How Monroya works.

A single prompt to ChatGPT can return different answers minute to minute. To make a score that means something, we run every tracked prompt multiple times per scan and aggregate the runs into a stable signal. Automated scans run weekly; you can also trigger a manual rescan any time. Outliers get smoothed; persistent patterns get surfaced. Your score reflects what AI consistently says, not what it said once.

Monroya scans ChatGPT (GPT-5 and 4-class), Claude, Gemini, and Perplexity in parallel. Each provider weights sources differently: Perplexity leans on cited domains; ChatGPT favors well-known brands; Gemini pulls from Google's index; Claude tends toward documentation. Looking at all four together is the only way to know whether a gap is universal or model-specific — and the fix differs.

A VP of Engineering asking "best observability tools" gets a different answer than a procurement lead asking the same question. Every prompt in your account is scanned with a buyer persona attached — role, funnel stage, intent. We track your visibility per persona so you see where you're strong with technical evaluators but invisible to economic buyers, or the reverse.

A mention is your brand named in the answer text. A citation is your domain linked in the source list. They behave differently: mentions build awareness, citations drive clicks. Most tools collapse them into one number. We track both, because the fix for a mention gap is brand presence on third-party sources, and the fix for a citation gap is your own pages getting indexed and cited.

Each gap becomes an opportunity card with three scores:

• Impact — how much your AI visibility should move if you fix it.
• Confidence — how often the gap shows up across runs and providers (the more consistent, the more real).
• Effort — how much work the fix actually takes.

Those three combine into a single Priority score that orders your queue. We show the underlying numbers, not just the ranking.

When an opportunity needs a new asset — a comparison page, a pitch to a publication, a Reddit answer, a fact sheet — clicking Execute drafts it. The draft uses the gap evidence as input and writes in your buyer's voice, not generic marketing copy. Every draft is fact-screened: invented stats, fabricated version numbers, and made-up customer counts are flagged before you ship. Stats-heavy formats retrieve fresh, dated sources at draft time so the numbers you cite are current. You edit and ship. No copywriter handoff, no two-week loop.

Tracked prompts don't sit static. When a prompt's citation rate drops or stalls, Monroya generates richer variants — different framings, persona angles, comparative phrasings — and runs them alongside the original. The variant that earns the citation gets promoted into your active set. You see the variant history and can accept, reject, or pin any version.

The site crawl doesn't just feed prompt generation. Every indexed page is graded for GEO readiness — schema coverage, citation-ready structure, freshness, internal-link depth — and surfaces a short list of one-edit quick-wins per page. The kind of fix that takes 15 minutes and shifts a citation rate the next scan.

When you mark an opportunity complete, we capture a baseline of your stage-inclusion rate. After the expected timeline passes and your next scan runs, we measure the delta. Once a given (action type, asset format, journey stage) combo has 3+ measured outcomes, its aggregate success rate re-weights future priority scores — confirmed winners get a +20% boost, consistent failures get a -15% penalty. Your verified playbook is auditable on the Progress page.

Every score traces back to the raw model responses that produced it. If your AI Share of Voice dropped three points, you can see which prompts moved, which providers shifted, and what the actual answer text was. Nothing about the methodology is hidden behind a black-box score.