How GEO works: from a client's question to an AI's answer
A cafe owner in Marrickville types a question into ChatGPT: "Who should do the books for my hospitality business?" Four seconds later she has three firm names and a sentence about each. Generative Engine Optimization is the work of making sure your firm survives everything that happens inside those four seconds.
You can only optimise a process you understand. So here is the process, step by step, as it applies to an Australian accounting firm.
Stage 1: The engine interprets the question
The engine first works out what the question needs. "Who should do the books for my hospitality business?" contains no suburb and no job title, so the engine expands it: this person wants a bookkeeper or accountant, probably nearby, with hospitality experience. If she is signed in, the engine may already know she is in Sydney.
This expansion decides which sources get considered. A firm whose site says "bookkeeping for cafes and restaurants in Sydney's Inner West" matches the expanded question almost word for word. A firm whose site says "comprehensive business advisory solutions" matches nothing in it.
Stage 2: The engine retrieves a shortlist
For anything local or current, the engine does not answer from memory. It runs a live search and pulls back a shortlist of pages: firm websites, Google Business Profiles, directories, review sites, sometimes a news mention or an industry association listing.
Two things decide whether you make this shortlist. Your pages have to be crawlable, meaning the engine's bots can fetch and read them without executing JavaScript. And your pages have to rank well enough in the underlying search index that the engine sees them at all. This is where GEO overlaps with traditional SEO: the retrieval step still runs on search infrastructure.
Stage 3: The engine reads and verifies
Now the engine reads the shortlist, and this stage kills most firms' chances. It reads like a sceptical researcher on a deadline. It wants facts it can extract and cross-check: what the firm does, who it serves, where it operates, registration details, review counts. When your website, your Google Business Profile, and two directories all agree on those facts, the engine gains confidence. When they disagree, or when the website offers nothing beyond "trusted advisors delivering tailored solutions", the engine moves on to a firm it can pin down.
Structured data earns its keep here. Schema markup states your services, location, and credentials in a format the engine parses without guesswork.
Stage 4: The engine writes and cites
The engine composes a short answer and names two or three firms. It tends to name the firms whose facts it extracted most cleanly, and it often quotes them near-verbatim: "Firm X specialises in hospitality bookkeeping across the Inner West." If your site handed the engine that sentence, you wrote the AI's answer yourself.
The firms that get named are the firms that made the engine's job easy at every stage: easy to find, easy to read, easy to verify, easy to quote.
Where the Australian layer matters
Engines match language, and Australian clients use Australian language. They ask about BAS lodgement, not VAT returns. They ask whether a firm is a registered tax agent, not a chartered practice. They mention the ATO, tax time, super, and their suburb. A firm whose content is written in those terms matches Australian questions; a firm running generic templated content matches nothing local.
The same goes for corroboration. Australian engines' shortlists lean on Australian sources: your Google Business Profile, Australian directories, the Tax Practitioners Board register, local review activity. Getting those consistent is Australian-specific work, and it is why we only operate in this market.
What GEO work looks like in practice
- Measure first. Run the real questions your prospects ask across each engine, repeatedly, and record who gets named. Answers vary run to run, so single spot checks mislead; averages across repeated runs tell the truth.
- Fix retrieval. Make every important page crawlable, fast, and readable as plain HTML.
- Fix extraction. Rewrite key pages around real questions, state citable facts, add schema markup.
- Fix verification. Reconcile your details across your site, Google Business Profile, and the directories engines actually check.
- Re-measure. Track your share of AI answers against named competitors, month over month.
None of these steps is exotic. The discipline is in doing them against measured data rather than guesswork, which is why every Reconcite engagement starts with a visibility audit rather than a proposal.