AI-powered legal search helps 62% more users find help on ILAO

Challenges

Improving access to legal help online presents unique barriers for organizations serving people unfamiliar with legal systems. 

Users know what problem they're facing (an eviction notice, unpaid wages, a custody dispute), but not the legal terms or categories that would help them search effectively. Keyword-based search failed them because it required the very vocabulary they lacked.

ILAO previously offered live chat services, but the costs weren't sustainable. Duplicate and misclassified content also made it harder for users to find reliable, actionable information.

Accuracy posed a separate challenge. Off-the-shelf AI models lacked legal specialization and risked generating responses not grounded in current Illinois law or verified sources.

Approach

We began by observing how people searched for legal help and where they struggled. This informed our decision to choose a retrieval-based approach anchored in ILAO's verified legal content, allowing responses to remain reliable, traceable, and easier to improve based on user feedback.

Design and engineering moved in parallel. User behavior shaped the conversational interface. Search refinement focused on reducing friction for users unfamiliar with legal terminology.

Our goal was to ensure every response reflects current Illinois law and links back to articles reviewed and approved by ILAO's legal team.

Solutions

Semantic search

A semantic search engine replaced ILAO's keyword-based system. Instead of matching keywords, it interprets what users mean and returns results based on legal relevance. Users can describe their situation in plain language (e.g., "my landlord won't return my deposit") without knowing the correct legal term. The system draws only from ILAO's reviewed legal content and supports both English and Spanish.

Conversational assistant

A conversational assistant provides plain-language responses through a chat interface. Users ask questions and receive structured answers without navigating multiple pages.

The assistant uses a retrieval-augmented generation (RAG) architecture: it retrieves relevant content from ILAO's legal articles, then generates a response grounded in that material. This ensures responses are traceable to verified sources rather than generated from the model's general training data.

Backend and infrastructure

The AI systems required infrastructure that could process ILAO's legal content and serve it reliably to both the semantic search and conversational assistant.

A data ingestion pipeline prepares ILAO's legal articles for retrieval. Content is processed and structured to improve the quality of search results and assistant responses. The pipeline updates as ILAO's legal team publishes new or revised content.

Vector storage holds the processed content and enables fast, relevance-based retrieval. The configuration was sized for ILAO's content library and query volume.

The system exposes API endpoints that integrate with ILAO's existing Drupal site. This architecture keeps the AI layer separate from the CMS, making it easier to maintain, update, and scale independently.

Outcome

Users knew their problem, but not the legal vocabulary to search for it. The keyword search failed them. Live chat was too costly to sustain. Semantic search and the conversational assistant changed how Illinois residents find legal help.
‍

More users served

Comparing February 1 to December 31 in both years, active users of the search experience grew by 62%, from 29,924 (2024) to 48,463 (2025). Search page views increased by 61%, from 85,505 to 137,527.

Deeper engagement

Average engagement time per user rose by 35%, from 1 minute 14 seconds to 1 minute 40 seconds. Users are spending more time with the content they find.

More meaningful actions

Session key event rate more than doubled, from 4.44% to 10.15%. Users are not just viewing pages but completing actions like accessing legal articles, using tools, or following referral pathways.

Multilingual access

The Spanish-language interface (/es/semantic-search) recorded higher engagement time per session than English, showing the system serves Illinois' multilingual communities.