the issue now with the data – Only the product title and and product description are being indexed in the embedding data even though this is selected.
In the admin of the plugin I have selected in woocommere: categories, sku, 20+ attributes but none of this data is being being synced through. I tried to regenerate this for a couple of products using the button you kindly added but still only syncs the title + description.
Thanks again
James
Hey @jamesc91 ,
In this new version (1.17.0), I added a new feature called Hybrid Search that lets you combine AI semantic results with traditional keyword matching. You can adjust how much weight each one gets in the final results. Hope you find it useful!
Hi Samual,
Thanks for the update.
Getting better results but it still appears to just be pulling the title and description and the woocommerce section of the plugin where you tick is not working. It doesn’t understand when searching for attribute specific. Category + attribute.
I have a test site with 1000+ products. Would it help you at all giving you access?
-
This reply was modified 2 months, 2 weeks ago by
jamesc91.
-
This reply was modified 2 months, 2 weeks ago by
jamesc91.
-
This reply was modified 2 months, 2 weeks ago by
jamesc91.
@jamesc91 in this last version, I made some updates for the users that are using their own OpenAI key.
Instead of text-embedding-ada-002 AI model, we’re now using text-embedding-3-small, which provides better results. You also have an option in your settings to switch to text-embedding-3-large, the most accurate model for AI search.
Enjoy!