Search FAQs

Why am I seeing a warning message that says, "There might be an issue requiring attention. Contact support for more information"?

This warning appears when our system can't complete the training process for your site's search functionality. The most common causes are:

Image Access: Our AI model needs to download product images from your CDN. This is a one-time process during the initial catalog sync. To ensure it works, add the following IP addresses to your allowed list:

3.92.38.28 3.127.189.219 3.127.189.219

3.127.189.219 3.227.221.249 18.158.108.19

18.192.189.87 18.195.163.252 18.197.234.165

54.145.37.112 75.101.154.167
Image Size: Product images must be smaller than 10 MB.
Image URL Path: Image URLs must be explicit. For example, use https://mydomain.com/myimage.jpg and not //mydomain.com/myimage.jpg or //myimage.jpg.
Missing Images: This message is triggered if our system can't access more than 15% of your product images (due to a 5-second timeout).

If you've addressed all of the these guidelines and the warning persists, try resyncing your product feed a few times. If the issue is still unresolved, open a support ticket so we can check whether it's due to an internal error on our end.

Should I implement the Experience Search API from the client-side or from my server?

The decision to implement the API from your server or the client is a technical one, and we fully support both methods. We provide different API keys for each implementation method.

Here's a quick summary to help you and your technical team decide:

	Server-side	Client-side
Best for	Security, performance, and SEO.	Faster implementation and reduced server load.
Key advantage	Your API key is kept secure on your server.	Has direct access to a user's device and browser info.
Considerations	Requires more development work on your server.	The API key is exposed in the front-end code.

Ultimately, the best choice depends on your team's technical skills and your company's security priorities.

Why am I not seeing clicks and revenue data in my reports, even though search data is present?

This indicates that the slot IDs are not being correctly passed from the search response to your listing pages. The slot ID is a unique identifier that connects a user's post-search actions (like clicks and purchases) to the original search query.

To fix this, review your Engagement API implementation. Ensure the correct slot ID for each product is passed to and wrapped around the corresponding items on your listing page.

Why am I not getting trending results in the Suggest API?

Trending terms are based on an aggregation of your users' search queries. It takes time for the system to learn and recognize these patterns. Trending terms are updated daily with a sliding window of 30 days, and depending on your site's traffic, it can take anywhere from 1 to 14 days for trends to appear across all of your locales. Be patient, the system is still learning in the initial days of your implementation.

Why are search response times high during tests in staging or preproduction environments?

Our servers allocate more AI computation resources to live, active environments. An "active" environment is one that has at least one request per minute. Latency is only noticeable in environments with no user activity. This is expected behavior.

In addition, our system caches responses for common queries. In a live production environment, users often search for the same things, and in 60-80% of cases, this caching cuts serving times by 80% or more. This is a primary reason for the performance difference you see.

To improve performance for your tests, do the following:

For load testing: Run a warm-up test with several queries before starting your official test.
For manual testing: Manually run several queries before proceeding with your planned tests.

Why is the search response time higher when I use a locale in the API?

We use a large language model (LLM) to translate user queries and your product catalog. While the catalog translation is done offline, user queries are translated in real time. This allows us to handle most languages while keeping our AI search contextual.

After a query is translated, it's cached, so subsequent searches for that same query are much faster. During the initial days of implementation, long-tail queries might have slightly higher latency until the system gathers and caches enough queries. Rest assured, this improves over time.

How long does it take until my Experience Search app is ready?

The training process for your catalog, which includes analyzing your images and product data, takes place only once when you first open the application. The time it takes depends on your catalog size and can range from 10 minutes to several hours.

How does Experience Search handle synonyms?

As a native-AI search solution, Experience Search models are pre-trained to understand synonyms across various industries and languages. This built-in intelligence removes the need for you to manually manage synonyms or even apply AI synonyms suggestions as required in other solutions, ensuring a seamless experience for your users.