Search as a Product

Search That Works: How We Fixed Discoverability

At the AMA, I discovered a major disconnect between how search was managed internally and what users actually expected. While users assumed they could search the company website and find relevant content from across all AMA platforms, the reality was very different. Each product and service operated independently, and no one owned the overall search experience. As a result, search had become fragmented, inconsistent, and easily influenced by internal priorities.

Role: Director of Product Management

Team: 5

The Problem

AMA had dozens of separate digital products. Users expected a unified search experience, but behind the scenes, each product team was making independent decisions. When executives received complaints that their content wasn’t surfacing, they would request higher weighting for their material. Over time, this created a disorganized system where search results were skewed and unreliable.

AMA website showing 429 search results for “CPT,” including filter options like news articles, documents, and videos. Demonstrates the structured output of unified search indexing.
Backend view of the Best Bet configuration panel, showing search logic for contact-related queries. Highlights use of “exact match” and “search contains” fields to trigger relevant top search results.
Front-end example of a Best Bet in action on the AMA website. A search for “address” triggers the Contact Form as the top result, using configured Best Bet logic for improved discoverability.

The Reset

We started by clearing out the existing search index and launching a clean instance of Apache Solr. Technically, this was easy. Politically, it was more complex. I had to work closely with executive leadership and product owners to build confidence in the new approach and explain how a unified search strategy would lead to better visibility and discoverability for everyone’s content.

Defining What “Good” Search Looks Like

To rebuild trust in the system, we created a simple framework for measuring search quality.

We reviewed the top 50 search queries and mapped them to existing topics. If we found content that matched the intent but performed poorly, we updated the on-page copy, context, or metadata, then retested. This process avoided manipulating Solr directly and instead focused on improving the content itself. It also helped improve our external SEO and organic traffic.

When a top query didn’t map to an AMA topic, we evaluated whether it represented a strategic content opportunity. This gave us early insight into emerging areas where we could create new content, often capturing valuable organic traffic with high engagement.

Negative Indicators:

  • Searches that ended with no click (excluding Best Bets or search connector use)

  • Searches that were refined without any click

Positive Indicators:

  • Searches that resulted in a click, including those assisted by Best Bets or connectors

Enterprise search results page demonstrating intent-based ranking, filtering, and structured results to help users quickly find relevant content at scale.
Mobile search results experience demonstrating intent-based ranking, filtering, and structured content to help users quickly find relevant information on smaller screens.
Enterprise search results page demonstrating relevance-based ranking, faceted filters, and mixed content types designed to help users quickly narrow and resolve COVID-19–related queries.

Search Contains Logic

We used Apache Solr’s Best Bets feature to deliver more intuitive results. If a search query didn’t match a Best Bet’s exact terms, the system would look at the “search contains” field for related matches.

  • If multiple Best Bets matched, the oldest one would be prioritized.

  • These matches appeared at the top of the search results, similar to promoted answers on Google.

  • Bet’s exact terms, the system would look at the “search contains” field for related matches.

This allowed us to guide users effectively, even when their search phrasing was unclear or non-standard.

For Example:

  • A query like “address” would trigger the Contact Us page.

  • A less direct query such as “I’m trying to contact someone” would also trigger the same result.

Universal Search

The long-term goal was to give users a consistent search experience across all AMA products. We created a Universal Search experience using APIs to connect independent platforms. When a search term had matches on a partner site, we displayed a callout on the results page showing how many results were found on those platforms. This helped users discover more without needing to search multiple sites separately, and it supported a more connected ecosystem across AMA’s offerings.

AI in Search

As AI technologies evolved, our methodology stayed the same. We continued to monitor behavior, adjust content, and ensure users could find what they were looking for.

  • Semantic search to better understand user intent

  • Predictive autosuggestions based on behavior patterns

  • LLMs to experiment with conversational interfaces and summarization

The goal remained consistent: deliver smarter, faster, and more relevant results that supported both users and business goals.

Previous
Previous

Platform Transformation

Next
Next

Sign-In & Identity