About AI Visibility Audit
AI Visibility Audit is an editorial and methodology resource published by NovCog (Novel Cognition) and operated by Guerin Green. It documents a repeatable way to measure how brands appear in AI-generated answers — with open evidence, clear limits, and no black-box scoring.
Use this resource to design an audit, document a repeatable sample, and evaluate the limits of its results. The worked example is a synthetic fixture; it does not report measured brand performance. For an inquiry about applying the method, contact NovCog.
Why this resource exists
Most discussions about AI visibility rely on vague claims or proprietary scores. This site takes a different approach: publish the method, capture the evidence, and let anyone rerun the process. The goal is to give marketers, founders, and researchers a practical starting point for understanding how AI assistants describe brands — without pretending the results are universal or permanent.
Repeatable evidence collection
Every audit follows a fixed protocol: define bounded prompts, record the assistant model and settings, capture responses verbatim, and store the raw output with timestamps. This makes each observation reproducible and comparable over time.
Transparent source and provenance
We show where each answer came from, which model produced it, and when. Sources are cited where they exist. If a claim cannot be traced, we say so. Corrections are published openly when new evidence changes a conclusion.
Real observations, clearly labelled
The worked example distinguishes between two kinds of content: actual observations, which would require captured answers and run metadata, and synthetic examples used to illustrate the format. The current worked example uses synthetic answers only. Synthetic fixtures are always marked as such. We do not fabricate rankings, citations, customers, or reviews.
Because AI models change frequently, every measurement is a snapshot. A result from one month may not hold the next. We document the model version and date so you can judge relevance.
Measured limitations
This method has clear boundaries. It does not claim to represent all AI assistants, all user queries, or all contexts. It measures what a specific set of prompts returns at a specific time. It does not predict search rankings, ad performance, or sales. It is a diagnostic tool, not a guarantee.
We also acknowledge that no audit can fully capture the nuance of how people phrase questions. The prompt set is deliberately narrow to stay reproducible. That is a feature, not a flaw — it means the results are comparable, even if they are not exhaustive.
Your reader journey
If you are new here, start with the method to understand the protocol, then follow the worked example to see it applied. The resources page offers checklists and templates you can adapt. For a full list of sources and tools used, see the credits.
If you have questions about the method, spot an error, or want to suggest a correction, please email Guerin Green directly. We treat corrections as part of the editorial process and will update the site with a note when a change is made.
Contact
Business email: [email protected]. For general inquiries, use the contact page.
Site analytics and privacy
We use a first-party image pixel and server request observations to understand which public pages are visited and which crawlers claim to request them. These are separate observations, not an additive visitor count. Browser labels alone do not verify crawler identity.
Our project analytics store the site, a fixed public page path where recognized, request method and response status, time, release, and a coarse crawler label in private Cloudflare storage. They do not retain raw IP addresses, full user-agent strings, query strings, form bodies, cookies, or referrers. The pixel uses a transient, keyed network-address counter for rate limiting. We do not set an analytics cookie or build a cross-site visitor profile. Cloudflare separately processes network traffic to deliver and secure the site. Questions can be sent to [email protected].
