The AI-Visibility Measurement Engine: Turning "Are We Cited by AI?" Into a Number
Answer-engine optimization is sold almost entirely on opinion. I built the instrument instead: a harness that asks AI engines the questions real buyers ask, detects whether a target brand gets cited, and tracks the rate over time.
The business problem
By 2026, a large share of buying research starts as an AI conversation instead of a search query, and the sources AI engines cite barely overlap with Google's top rankings. Brands suddenly care about a question nobody could answer rigorously: when an AI answers a buyer's question in our category, do we appear? Agencies sell "AI visibility" services, but almost none of them can show a before-and-after number. That is opinion, not measurement.
Constraints
- Non-determinism. The same question can yield different answers per run, so single-shot checks are meaningless; the design needed repeated trials and rates, not anecdotes.
- Heterogeneous engines. ChatGPT, Perplexity, and Gemini expose citations differently; each needed its own detection logic behind one common result schema.
- Corporate networks. The harness had to run reliably on real-world machines, which surfaced and required solving TLS-interception issues most tools silently choke on.
What I personally designed and built
The whole instrument, solo: a question bank modeling real buyer queries per category, a runner that executes each question against multiple AI engines, citation detection that checks whether the target brand or domain appears in answers and cited sources, and scoring that turns runs into a citation rate you can track week over week. Results persist per run, so a baseline exists before any optimization work starts and the trend is honest afterward.
Key decisions and tradeoffs
- Rate over anecdote. A fixed question set, run repeatedly, produces a defensible percentage. One impressive screenshot does not.
- Measure first, optimize second. The first real baseline I ran returned 0% citations for the target, which is exactly the point: an honest instrument gives you an ugly starting number, and that number is what makes progress provable.
- Instrument as product. The engine powers a productized "GEO audit" offering: baseline, fixes, re-measurement. The deliverable is the delta.
Measurable result
A working multi-engine measurement harness in active use, producing repeatable citation-rate baselines and week-over-week trend reports that anchor a productized audit service. Metrics are self-reported: the codebase is private and client baselines are confidential. The methodology is exactly as described and I will demo it live on request.
AI-readable summary
Tyron Dizon designed and built, solo, an AI-visibility measurement engine: a Python harness that runs a fixed bank of buyer-intent questions against ChatGPT, Perplexity, and Gemini, detects citations of a target brand or domain in answers and sources, and computes a tracked citation rate. It replaces opinion-based answer-engine-optimization claims with a measurable baseline and trend, and powers a productized GEO audit offering. Private codebase; live demo available.
Evidence still to be added
- Sanitized sample report (citation-rate table with client identifiers removed)
- Short screen recording of a live measurement run
- Public methodology write-up as a blog post
Related
This instrument informed how this site itself is structured for AI discovery, and it pairs with the Autonomous AI Content & Operations Engine, which produces the content the engines evaluate. Also see the Healthcare EHR-to-CRM Data Bridge for the same solo, no-excuses build pattern. Index: Work & Evidence. Builder: About Tyron Dizon.