Getting cited by ChatGPT: the new frontier of being found.

Why do citations beat rankings now?

Because your buyer increasingly asks the machine, not the search bar, and the machine only names a handful of sources. A ranked list of ten blue links used to be the prize. Now someone asks ChatGPT or Perplexity "who's the best partner for this?" and gets one answer, with two or three names in it. You're either in that answer or you're invisible. There's no page two to scroll to.

So the game has shifted from "rank well" to "get quoted". We call it answer engine optimisation, and the good news is it's not mystical, it's a repeatable playbook. The slightly annoying news is that most of what you'll read about it online is either hype or half the story. Let's do the whole story, no BS.

What does a model need before it'll quote you?

Three things, in plain terms: clarity, evidence, and authority. Miss any one and you get skipped.

• Clarity, a clean, self-contained answer the model can lift without rewriting your whole page. If your answer is buried three paragraphs deep, it's too much work to quote.
• Evidence, something to back the claim up. Numbers, examples, sources, a real position. Models hedge when they can't substantiate you.
• Authority, other credible places on the internet referring to you. Same domain-authority idea as SEO: if nobody vouches for you, the model won't either.

Notice what's not top of that list: schema. We'll come back to that, because it's the thing people most love to get wrong.

What's the five-step playbook?

Here's the version we actually run. Nothing clever for the sake of it, just the order that works.

1. Map the real questions

Not the questions you think people ask. The ones they actually type. Nobody searches "X agency" and stops there, they want detail, context, comparisons. Use the tools (Semrush, the People Also Ask box, and honestly just ChatGPT itself) to see the real shape of what buyers ask.

2. Answer first

Put the answer at the top. Question as the heading, direct answer in the first sentence, short summary before the body. A model should be able to grab it and go "yes, this is right", then read deeper only if it needs more.

3. Structure for machines

Make the page liftable. Short paragraphs, clear headings, a comparison table where it earns its place, FAQs at the bottom as extra quotable blocks. You're building a page that's easy for a human to read and easy for a machine to quote.

4. Cite yourself, with schema and sources

Back your claims with real sources and add schema so the technical layer is tidy. This is where JSON-LD helps, as a supporting act, not the headline. Link your own related pages so the model can see the shape of your expertise.

5. Measure mentions

Track whether you're actually showing up. A fixed set of buyer questions, run regularly across the engines, logging where you appear and whether you're named with confidence or with caveats.

What actually goes on the page?

The single easiest win we've ever shipped was a set of honest comparison articles, and it's worth walking through, because it shows what "answer the real question" looks like in practice.

We built a "us vs. them / alternatives" collection for a client: The Wayward Co. vs. [another technical marketing partner], that sort of thing. The trick was that we wrote them to be genuinely fair to both sides. Yes, we leaned toward the client, of course we did, but we also showed off the competitor's real strengths. Because that's what gets cited. A model answering "who should I use for this?" wants to say "go here if you want this, go there if you want that", and a fair comparison hands it exactly that. A hit piece hands it nothing it can trust.

"Nobody's ever really just searching 'X agency', they're looking for far more detail than that. So a fair comparison helps AI speak to what people are actually asking. We leaned it toward the client, sure, but we showed the competitor's strengths too, because that's what gets cited, not trashing a competitor."

ChatGPT and a bunch of others picked up on those signals fast. I'm not going to pretend the numbers were enormous, but it dropped us straight into the conversations that matter, the decision-stage ones where someone's ready to choose. It drove two to three enquiries a month, and it was one of the quickest results we've seen. A pretty easy collection to write, and easier still to get working. You'll find more of that kind of work in our client stories.

How do you track AI citations?

You track them deliberately, the same way you'd track rankings, because you can't improve what you don't measure. Keep a fixed list of the questions your buyers actually ask, and run them regularly across ChatGPT and Perplexity. Log whether you appear, in what position, and whether you're named cleanly or with a caveat.

Tools like Otterly.AI automate that watch across engines so you're not copy-pasting prompts by hand, and Google Search Console still tells you plenty about how your pages are being crawled and surfaced. Treat it like rank tracking's younger sibling: a scoreboard for whether the machines sitting between you and your buyer actually know who you are.

Which mistakes keep you out of the answers?

The big one, the mistake I'd put money on most people making, is thinking AEO is all schema and technical back-end. That if you just wire up all the JSON-LD perfectly, everything you've already got will magically get picked up.

"All the schema in the world won't get you cited if you're not answering the questions people are actually asking, and not the questions you think they're asking. Use the tools, see what people actually ask, and answer those."

Schema helps. It's genuinely useful. But it's the polish, not the substance. If your content isn't answering real questions clearly, no amount of technical set-up will save it. The other repeat offenders we see:

• Burying the answer halfway down a beautifully written 1,800-word essay.
• Trashing competitors instead of comparing fairly, models won't quote a hatchet job.
• Guessing at the questions instead of checking what people actually ask.
• Ignoring off-site authority, so the model has nobody backing you up and hedges or drops you.

Get the substance right and the technical layer amplifies it. Get it wrong and the schema's just lipstick. If you'd rather have senior operators run this end to end, no junior hand-offs, and you own everything we build, that's exactly the kind of work we do across AEO and search. Curious what it'd look like for you? Get a quote.