A buyer opens ChatGPT and types "who are the best companies for precision sheet metal fabrication for aerospace." It answers with a few names. If yours isn't one of them, you never knew the question was asked. Learning how to get cited by ChatGPT, and by Claude, Gemini, and Perplexity, is about making sure your company is one the machine can read, trust, and name.
There's no single trick that makes an AI cite you. But the mechanics are knowable, and most established B2B sites are leaving them on the table. Here's what actually moves it.
First, see what the AI says today
Before changing anything, run the test yourself. Ask ChatGPT to compare your company to two competitors, or to recommend companies for the work you do, and read the answer. You'll usually learn one of three things: it doesn't mention you, it describes you vaguely, or it says something a few years out of date. That's your baseline, and it tells you where the gap is. (This is the practical side of answer engine optimization.)
Get your capabilities out of PDFs and into plain text
This is the single biggest lever. An AI reads web pages far better than it reads a PDF spec sheet, and most industrial companies keep their real substance, tolerances, materials, certifications, in downloadable documents. Put that information in actual page text. A model can only cite what it can read, and it can't reliably read a scanned brochure.
Answer the questions buyers actually ask
Models pull from pages that clearly answer a question. So write the answers plainly, on the page. What do you make. Who do you serve. What industries, tolerances, and standards. A well-built FAQ helps here, not as a keyword trick, but because it maps your content to the exact questions a buyer would ask the machine. The clearer the match, the more likely you're the quoted source.
Build consistent, verifiable authority signals
AI models weigh trust. They lean toward companies whose facts line up across the web: the same name, the same capabilities, the same clients and certifications on your site, your profiles, and any third-party mentions. Inconsistent or thin information makes a model hedge and reach for a competitor it's more sure about. Named clients, real certifications, and industry directory presence all feed this. It's the same authority work that helps you show up in Google.
Use structured data so the machine isn't guessing
Schema markup (organization, product or service, FAQ) tells a machine what a page is, in a format built for machines. It won't fix vague content, but on top of clear content it removes ambiguity about who you are and what you offer, which makes you a safer citation.
Optimize for citations, not tricks
There's no keyword-stuffing shortcut here, and trying one tends to backfire. Durable AI visibility rewards the same thing a good buyer rewards: real capabilities, made easy to read and verify. For established manufacturers, this overlaps almost entirely with SEO for manufacturers, and we run it as a program under AI SEO.
What to expect
Unlike a ranking, an AI citation isn't a single number you watch daily. You track it by re-running the prompts over time: are you named, more accurately, against stronger competitors than before. Progress here compounds the same way SEO does, and it starts the day you make your real work readable to a machine.
How common the gap actually is
It helps to know how low the bar is. When we audited more than 55,000 US B2B websites for The State of the Established B2B Website, 82% weren't structured for AI search to read them, and only 9% used structured data at all. Roughly half kept their specifications inside PDFs. So the competitor you're worried about probably isn't doing this either, and the work below is less about getting ahead than about being readable at all.
The same audit found that 68% of these sites never plainly say what the company makes or who it's for. An assistant summarising your market has to work with what the page gives it, and a homepage that opens with a slogan gives it nothing to attach you to. Clarity for a person and legibility for a machine turn out to be the same job.
How to tell whether it's working
Set a baseline before you change anything, because this is easy to feel good about and hard to measure. Write down five prompts a buyer would actually use, phrased the way they'd type them: "who makes X in the US", "best supplier for Y", "alternatives to Z". Run each one in ChatGPT, Perplexity and Google's AI answers, and record three things: whether you appear, which page gets cited if you do, and which competitors show up instead.
Re-run the same five prompts monthly. What you're watching for isn't a ranking, it's a pattern: appearing on more of the five, moving from a passing mention to a cited source, and being described in specifics rather than in a vague sentence. Assistants vary between runs, so a single check tells you very little and a monthly series tells you a lot.
One caveat worth setting expectations on. Being cited depends on the model's training and retrieval, so changes don't show up on a schedule the way a page edit does. Treat a quarter as the honest evaluation window, and treat the page work as worth doing regardless, because everything that makes you citable also makes you clearer to the person reading.
