Your agent finds out who the AI engines cite.Then builds what gets quoted.

An AI answer is built from a question and the sub-questions the engine fans it out into — most of them have no measurable search volume, and they carry about a third of all citations. The questions come from data, not guesswork: the ones Google hangs on the results page (seo serp → People Also Ask and related searches), the question-shaped keywords your competitors live on (seo competitors), the questions Search Console already shows you for, the forum threads on the page, the sub-questions ChatGPT actually searched (--engine), the stored AI answers that cite a rival (competitors --ai) — and the questions real people sent to your own inbox.

invent questions, or call an estimate a demand curve. Every row carries its source and whether a real person asked it.

Google's top ten is not the AI answer: only 38% of AI Overview citations come from it, and 8% of ChatGPT's. One seo serp — or clize geo, the same read with the AI face on by default — gives you the Google face (who occupies the page, which roundups you could be in) and the AI face — who the AI Overview quotes, with the sentence it lifted. Add --engine chatgpt,gemini,aimode,perplexity for who each engine cited, which brands it named and what it searched. Each is one sample on demand, about $0.02 per engine, cached 30 days. What it tells you is the shape: vendor docs, a publisher roundup, a forum thread, a wall of official docs — and that shape decides the next step.

keep a score or a schedule. There is no "were you cited" field and no cron — the same prompt rarely returns the same answer twice, so a percentage would be noise dressed as a number.

Every question ends in one decision: fix a page the engines cannot fetch, build the answer that is missing, get the evidence first when your page would only restate what the cited pages say, record that a rival is named where you are not, or defer. What gets a small site quoted is something the roundups cannot restate — an integration you actually ran, a comparison that can be reproduced, original numbers, exact pricing and limits. The free skill carries the page shapes engines quote (a category comparison with a real table; a short answer page per question cluster), and build site check warns on the three parts a machine can see — a comparison page that never names you in its first 200 words, one without a table, a content page with no visible date.

generate content, pad to a length, or serve engines a page people never see. The writing is your agent's brain; Clize is the hands.

Engines reach you two ways — live retrieval when the question needs current facts, and model memory that lags months — and you only control the door: search and training crawlers allowed in robots.txt, no edge block (domain check reads Cloudflare's AI-crawler policy as its fifth layer; new zones default to blocking since 2026-09-15), body text in HTML, a URL that resolves in one hop, a sitemap whose lastmod is a real content change, IndexNow for Bing's index. Clize-hosted sites ship all of it on deploy; for any other site the skill carries the checklist.

submit your URL to ChatGPT. There is no such thing — anything that claims otherwise is selling something.

seo check lists the visits AI engines actually sent, page by page — the milestone is the first non-zero. Above that sit two reports that only exist as web pages, Search Console's generative-AI impressions and Bing's AI Performance citations, which the skill reads as a time series against yourself. Outcomes are recorded in layers — fetched, cited, named, recommended correctly, visited, converted — because being cited says nothing about being named, and being named nothing about being described correctly.

report an AI visibility percentage. Sampling dressed as a share is a product-integrity failure, and we would rather show you a zero.