Illustrative example: grounding a publisher's AI assistant in a 15-year archive

Illustrative example. This is a composite scenario written to show how AACSearch can be used. It does not describe a named customer, and the figures are illustrative, not measured results.

The problem

The group's archive — 120,000 articles across regional news, lifestyle and classifieds — was effectively invisible: site search matched titles only, and readers rarely went past page one. The team wanted an AI answers product, but every off-the-shelf chatbot either hallucinated quotes or refused to cite anything.

The setup

The archive went into AACSearch through the import API — body, rubric, tags, author and publication date as fields; paywall level as a server-side enforced filter. Site search moved to AACSearch widgets with autocomplete driven by real query suggestions.

The assistant is bound to the archive collection: every answer is assembled from retrieved articles, streamed with citations, and scoped so that paywalled content never reaches an unauthorized reader. A knowledge graph of people, organisations and places sits behind retrieval, so «who is the mayor mentioned in the flood story» resolves to the right entity, not a fuzzy match.

What could change (illustrative figures)

• Internal searches per visit: ×3.4 (illustrative) — readers treat the search box as the primary way into the archive now.
• 89% of assistant answers with a citation click (illustrative), which editors would read as the trust signal they couldn't get from a generic chatbot.
• Zero editorial hours on search upkeep (illustrative). New articles are searchable on publish; synonym suggestions from zero-result queries arrive monthly for review, not as a weekly firefight.

Why it matters

The citation requirement did the heavy lifting: once every answer had to point back to the archive, the assistant became a navigation product the editors could defend, rather than a novelty that occasionally invented a quote.