AI Visibility Audit for publishers · Pilot engagements
Know what AI answers do with your work
Test the questions your audience actually asks, see which sources AI systems use, make a real change, and measure what happens next.
Why we’re building this
Lyra Forge’s AI Visibility Audit is designed for a specific publisher decision: when an AI answer becomes someone’s first encounter with a subject, a publisher needs more than a visibility score. It needs to know which audience needs are being served, where its evidence is being missed or misused, and which changes are worth making.
Start with decisions, not a keyword list
We begin with the people you serve, the decisions they are trying to make, and the evidence they need. That work determines which questions to test. Search demand, modeled AI demand, referrals, and customer outcomes remain separate signals rather than being rolled into a made-up audience number.
Keep the evidence inspectable
The audit preserves the exact question, model or interface, date, answer, citations, matching source pages, and known limitations. It distinguishes whether an answer is useful from whether it cites you. Those are related questions, not the same score.
Ship a change, then retest
A recommendation is only a hypothesis. We identify a useful page, make an evidence-backed editorial or technical change in the real publishing stack, and repeat the relevant observations. The handoff shows what improved, what did not, and what is worth doing next.
Use it when a decision cannot wait for a new team
The best fit is an evidence-rich publisher or paid information product with a live product or distribution decision that needs an independent baseline, interim capacity, or hands-on implementation. If the only need is an automated share-of-voice chart—or no editorial, product, technical, or policy decision would change—monitoring software is likely the better tool.
A live test on NJ School Data
We tested 12 parent questions on three current AI surfaces, improved one public discipline-and-safety page, and initially saw better page readiness but no citations. The broad question was ambiguous, so we kept it unchanged as an ambient-discovery measure and added a separate district-grounded probe. On that probe, Codex cited the current South Orange-Maplewood district page first and used its figures; Claude and Google did not. One result does not establish repeatability or demand, but it shows the method can separate a useful site change from an observed retrieval result.
What an engagement leaves behind
You receive the audience-and-question rationale, raw answer and citation evidence, a source-linked diagnosis, prioritized changes, the implemented work that is in scope, and a retest receipt. The artifacts remain usable outside our tools, so the engagement does not require an ongoing software subscription.
Evidence, not a showcase score
A real test, including what did not improve
Follow one audience question from selection through source audit, a shipped page change, and live-surface retesting.
See the NJ School Data field test →Pilot work is available for evidence-rich publishers and information products with a concrete decision to make. AI answers vary, external systems control retrieval, and no intervention can guarantee a citation. The NJ School Data result is an observed case, not a promise of lift.
Contact
To discuss a publishing problem or ask about this tool, get in touch through Lyra TK.