Answer engine optimization (AEO) is the part of the work that starts from the question: finding what your buyers actually ask AI assistants and search engines, and publishing an answer in a form an engine can lift and quote. It is for content leads, SEO teams and agencies who want to write to questions rather than to keyword lists. In The Ranking Factory, question discovery feeds one topic list, and every answer is written to a shape that two or three of your own AI providers agreed on before a word was drafted.
Discovery and writing are the two halves of this page. Measuring whether the answers then name or cite you is covered on the AI visibility and AI citation tracking pages.
Seven sources of questions feed one topic list; four steps turn a topic into a page.
The topics you confirm at setup, and an optional “What do customers ask you?” box: up to ten questions at once, each saved as a tracked prompt and asked weekly. Nothing in setting up waits for it.
With Search Console connected, the real queries people ran to reach your site, with impressions, click-through and position. A search where you sit on page one and almost nobody clicks is flagged, because it is consistent with an AI Overview answering in your place, or with another results feature. No search volume is estimated or bought.
For a tracked prompt the ChatGPT app answers, the searches the app ran to answer it are listed as “N of M answers”, each marked as covered by one of your topics or not. A search no topic covers has an Add as topic button. Nothing is added automatically.
Each cycle mines recent AI answers for topics you are emerging for and topics competitors are cited for. Competitor gaps lists searches where a rival holds page one and you are absent, from rank checks already running, and hands any of them to the Evidence Engine as a topic.
Google and YouTube autocomplete expansion, question clusters and related terms. Reddit discussions already asking what you answer are opened as tasks with guidance on replying; the product never posts, and you paste your reply’s link back.
Near-duplicate phrasings cluster under one parent so a question is not measured twice. Each topic carries a buying-journey stage (Buying, Comparing, Learning) that orders the queue. Up to 250 main topics per project.
For each topic, two or three of your own AI providers from different companies are asked what kind of page the question needs (an FAQ, a how-to, a comparison, a glossary, a checklist, a dataset) and what it must contain. The majority answer is stored as the page’s contract. Where they cannot agree, the topic is written as a guide.
The writer gets your own pages and videos on the subject and what the strongest outside sources actually say. Each section is written to stand on its own when quoted. Afterwards every element of the contract must be proven by a quote from the page; a missing one gets one redraft naming it. A product of yours that nothing explains is not written about until you say what it is.
An FAQ is built as real questions and answers, a how-to as steps, a glossary as definitions, a dataset as a table. The FAQ shape is kept for the structure, not for a rich result: Google withdrew the FAQ rich result on 7 May 2026.
Where you switch grading on, each page on your own site is put to two or three AI models with web search off, and each must quote the passage that answers the real Search Console query that showed the page. The verdict is Answers, Does not answer, or Contested. A page in the wrong shape can be rebuilt from the models’ brief, keeping its verified lines word for word.
Worked example: an illustration
The shapes are the product’s: the setup box, the Search Console row, the “N of M answers” searches, a shape contract with its elements checked by quote, and a graded verdict. The business and every figure are made up: an installer of home backup batteries.
| Provider | Kind of page | Must contain |
|---|---|---|
| Provider A | Comparison | A side-by-side table; when a generator is the better choice; a source for any run-time figure |
| Provider B | Comparison | A table of cost, run time, noise and upkeep; the installer’s own service area |
| Provider C | FAQ | Run time by household size; whether it works off-grid |
Majority: Comparison. The majority answer becomes the page’s contract: four elements here.
Three different sources pointed at one question, and none of them needed a volume estimate: a customer asked it, Google showed the site for it, and ChatGPT searched for it. The page was then written to a contract, and the contract was checked against the page’s own sentences before anything went out.
“Answers” means the page contains a passage that answers the query, as quoted by models that could not search. It does not mean an engine will retrieve the page or cite it. That is measured separately, on later runs.
Finding the question and answering it well are within reach. Whether an engine picks your answer is not, and the product keeps the two apart.
A page that answers its question is not a page an engine will surface. Answerability says nothing about retrieval, and it is never averaged with visibility or plotted beside it.
The panel’s choice of shape is an opinion of the providers you connected, and the screens label it as one. A verdict exists only where a real Search Console query was tested; a page with no such query is asked about its own subject and carries no verdict.
The searches ChatGPT ran are read only for tracked prompts, only from the ChatGPT app (the Gemini app reports none), and they change from one answer to the next, which is why each is shown as N of M answers.
No keyword volume is estimated. Where Search Console is not connected, you are told the evidence could not be reached, never that none was found.
Where the panel says a question is answered best by a kind of source a business cannot be on its own site, such as an independent review, the page is still written and you get one task naming the venues suggested. The product never places content on other people’s sites.
[SCREENSHOT: /tools/visibility-audit, showing an opened tracked prompt with its “Searches ChatGPT ran” block: each search as N of M answers, covered or not by a topic, and the Add as topic button]
Question discovery and answer writing are on every plan, on your own AI keys; automatic discovery each cycle starts on Growth. Growth monthly starts with a 14-day free trial; annual billing is ten months for twelve.
$49 a month
Topics, tracked prompts, keyword research and the Evidence Engine when you press Run; discovery when you run it.
$129 a month
Automatic topic discovery every cycle, the engine every 6 hours. 14-day free trial.
$249 a month
The engine every 4 hours, multi-location onboarding and white-label reports.
$499 a month
The engine every 3 hours, and the largest page-by-page scans.
The shape panel costs two or three model calls per topic, once, on your own keys. Grading costs two or three calls per page, once, and runs only where you switch it on. The passage check is one call per section, at most 12 a page, and is off until you turn it on.
Asking the ChatGPT app your tracked prompts, which is what lists its searches, runs on your own DataForSEO credit at about $0.0012 a question.
The same product, from another task
Find them in your own data and in the engines’ own searches, then publish answers written to a shape that was agreed and checked before they went out.
Start your 14-day free trial