--- title: "Announcing Query Logs" author: "Max Irwin" date: 2026-08-17 canonical_url: https://bonsai.io/blog/announcing-query-logs/ license: CC BY-SA 4.0 license_url: https://creativecommons.org/licenses/by-sa/4.0/ copyright: Bonsai.io, 2026 --- We're excited to officially launch Query Logs, available now for all paying Bonsai customers! If you don’t know what your customers are searching for, you can’t improve your search. But, beyond that, search is the purest expression of what your customers actually want from your product. When working with customers who want to improve their search, the first thing I ask is, “Show me your queries.” By this I mean the actual search terms their customers are typing into the search bar. This is the first thing I said to Nick on my very first day at Bonsai in early 2025. I told him that Bonsai customers need this data, so they can understand where their search is falling short and what to fix. Over 18 months ago, during our first week together, we riffed on an early concept of this feature. We’re finally launching it today. Queries uncover _everything_ that’s right or wrong with your product. They tell you: - The things people want or need from your site or app - Trends and reactions to updates - The terminology that your customers use - What content or data you lack - What content or data is the most important - How much people need to dig through to find what they want - How slow your search is - How often things come up empty You can get this all from queries and some basic response stats, even without capturing result clicks! Often, customer queries are things that you didn’t even think of. Three of my separate search relevance war stories involve a product owner telling me that their customers were not searching for something, and then I revealed the query logs and pointed out exactly what they said customers weren't searching for. Astonishment ensued. In general, query logs are used as a priority decision tool - if customers aren't looking for it, then it's probably not that important. If they are looking for it and not getting it, then that's a target to fix! I’d say about 80% of the time, clients don’t capture queries. They might accidentally be in application request logs, but almost impossible to get without a big project. Then 10% of the time, even if queries are purposefully captured, they only exist “somewhere,” requiring digging or coordination with another team. Some teams also rely on tools like Google Analytics for this data, but it’s an incomplete picture, since lots of users block cookies and trackers. So today, Bonsai is rolling out the feature that you won’t find on AWS OpenSearch nor Elastic cloud: Query Logs. ![Bonsai Training Example Dashboard with Query Logs](https://bonsai.io/blog/announcing-query-logs/bonsai-training-example-dashboard.png) You will see queries as an additional and prominent item in your dashboard, with the following columns: - _Query_: The query text - _Volume_: How many times the query was used - _Hits_: How many search results (on average) were matched for the query - _Duration_: How long (on average) the response took The queries are normalized and grouped together, then sorted by frequency. Normalized means that we put similar queries together even though they might have different casing and punctuation. We strip punctuation and lowercase them. ‘Shoes’ will end up the same as ‘shoes’. Let’s break down how they will help you improve your search! ### Improve Relevance By seeing the most common queries, you’ll instantly understand the main interests of your customer base. Good practice is to take those queries and make sure the results that come back for them are relevant! Just run them in your search bar and do a quick check to ensure they all look good. In addition to the most common, also look for interesting or unexpected queries that jump out at you, and search for those too! Make sure you're seeing good results, and tune if you need to. In the long term, we recommend using a tool for repeating this and getting a relevance metric such as nDCG or ERR for understanding quality improvement. ### Fix language or content gaps Now, look at the queries that have zero hits on average - we call these ZRS or “zero result searches”. These tell you that people aren’t finding things for certain terms. Maybe you need to add some synonyms or integrate [vector search](https://bonsai.io/blog/when-to-use-vector-search/), and maybe you even need to modify or add some content! For example, if people are searching for “support” and not getting hits, make sure that your support page is included in the index. ### Improve performance Response duration is less helpful for relevance, but if this number is above 500 milliseconds we recommend taking action to improve speed. Spoiler alert: we have another feature in the works that will help diagnose this a bit better :) ## Practical implementation Theoretical “here’s what to do” is helpful advice, but I’m dogfooding our own feature and I’ll now give you concrete examples. If you remember some of my previous posts [OpenSearch and Elasticsearch on Render with Bonsai](https://bonsai.io/blog/managed-search-on-render-with-bonsai/) and [Agentic Search with OpenSearch and Elasticsearch](https://bonsai.io/blog/agent-search-with-sid/), I used a Bonsai cluster “bonsai-training-examples.” Here are the actual query logs for the past 7 days: ![Bonsai Training Example Query Logs](https://bonsai.io/blog/announcing-query-logs/bonsai-training-example-query-logs.png) Notice anything? The Hangul (Korean) queries are all giving ZRS! This is where I’d now turn to implement vector search for the corpus using a multilingual model, such as [Qwen3-Embedding-0.6B](https://huggingface.co/Qwen/Qwen3-Embedding-0.6B), which is what we offer for enterprise customers in our fully integrated vector search component. It supports over 100 languages including Hangul. I’ve added this one to our backlog (and of course, as a future blog post!) The second thing I did was run the most popular query, `japanese short stories`. Here are the top 4 results. ![Japanese War Stories query results page](https://bonsai.io/blog/announcing-query-logs/japanese-short-stories.png) _Results for "japanese short stories"_ Notice, the 2nd result is Chinese, not Japanese! That’s another item for the backlog - tune the relevance to improve the outcome of this search, as I want at least the top four of the results to be relevant. ## Getting all the data Depending on your search deployment, you likely have lots and lots of queries - far more than we show in the dashboard. To get the full set of queries for the time period, you can Export as CSV. For now, this will email you an expiring download link. We don't keep all this data forever. We aggregate for the past 7 days and then we need to let it go. So for now if you want to do long term analysis on queries, then you'll need to download these and keep them for later. We're still working on refining this process, and we're open to feedback - so reach out about your use case and we'll gladly take it into consideration on how to improve. ## Try it out! We’re really excited to launch this feature. It’s the first thing I reach for in my bag of tricks for search quality improvement, and I am delighted that it’s now in all of yours. _Copyright ©️ Bonsai.io, 2026 · By Max Irwin · Originally published at https://bonsai.io/blog/announcing-query-logs/ · CC BY-SA 4.0_