We’re building the measurement layer for AI search.

Ansengine measures whether an AI answer mentions your brand, whether it cites your pages, and whether what it says about you is supported by a source. Three rates, one prompt at a time, one model at a time.

Every rate carries the number of answers behind it and the range it could honestly be. A rate built on four answers prints as a rate built on four answers, because the count is the difference between a finding and a feeling.

There is no rank position anywhere in the product. Position in an AI answer is not stable across repeated runs, so a rank is a number that looks precise and means nothing. We would rather sell a smaller true number.

Every prompt is measured every week, because the answer changes from one asking to the next. One answer is an anecdote; the count over the weeks is the measurement, and it is what can tell you whether anything you did worked.

Every model is measured on its own and reported on its own. They agree on only a small share of the sources they cite, so an average across them describes no surface a buyer ever opens.

Movement is reported only when the ranges stop overlapping. A difference that could be noise is labelled normal variation, on every plan, at every price. Anything unmeasured renders as a dash and never as a zero.

Every plan includes 3 models, ChatGPT, Google AI Overview and Google AI Mode, and can add 3 more per prompt, on request, with 2 developer APIs on Agency Scale and custom plans. The included set is the apps buyers actually open, not the cheapest endpoints to poll.

ChatGPT is read on chatgpt.com, the app your buyers use, rather than only through the developer API. The two do not return the same answers, and reporting the API alone would flatter every number on this site.

Developer APIs are measured and reported on their own line, never averaged into a headline with the consumer apps.

The free report is a real baseline and not a teaser: 10 prompts across 3 models. That is 150 measured answers, every one stored with its sources, and no card.

On a plan the measurement continues on a cadence, opportunities are ranked by what it would move, drafts are written from facts you verified, and nothing reaches your own site without an approval from your team.

Receipts confirm a win only when the ranges separate. Every answer downloads as CSV and every payload as JSON, so any number we print can be checked outside this product.

Schema markup and llms.txt are hygiene. They are not ranking levers for AI answers, and we say so on the page where a competitor would be selling them to you.

No rank positions. No invented statistics. No star ratings, review counts or aggregate ratings on our own pages, which is the shortcut this category takes by default.

Nothing publishes without an approval from your team, and an API key or connected agent you authorise can give that approval. The agent drafts, the approval releases it, the system acts.

Our own grades are published on the methodology page: a hand labelled golden set for each stage of the pipeline, the score it reached, the bar it has to clear, and the date it last ran. A stage that did not run says so rather than showing a zero.

The enforcement page mentions the test that fails if we break one of the rules above, so a claim on this site is backed by something that can go red.

Every page on this site carries the commit that served it, in a meta tag mentioned ansengine-build. Any claim here can be checked against the exact code that made it.