Technical (engineers, DevOps, data)

AI search visibility for engineers

How is AI-search visibility measured, and can I reproduce the measurement myself?

Measured: at least 10 runs per engine, spent adaptively, across 5 AI engines · 95% confidence interval on every figure · open data (DOI 10.5281/zenodo.21612952 (opens in a new tab)) · how we measure

Read the method, or run the open-source tool against your own domain.
Every number we publish is a median across at least ten runs per engine, carried with a confidence interval and the date it was captured. We ask each engine -- ChatGPT, Perplexity, Claude, Gemini and Grok -- a fixed, published set of buyer prompts, record which sources each answer cites, and report share of model rather than a rank. The prompt set, the run count and the extraction rules are published, so you can disagree with the method on the evidence rather than on the marketing.

What is actually measured

A single question asked once is noise: the same engine answers the same prompt differently across runs, and the difference between two companies is often smaller than that run-to-run spread. So the unit is a distribution, not an observation. For a category we hold the prompt set fixed, run it repeatedly against each engine, and report (a) share of model -- the share of answers in which a company is named -- and (b) citation share, the share of cited URLs that belong to it. Both carry an interval and an as-of date, because an engine's behaviour on 2026-09-01 is not evidence about its behaviour today.

Run it against your own domain

The audit tool is open source and runs offline in mock mode, so you can read exactly what it sends before you give it a key. The AI Visibility Index publishes its category datasets as CSV and JSON under an open licence, which means you can re-derive our published figures from our published data and tell us if they disagree. If you would rather not run anything, the free teardown does one pass for your domain and hands back the raw runs alongside the summary.

Where this method can be wrong

Engines change continuously and without notice, so any measurement is a point in time -- we date every figure for that reason. Answers are non-deterministic, so a small sample can move a company several places for no real reason; that is what the interval is for. Retrieval depends on the asking region and on what the engine has cached, and we control for neither perfectly. Where something is not measurable with the runs we have, the page says UNVERIFIED rather than estimating it. Corrections are published with the same prominence as the original figure.

Claims we refuse to make

No engine publishes a ranking formula. Any breakdown that assigns fixed weights to ranking factors -- the 40/35/25 split that circulates in AEO marketing is the common one -- is invented, and we will not repeat it. We do not guarantee a citation, a ranking or a position, because nobody controlling those systems has offered anyone that guarantee. What we will commit to is the method: reproducible runs, published prompts, intervals on every number, and the raw data so the conclusion is checkable.

What counts as proof for this reader

The method, the raw runs, and the confidence interval -- a number without its n and its CI is noise to this reader.

No figure is quoted on this page. The measured figures live where their runs, intervals and capture dates live: the AI Visibility Index and the methodology.

Read the method, or run the open-source tool against your own domain.

A free AI-Visibility Teardown measures where you are recommended and where you are not — across the 5 AI engines — ChatGPT, Perplexity, Claude, Gemini, and Grok — plus Google's AI surfaces (AI Overviews & AI Mode), measured separately — reproducibly, with the raw runs attached. No call.

Start the free teardown

Reading for someone else? Find your fit.

Our open-source tools

Clear Cited is a Canadian practice based in Markham, Ontario, serving the US and Canada; US-market measurement is the default and billing is in USD.

We do NOT guarantee rankings. We guarantee rigorous, reproducible measurement plus evidence-based work against a measured share-of-model target and a fixed monthly deliverable cadence.