HOW TO GET YOUR BUSINESS CITED IN AI ANSWERS, WITHOUT GAMING ANYTHING

A prospect told me last month that half his new enquiries now open with "I asked ChatGPT and it suggested I look at companies like yours." Not "I googled you". Not "I saw your ad". The AI recommended a shortlist, and he wanted to be on it. That is the whole game now, and most businesses have no idea whether they are on those lists or not.

This is a field guide to getting on them. I am writing it because we built the same thing for ourselves before we sold it to anyone, so this is what actually moved the needle, not a theory I read.

First, understand what changed

For twenty years, search meant ten blue links. You optimised a page, you ranked, you earned a click, the visitor landed on your site. That still matters. But a growing share of buyers now get an answer before they ever see a list of links. They ask ChatGPT, or they type a question into Google and read the AI Overview at the top, or they use Perplexity, and the machine hands them a synthesised answer that names a few sources.

The important shift is in that last part. A blue-links page shows ten results and lets you choose. An AI answer picks for you. It cites a handful of sources, often three or four, and everything else is invisible. Being on page one is no longer the finish line. Being one of the sources the answer is built from is.

People have started calling the work of getting cited "generative engine optimisation", or GEO. I am wary of the acronym because it makes it sound like a new dark art with a new set of tricks. It is not. GEO is mostly the oldest idea in marketing wearing new clothes: be the clearest, most trustworthy source on a question people are actually asking. The difference is that your reader is now partly a machine, and machines are ruthless about clarity.

The uncomfortable truth about who this is for

If you have never touched AI, this is not your first problem. Go and use ChatGPT for a week first. This guide is for the operator who has already used it, probably has a few subscriptions, has data and tools scattered across a dozen places, and is now realising the ground has shifted under their marketing. You are not starting from zero. You are mid-transition, and the job is turning that scattered effort into a system that compounds. That is a very different, and honestly more valuable, position than starting cold.

The four levers that actually work

Here is what we do, in the order that matters.

Answer the question directly, at the top. Language models lift text that states a claim cleanly. If your page buries the answer under three paragraphs of throat-clearing, the model has to guess what you are saying, and it would rather quote someone who just said it. So we lead with the answer. On this very page, the takeaways at the top are not decoration. They are the most liftable summary of the article, written so a machine can quote them without misreading us.

Make the facts machine-liftable. This is the part most people miss. A model trusts a clear number in a table far more than the same number floating in a sentence. So we structure our facts: proper headings, tables, and schema, the invisible layer of markup that tells a machine exactly what a page contains. Every data chart we publish is encoded three ways at once, as a picture for you, as a plain table a search engine can read, and as structured data an AI can cite. That sounds like overkill until you realise it is the difference between being quoted and being skipped.

Earn mentions where the models read. AI answers are built from the wider web, not just your site. If reputable sites, industry publications and the right forum threads mention you in a genuine, useful way, you show up in the training and retrieval that answers draw on. This is not link spam. It is being genuinely present and helpful in the places your buyers already trust, which, conveniently, is also just good marketing.

Measure your citations, then repeat what works. You cannot improve what you cannot see. Most businesses have no idea whether AI answers mention them, so they fly blind. We track how often the major AI answers cite a domain, for the queries that matter in its space, and against the competitors who currently own those answers. That number is the scoreboard. Everything else is in service of moving it.

What a liftable fact actually looks like

Let me make "machine-liftable" concrete, because it is the lever people find hardest to picture.

Say you run a plumbing firm and you want AI answers to recommend you for emergency callouts. The weak version, the one most sites publish, reads like this: "We pride ourselves on a fast, friendly and reliable emergency service across the region, with highly trained engineers you can count on." A model reads that and learns nothing it can quote. Fast compared to what? Which region? How fast?

The liftable version states the facts a machine can lift and a buyer actually wants: "We answer emergency callouts across Staffordshire within 60 minutes, 24 hours a day, with a fixed callout fee of ninety pounds." Now there is a claim. A specific area, a specific response time, a specific price. When someone asks an AI "who does fast emergency plumbing in Staffordshire", that sentence is quotable, and yours is the firm that gets named while the "fast and friendly" crowd stays invisible.

The pattern holds for every business. Replace adjectives with facts. Replace "leading" and "trusted" and "bespoke" with numbers, places, timeframes and named specifics. The models do not care how you feel about your service. They care what they can safely repeat on your behalf.

Finding the questions worth answering

None of this works if you answer questions nobody asks. Keyword tools tell you what people type into Google. They are a weaker guide to what people ask an AI, because people talk to ChatGPT in full, messy, specific questions, not two-word searches. "Best CRM" is a Google search. "Which CRM should a 12-person recruitment agency use if we already run everything through Google Workspace" is an AI prompt, and it is a far better question to own because the person asking it is closer to buying.

So we mine the real questions: the ones clients get asked on sales calls, the ones that show up in support tickets, the long specific ones buyers are now comfortable typing into a chat box. Then we write the clearest answer on the internet to each one. That is the raw material. Everything else is structure and distribution.

Why "just write more blog posts" fails

The instinct, when someone tells you to do content, is to publish more. More posts, more words, more keywords stuffed in. That is exactly the wrong response, and it is the same mistake I see people make with AI outbound: confusing volume with the thing that works.

A model does not reward you for publishing forty thin articles. It rewards one genuinely authoritative, clearly structured answer to a question people ask, on a domain it has reason to trust. One page that is the best, clearest source on a real question will get cited over and over. Forty pages that half-answer forty questions will get cited never. The work is not typing more. It is being more useful and more liftable than whoever the answer currently names.

This is also where the scattered-tools problem bites. If your content, your data, your product facts and your proof all live in different silos, you cannot present a machine with a single clear, trustworthy source. Half the GEO job is just consolidating what you already know into one authoritative place. That is unglamorous, and it is most of the win.

Where honesty comes in

I will not pretend there is a button. Getting cited is earned, it takes weeks not days, and anyone selling you an instant AI-ranking hack is selling you the version that gets your domain flagged. The models are getting better at spotting manufactured authority, not worse. The durable move is the boring one: be the real answer.

I will also be honest that the measurement side is new and imperfect. The tools that track AI citations are a couple of years old, not twenty. But imperfect measurement of the right thing beats perfect measurement of the wrong thing, which is what ranking-only reporting has become. I would rather know roughly how often AI names you than precisely where you sit on a links page nobody reads anymore.

Start by finding out where you stand

The reason most people do nothing about this is that it feels abstract. "Am I in AI answers?" has no obvious answer, so it stays a worry rather than a task. So make it concrete. Find out, for your actual domain, how often AI answers cite you against the competitors who currently own your space. Once you see the gap in black and white, the work stops being abstract and becomes a to-do list.

That is the entire reason we built a free scan for it, and it is the honest first step: see the number, then decide if closing it is worth your time. For most businesses I talk to, the moment they see a competitor cited ten times and themselves cited once, the decision makes itself.