Generative Engine Optimisation: how to get cited by AI

Search is splitting in two. People still type things into Google, but a fast-growing share now ask an AI instead, and the AI doesn’t hand back ten blue links. It writes one answer and names a handful of sources. Google’s AI Overviews reached around 2 billion monthly users by early 2026, and ChatGPT passed 900 million weekly users around the same time. When someone asks one of these for a recommendation in your category, you are either in the answer or you don’t exist for that question.

Generative engine optimisation, GEO, is the work of becoming the source the AI picks. It’s the AI-search cousin of SEO, and the honest headline is this: it’s mostly good SEO done for a machine reader, plus being talked about in enough other places that the machine trusts you. There’s no secret file and no trick. You become the obvious answer. This page is the map of how, and it’s built in three levels: the standard work any operator can do (steps 1 to 7), an AI-native layer where you use the AI tools themselves, and a bleeding edge where you build your own monitoring. Start at the top and go as far as makes sense.

L1 Standard work Answer first, get specific, get indexed, get mentioned.

L2 AI-native Use the AI tools to audit and improve.

L3 Bleeding edge Build your own visibility tracking.

1. Understand how AI search actually decides who to mention

Before you change anything, picture what’s happening when someone asks an AI a question, because the whole game makes sense once you do. The AI doesn’t know the answer about your industry off the top of its head. It runs a live search, reads the pages that come back, and writes an answer grounded in them, naming the ones it leaned on. The jargon for this is RAG, retrieval-augmented generation. Ignore the term. It just means the AI looks things up before it answers, the way you’d skim a few tabs before giving someone advice.

Three things follow from that, and they’re the levers for everything below. First, where it looks matters: ChatGPT’s search leans heavily on Bing’s index, with one study finding about 87% of its citations matched Bing’s top results, while Perplexity crawls the live web and Google’s AI Overviews use Google’s own index. Second, it splits your question into several smaller ones and searches each, so covering a whole topic beats ranking for one phrase. Third, it favours sources that agree with each other, that are recent, and that already have authority. For a clear walkthrough of the mechanics, Ahrefs has a solid explainer, embedded below. Everything from here is about making your page the one that gets retrieved, and trusted, and quoted.

2. Answer the question first, in the first two lines

This is the single highest-impact change, and almost nobody does it. The AI lifts short, self-contained passages out of your page to build its answer. So every section has to make sense pulled out on its own, and the answer has to come first, before the build-up. A section that only makes sense after you’ve read the three above it will never get quoted.

The move is simple. Take the exact question a customer asks, make it the heading, and put a direct 40 to 60 word answer in the very first line underneath. Then add the detail. Lead with the answer, prove it below. If your page opens a section with “When considering the many factors involved in choosing a plumber…”, a machine has nothing clean to grab. If it opens with “A blocked drain costs $150 to $400 to clear in Brisbane, depending on the cause”, that’s a passage an AI can quote word for word. Write every section to be liftable.

3. Be specific: real numbers, named facts, quotes

Vague content doesn’t get cited. Specific content does, and this isn’t a hunch, it’s the clearest finding in the research. The original academic study on GEO, Aggarwal et al. at Princeton, tested what actually moves the needle and found that adding citations, direct quotations and statistics to a page lifted its visibility in AI answers by roughly 30 to 40%, with quotations the strongest single lever. Concrete beats abstract, every time.

So strip the fluff and put in the facts. Real prices, real dates, real measurements, named examples, a quote from someone credible. “We’ve fitted over 400 of these in the last two years” beats “we have extensive experience”. “Most jobs take two to three hours” beats “we work efficiently”. The machine is looking for a sentence it can stand behind, and a number it can repeat. Give it those, and you’ve done more for your AI visibility than any amount of keyword tweaking.

The trap. Don’t invent specifics to look quotable. AI engines cross-check sources against each other, and a number that nothing else on the web supports is more likely to be ignored than quoted. Use real figures from your own business or a named source. Honesty isn’t just ethics here, it’s what gets you cited.

4. Make your pages easy for a machine to read

A human can read a messy page. A machine wants structure. The clearer the structure, the more reliably an AI can pull your point out without dragging in noise, so clean structure is a ranking factor in disguise.

Three things do most of the work. Use plain, descriptive headings that match how people ask, so each section announces exactly what it answers. Keep one idea per section, because a tidy section gets quoted and a sprawling one gets skipped. And add structured data, which is jargon for a bit of code that spells out, in a format machines read, what your page is: a list of FAQs, your business details, a how-to. The most useful for AI is FAQ markup, which lets an engine lift a clean question-and-answer pair straight off your page. If your site runs on WordPress, a plugin like Yoast or Rank Math adds it for you. If it’s custom-built, it’s a small job for whoever built it. You don’t need to learn the code, you just need it there.

5. Get into the index each engine actually reads

You can’t be retrieved from an index you’re not in, and the index isn’t always Google’s. This is the step the content farms skip, and it’s pure plumbing.

For ChatGPT, the index that matters is Bing’s, because its search runs on it. So submit your sitemap to Bing Webmaster Tools, not just Google Search Console. It’s free, it takes ten minutes, and being in Bing is a precondition for being cited by the biggest AI engine. Then make sure you’re letting the AI crawlers in at all. Your robots.txt file (the small file that tells bots what they may read) should allow GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, PerplexityBot and Google-Extended. And if your site sits behind Cloudflare, check its settings, because since mid-2025 Cloudflare blocks AI crawlers by default on new sites. You allow them back in one click under AI Crawl Control. Miss this and every other step on this page is wasted: you’ve written the perfect quotable answer and locked the door on the things meant to quote it.

The trap. People do all the content work and never check the door is open. Before anything else, confirm the AI crawlers aren’t blocked, in robots.txt and in Cloudflare. It’s the most common silent killer of AI visibility, and the easiest to fix.

6. Get talked about off your own site

Here’s the part that stings: your own website is not enough. AI search leans hard on what others say about you, because agreement across independent sources is how a machine decides you’re trustworthy. One well-known finding is that spreading your content and mentions across many sites can lift AI citations dramatically compared with publishing only on your own. The engines favour earned media, meaning third-party coverage, over your own marketing.

So the work is to be mentioned where the AI looks. Get listed in the directories and review sites for your industry. Earn a few genuine mentions in local press or trade publications. And do not underestimate Reddit and Q&A sites like Quora: AI engines quote them heavily because they’re real people answering real questions, and a thread where your business comes up well is gold. You can’t fake this convincingly, and trying gets you caught. You earn it by doing work worth talking about and being present where your customers already discuss the problem you solve. The more places that name you, consistently, the more confident the AI is to name you too.

7. Show who’s behind the page

AI engines, like Google, lean on trust signals before they quote you, and the clearest one is showing there’s a real, credible person and business behind the words. The industry shorthand is E-E-A-T: experience, expertise, authoritativeness, trust. You don’t need to memorise it. You just need your pages to prove a human who knows their stuff wrote them.

In practice that’s small, concrete stuff. Put a named author on your articles, not “admin”, with a line on why they’d know. Show a visible publish date and an updated date, because freshness counts and AI skews toward recent sources. Have a real About page and contact details. If you’ve got credentials, licences or years in the trade, say so plainly. None of this is a trick, it’s just the difference between a page that looks like it came from a real business and one that looks like it came from a content mill, and the engines are increasingly good at telling those apart.

The platform underneath your pages

Every step so far assumes your page can be read cleanly, and that partly depends on what your site is built on, which is the consideration most GEO advice skips. Plenty of small businesses run on WordPress, Webflow or Squarespace: platforms built for people editing pages, not for being read by a machine. They tend to carry weight, heavy themes, stacked plugins, and content that often only appears once JavaScript has run. Google renders JavaScript and copes. The AI crawlers mostly don’t: analysis of their traffic shows GPTBot, ClaudeBot and PerplexityBot fetch the raw HTML and don’t execute JavaScript, so anything that only appears after your scripts run can be invisible to them. A lighter page, served as plain readable HTML, simply gets read more completely.

This is where the ground has shifted. Building a fast, hand-made site used to need a developer and a budget. Now an AI coding agent, like Claude Code, Cursor or Lovable, can build a clean, lightweight site in plain HTML, or in a static framework like Astro that ships almost no JavaScript and serves your content straight as HTML. It loads faster, reads cleaner, and you own the code outright, with no platform deciding what you’re allowed to change. For getting cited by a machine, that’s a genuine edge.

It’s a judgment call, though, not a rule, and the deciding factor is what you’d be leaving behind. A small, ageing site, a handful of pages on a bloated CMS that’s earning you nothing, is a good candidate to rebuild light: the risk is low and you’ll likely come out faster and cleaner. A large site with years of history, hundreds of pages and real ranking authority is the opposite, because re-platforming it is a serious migration where you can lose the rankings you already have, and you’re usually better improving the pages in place. Match the move to the situation.

The trap. Don’t re-platform a site that’s already ranking just because the stack is unfashionable. Hard-won authority is easy to damage in a migration and slow to win back. The lighter-site advantage is real for a fresh build, or a stale one worth replacing, not a reason to tear down something that already works.

Level two: use the AI tools themselves

Once the standard work is in place, there’s a newer level most guides skip, and it costs you nothing but an hour. The AI engines are the best tool for checking and improving your AI visibility, so use them directly.

Start by auditing yourself. Open ChatGPT, Perplexity, Claude and Google’s AI Mode, and type the ten questions a customer would actually ask in your category. “Best electrician in Newstead.” “Who can do a commercial fit-out in Brisbane.” Write down whether you’re named, and if not, who is. That list is your real scoreboard, and it tells you exactly which competitors the engines currently trust more than you. Then go read their pages and you’ll see steps 2 to 7 in action.

You can also put the AI to work on your content. Paste a page into Claude or ChatGPT and ask it to rewrite each section answer-first, in the customer’s own words, and to draft FAQ structured data from the questions it contains. It’s quick, and it’s the same job a GEO agency would charge you for. If you want the auditing done for you on a schedule, an entry-level tracker like Otterly runs those category prompts automatically and charts your share of mentions over time, from around $29 a month.

The maturity call. This rung is real and safe to do today. The one caution: AI answers vary run to run, so don’t panic over a single check. Look at the pattern across a few weeks, not one bad result on a Tuesday.

Level three: build your own visibility tracking

The deepest level is automating the monitoring so you’re not checking by hand, and this is where you stop reading and bring in someone who builds. The shape of it: a scheduled job runs your set of category questions through the Perplexity Search API and the OpenAI API, records every answer, flags whether you and your competitors were mentioned and cited, and alerts you when something shifts. You wire it together in a tool like n8n or Make, or build it directly with Claude Code, and pipe the results into a simple dashboard. The same setup can generate and maintain structured data across your whole site automatically, rather than page by page.

This is the rung the funded tools are built on. Profound, Peec AI and Ahrefs’ Brand Radar all do versions of this at scale, so for most businesses the honest call is to buy one of those rather than build, unless you want something specific they don’t offer.

The maturity call. This is the experimental rung, so no pretending. A custom tracker is a build that needs upkeep, the APIs change, and for a non-technical operator it’s the point where the answer is “get someone to build it once and hand it over.” The off-the-shelf trackers are the safer path unless you’ve got a real reason to go custom. Very little of this is written down, which is exactly why it’s an edge.

What’s overhyped: the llms.txt question

Because GEO is new, the advice around it is full of quick tricks, and the loudest right now is llms.txt: a small text file you add to your site that’s meant to tell AI models how to read it. You’ll be told it’s essential. The data says otherwise, for now. One study found the dedicated llms.txt file was fetched in about 0.1% of AI crawler visits, and Google has said its AI Overviews rely on ordinary SEO signals, not llms.txt. It’s near-zero effort, so add one if you like, but don’t believe it’s the thing getting you cited. The unglamorous steps above are.

That’s the pattern to apply to every new GEO tip you hear. Ask whether it changes one of the real levers: a clearer answer, a more specific fact, a more readable page, a wider index, more mentions elsewhere, more trust. If it doesn’t, it’s probably noise.

Do it yourself, buy, or get help

Do it yourself for the standard stack. Answering questions first, being specific, tidying your structure, submitting to Bing, opening the crawler door and getting mentioned elsewhere are all within reach of any operator, and they’re where the real win is. The AI-native auditing in level two is a free hour that tells you exactly where you stand. Start there, this month.

Buy a tool when checking by hand gets tedious, not before. A tracker like Otterly or Profound earns its place once you’re actively working on visibility and want to see the trend without running prompts yourself. It’s a measurement spend, so buy it when you’ve got something to measure.

Get help for the build and the heavy lifting. A custom monitoring pipeline, site-wide structured data, or a proper earned-media push is where real expertise saves you weeks, and it’s the rung where “get someone to build it and hand it over” becomes the honest answer. Whatever you do, be wary of anyone selling a guaranteed AI ranking or a secret GEO trick. There’s no ranking and there’s no secret. There’s being the clearest, most trusted, most talked-about answer in your category, and the work above is how you get there.