Tuning uses AI to train your search algorithm to prioritize the most relevant content for your visitors. By learning from real user behavior and your training data, it continuously improves the accuracy and relevance of every search.
Use both your expertise and real data, such as click-through rates, to improve site search relevance. Input specific search queries and their ideal results to teach the algorithm what your visitors are looking for. Meanwhile, when users click specific results more frequently for a given query, the system interprets this as a relevance signal and adjusts ranking weightings automatically. Over time, click-through behavior improves how similar searches are ranked.
AI technology tests various combinations of over 70 ranking signals (content freshness, authority, document type, keyword relevance) to find the optimal settings for your site search algorithm. One training session automatically improves similar searches across your site.
Video: How do you input training data? . Captions and transcript available on playback.
See how you can input search queries and expected results to train your algorithm. No search specialist needed.
Squiz Funnelback Search uses over 70 ranking signals to determine the order of search results, including content freshness, authority, document type, and keyword relevance (you can see the full list of ranking signals here).
While you can manually adjust these ranking signals, it's not recommended. When people try to manually tune search, they often test with just a handful of queries and tweak only a few settings. What fixes search results for one query often breaks them for dozens of others. Our automated tuning system tests hundreds of different combinations to find the optimal settings across your entire site, delivering much better results than manual adjustments.
We recommend starting with 50-100 training examples for best results. This gives the automated tuning system enough examples to systematically explore different parameter settings and optimize performance across your entire search.
Most organizations begin with their most common search queries (the ones that represent 80% of user searches) and gradually expand from there. The beauty of tuning is that each training example you provide helps improve not just that specific search, but similar queries throughout your site.
Yes. The system automatically tracks anonymized click-through behavior and engagement patterns. When users consistently select certain results for a query, the algorithm interprets this as a positive relevance signal and automatically adjusts ranking weightings. This process is powered by Learning to Rank (LTR), a machine learning approach that analyzes both user behavior and training data, creating a continuous feedback loop that improves search accuracy over time.
Yes, tuning includes preview capabilities so you can see how algorithm changes will affect search results before publishing them. You can run test queries to see the difference between current results and improved results, helping you validate that the training is working as expected. This preview function lets you refine your training data and ensure you're happy with the improvements before they go live for your users.
