Last month a mate rang me for a plumber. He did not open Google. He opened ChatGPT, typed "best emergency plumber on the Northern Beaches", read the three it suggested, and called the first one. That plumber did not run a single ad. He just happened to be the business the AI decided to name.
This is the quiet shift under everyone's feet in 2026: a real and growing slice of "who is the best", "recommend me a", and "where should I buy" questions now go to ChatGPT, Claude, Perplexity, Gemini and Copilot instead of a search bar. If your business is invisible to those tools, you are invisible to those buyers, and most Australian businesses have given the AI almost nothing to work with.
The good news: this is early, the rules are learnable, and the businesses that get visible first will own the "just book me one" requests while their competitors are still arguing about Google rankings.
Quick answer
To get recommended by AI assistants, you need to be present in the three places they draw answers from: their training data, the live web they search, and the third-party sources they trust (directories, reviews, "best of" lists, forums). Practically, that means: make your content answer-first and easy to quote, get your business identity crystal clear (consistent name, address, phone and a Wikidata or knowledge-panel presence), add structured data plus an llms.txt file, earn mentions on the review and comparison sites the models cite, and, increasingly, make your business transactable by AI so it can actually book or buy. This discipline has a name now: Generative Engine Optimisation, or GEO.
How an AI actually decides who to recommend
It helps to know what is happening when someone asks an assistant for a recommendation. There are three ingredients, and you want to be in all three.
1. What the model was trained on. Large language models learn from a huge crawl of the public web. If your business, its category and its reputation appear across enough credible pages, the model has a baseline sense of who you are. You cannot edit training data directly, but you influence it over time by being well represented across the web, not just on your own site.
2. What it retrieves live. Modern assistants do not rely on memory alone. When you ask ChatGPT or Perplexity a "best X near me" question, they run live web searches and read the top results before answering. This is retrieval-augmented generation, and it is the part you can influence fastest: if your pages are clear, current and citable, you get pulled into the answer today, not in the next training run.
3. Which sources it trusts. Assistants lean heavily on third-party sources that look neutral: directories, review platforms, "top 10" and comparison articles, industry bodies, and community threads like Reddit. A recommendation on your own site is marketing. The same claim on a review site or a respected list reads as evidence, and the AI weights it accordingly.
Notice what is missing from that list: keyword density, meta-tag tricks, and the tired bag of old SEO hacks. LLMs reward clarity, consistency and credible evidence, not repetition.
The GEO checklist: how to become the answer
1. Make your content answer-first and quotable
Assistants lift clean, self-contained answers. If your key pages bury the answer under three paragraphs of throat-clearing, the model cannot easily quote you, so it quotes someone clearer. Rewrite your important pages, your services, your pricing, your "how it works", so each opens with a direct, factual answer a machine can lift in one or two sentences. Put real numbers, ranges and comparisons in plain text (not locked inside images), because citable facts are exactly what a model reaches for.
2. Get your identity unambiguous
AI will not confidently recommend a business it cannot pin down. Make sure your name, address and phone number are identical everywhere they appear online, your Google Business Profile is complete, and, where you can, you have a presence in the structured knowledge graph the models read: a Wikidata entry, consistent sameAs links to your real social and directory profiles, and clear "who we are, where we serve, what we do" content. This is the difference between the AI saying "a Sydney landscaper" and the AI saying your actual name.
3. Add structured data and an llms.txt
Structured data (schema markup) spells out your business, services, prices and FAQs in a format machines parse without guessing. On top of that, a growing convention is llms.txt, a simple text file at your domain root that gives AI crawlers a clean map of your most important pages and facts. Think of it as a robots.txt for the AI era: it does not force anything, but it makes your canonical information trivially easy to ingest. We wrote one for this very site.
4. Win the sources the AI cites
This is the step most businesses skip, and it is often the highest leverage. Get genuinely well represented on the platforms assistants trust: the relevant directories, the review sites (with real reviews, not fabricated ones), the "best plumbers in", "top agencies for" style comparison pages, and the community threads where people ask for recommendations. You are not gaming anything here; you are making sure that when the AI goes looking for neutral evidence about your category, you are actually in the picture.
5. Make your business transactable by AI
Being recommended is step one. The next wave is customers asking the assistant to act: "find me a physio in Cronulla and book Tuesday afternoon." For that to reach you, an agent needs a safe, structured way to talk to your systems, which is what the Model Context Protocol (MCP) provides. Businesses that expose booking, quoting and availability to AI agents will win the "just sort it for me" requests. It is early, which is the whole opportunity.
6. Keep it fresh and honest
Assistants favour current, trustworthy sources, so a stale site with 2022 prices signals neglect. Keep your facts current, date your content, and never fabricate reviews or ratings to trick the model. Beyond being against every platform's rules, it backfires: the AI cross-checks sources, and a business whose self-claims do not match the independent evidence gets quietly dropped.
How to tell if it is working
You measure GEO differently from SEO. There is no single ranking to watch; instead you track your "share of AI answers". Pick the twenty or thirty real questions a customer would ask an assistant to find someone like you, then run them across ChatGPT, Claude, Perplexity and Gemini every month and record two things: how often you are mentioned, and how often you are cited as a source. As you work the checklist above, both numbers should climb. It is the same instinct as rank tracking, pointed at a new set of answers.
What this is not
It is not a magic prompt, and anyone promising a guaranteed answer from ChatGPT is bluffing, nobody controls exactly what a model says. It is not a replacement for Google either; classic SEO still drives most discovery, and the two overlap because both reward clear, authoritative, well-structured content. GEO is the additional layer that captures the buyers who have quietly switched to asking an assistant. Ignore it and you do not get penalised, you just get left out of a conversation that is growing every month.
Where to start
If you do nothing else this quarter: rewrite your top three pages to lead with a clear, quotable answer, make your business name and details identical everywhere online, and get yourself onto two or three of the review or "best of" pages your customers actually read. That alone moves the needle.
If you want it done properly, this is exactly what our LLM Recommendation Optimisation service handles, from citable content and entity clean-up to prompt-testing your visibility across every major model. Pair it with an MCP server so the assistants that recommend you can also book you, and keep your classic search foundations strong with programmatic SEO. Not sure where the budget should go? Our cost guides lay out real AUD ranges, or post a brief and compare three free quotes from vetted Australian developers within 24 hours.