ChatGPT cites businesses it can find, verify, and quote across the surfaces it searches at answer time: your website, Bing's index, review platforms, and third-party mentions. You earn citations with pages that answer one buyer question each in specific, verifiable terms, indexed where the model actually looks. We are testing this process on our own site, day by day.
Why does most advice on this query fail you?
Two reasons. First, the playbooks are stale. Google retired FAQ and HowTo rich results in early 2026, so half the standard advice you will read is dead weight: markup that no engine rewards and checklists built for a results page that no longer exists. Second, guide authors never show their own citations. They tell you the method works and offer no logged queries, no dates, no screenshots. Demand receipts from anyone who sells this, including us.
Where does ChatGPT actually look?
ChatGPT's browsing queries Bing's index. Perplexity runs its own index plus Bing. Google's AI Overviews pull from Google's index, where 62% of citations now come from outside the top 10 organic results, per Ahrefs' 2026 study. The shared dependency across the first two is Bing, and most businesses have never opened Bing Webmaster Tools. That gap is the cheapest fix in this whole discipline.
What is the 5-step process?
Step 1: find the questions buyers ask AI
Pull them from sales calls, support tickets, and the cost and comparison questions buyers raise before they commit. Run each question through ChatGPT and Perplexity and log who gets named. Queries where the incumbent answer is vague are your targets: the model wants a better source and you can become it.
Step 2: publish one page per question
Each page answers its one question, with a number or a date in the first 50 words. Use question-form H2s, and open each H2 with a 40 to 60 word direct answer before any elaboration. Models lift the opening block; make it liftable.
Step 3: become legible to machines
Ship lean JSON-LD: Organization, LocalBusiness where relevant, and Article with a named author and dates. Skip the retired FAQ and HowTo markup. Keep your name, category, and profile data consistent everywhere a model reads, because conflicting facts make the model hedge, and a hedging model names your competitor.
Step 4: get indexed where engines retrieve
Open Bing Webmaster Tools and Google Search Console, submit your sitemap, and use IndexNow for new URLs. Then run the retrieval test: ask the AI tool to summarize your URL. If it can read the page back to you, you are in the index it retrieves from. If it cannot, fix that before you write anything else.
Step 5: build the second surface
AI answers cite Reddit threads, LinkedIn articles, and YouTube videos alongside business sites. Our own day 0 logs caught LinkedIn articles cited inside Google AI Overview panels. Publish your core answer natively on whichever of those surfaces your niche triggers, so the model meets your answer twice.
How long does it take?
First citations on long-tail queries take 2 to 4 weeks, and the documented cases cluster around week 3. Head terms take months. Nobody can schedule placement in model training data, and anyone who promises a date for that is guessing on your invoice.
Want the map for your own business?
The $397 audit maps 50 buyer questions across 6 AI platforms and ends in a ranked fix list. The free 48-hour scan is the short version. Both start at the same place.