Someone in Hawthorn asks ChatGPT “best accountant near me for a small trades business.” Instead of ten blue links, ChatGPT hands them a paragraph answer with three named firms and links to the sources it pulled from.
If your accounting practice isn’t one of the three, you don’t exist in that conversation. There’s no page two to scroll down to.
That’s the thing most small business owners I talk to haven’t quite clocked yet: AI search is already here, your competitors are showing up in it, and being quotable enough to make it into the answer is now part of the job. That’s what GEO is about.
So before anyone sells you an AI SEO package, look for yourself. It takes about 25 minutes and a notepad.
Start here
Outcome: find out whether AI tools name your business when customers ask, then fix the basics that make you easier to name. Time: about 25 minutes to check, about two hours to fix. Cost: nothing, if you write the schema yourself.
Do first: the AI check · the homepage opening sentence · your Google Business Profile · crawlability · one real FAQ section.
Leave for later: llms.txt · new “GEO tools” · ongoing “AI SEO” packages.
Done when: you know how often the AI tools name you, your main pages say what you do, who you do it for, and where, and your key pages are crawlable and indexable.
What is GEO?
Generative Engine Optimisation (GEO) is the practice of structuring your website content so that AI tools like ChatGPT, Perplexity, Google AI Overviews, and Gemini can find it, understand it, and cite it in their answers. The term is usually credited to a 2023 paper by Aggarwal, Murahari and colleagues at Princeton, which introduced GEO as a discipline and measured which techniques actually improve citation rates in generative search (arXiv:2311.09735, published in KDD 2024).
The short version: AI tools read across many sources, pick the ones that give them a clear, specific answer, and then paraphrase or quote those sources in their response. Getting cited is the new getting ranked. GEO is how you make that more likely. More likely is the honest framing, too. Nobody can promise you a citation, and anyone who does is selling something.
GEO sits on top of SEO rather than replacing it. You still want to rank in traditional search. You just also want to be the source the AI tool pulls from when someone asks a question in natural language instead of typing keywords into a box.
And the people asking are already here. ChatGPT crossed 900 million weekly active users in February 2026, and a survey Google Australia published found 49% of Australians used generative AI in the past year, up from 38% in 2023. Some slice of that will be exactly the questions a small business in Fitzroy or Coburg wants to show up for: who’s good near me, and who should I call.
Check what the AI tools say about you (about 25 minutes)
This is a cut-down version of the manual AI-search check I run in a Digital Health Check, something you can do with a cuppa and a notepad. You don’t need any tools or logins beyond the free versions.
1. Write five questions your customers would actually ask (5 min)
Use their words, not yours, and put a place in most of them, because that’s how people ask. Three “who should I use” questions (“who’s a good bookkeeper for tradies in Preston?”), one comparison (“is it better to use a local accountant or an online one for a small café?”), and one that names you directly (“what do you know about [your business name] in [your suburb]?”). That last one tells you whether the tools know you exist at all, and whether what they know is right.
2. Ask each question a few times in each tool (15 min)
Run all five through ChatGPT, Gemini and a normal Google search (where you’re looking for the AI Overview that sometimes sits at the top). Ask each question at least twice per tool.
Try this: in ChatGPT, open a temporary chat and choose Unpersonalized before you type the first question. OpenAI’s help page says that version doesn’t use your memory or custom instructions, so the answer isn’t shaped by everything you’ve told ChatGPT before.
The repeats aren’t me being fussy. SparkToro had 600 volunteers run the same prompts through ChatGPT, Claude and Google’s AI a combined 2,961 times, and found that, asked 100 times, there was less than a 1 in 100 chance that ChatGPT or Google’s AI would give the same list of brands in any two responses (January 2026). What did hold up across runs was how often a brand appeared at all. So count appearances, and ignore the order.
3. Write down what came back (5 min)
For every run, note whether you were named, who else was, anything said about you that’s wrong, and which sources the answer linked to. There’s a copy-ready table for this in Tools & Resources. Screenshot anything odd, with the date, because you’re unlikely to see the exact same answer again.
How to read what you found
| What you found | What I’d do |
|---|---|
| Named in most runs, and described accurately | Not much. Re-run the check in a few months to see if it holds. |
| Named, but with the details wrong (old hours, a service you dropped, the wrong suburb) | Fix it where you’re described: your own website first, then your Google Business Profile and any directory listings, so they all say the same thing. |
| Not named, and competitors are | Open the pages the answers linked to and note what they say plainly that yours don’t. That gap is your to-do list, and the plan below covers most of it. |
| Not named, and nobody local is either | The answers are generic, so the field is open. The plan below is how you get in early. |
| Asked about you by name, and the answer is blank or muddled | The tools can’t tell who you are yet. Make your homepage, About page and Business Profile state the same name, address and services in the same plain words. |
None of this is a ranking, and a handful of runs is a rough sample. It’s still the most honest picture you’ll get for free.
What’s worth doing, and what I’d skip
Start with what Google itself says, because it’s blunter than the “AI SEO” industry would like. In May 2026, Google published an official guide to optimising for AI features in Search. Google says its AI features in Search are rooted in its core Search ranking and quality systems, and they pull pages from the same Search index. In Google’s words: “To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. There are no additional technical requirements.”. It also says you don’t need llms.txt (“Google Search itself doesn’t use them”), a special “AI-friendly” voice, or dedicated schema markup for AI answers.
So for Google, good SEO is still the groundwork. The tactics below earn their keep mostly with ChatGPT and the other tools, which each appear to behave a bit differently (treat any per-tool rule of thumb as a pattern that will shift). They also make the page better for people, which is why I’d do them anyway.
What I’d prioritise, and what I’d skip for now
| What I’d prioritise | What I’d skip for now |
|---|---|
| Answer-first formatting in every section | Buying an “AI SEO” package for $2,000 a month |
| One-paragraph definitions of what you do | llms.txt |
| FAQ sections with direct answers | Chasing every new “GEO tool” |
| Fresh content with clear dates | |
| Specific, verifiable numbers |
Answer-first formatting in every section. If someone only read the first two sentences of each section, they should get the answer. This matches the research and, honestly, just makes the article better.
One-paragraph definitions of what you do. Your homepage, your service pages, your About page. Each should open with a single sentence an AI tool can quote verbatim. “We’re passionate about delivering solutions” is not quotable. “Nathan Schram is a Melbourne-based digital marketing consultant specialising in SEO, GEO, and Google Ads for small businesses” is.
FAQ sections with direct answers. FAQ content maps almost exactly to the kind of question people ask AI tools. Google retired the FAQ rich result in May 2026, so FAQPage markup no longer earns you the expandable Q&A box in Google’s results. The value now is the AI-citation angle and the clean structure, not a Google SERP feature. Write FAQs. Add the schema where the questions are real. The cost is low.
Fresh content with clear dates. Several AI engines appear to weight recency, Perplexity especially, so every article I publish has a “last updated” date.
Specific, verifiable numbers. “We’ve helped many businesses” is worthless. “I’ve audited Google Ads accounts for trades, healthcare, real estate, and professional services since 2013” is fact-checkable and citable. It’s also where the research points: in the Princeton tests, citing sources, adding statistics and using quotations did most of the heavy lifting, and certain techniques boosted visibility in AI responses by up to 40%.
Buying an “AI SEO” package for $2,000 a month. Most of what’s in those is either standard SEO or basic schema implementation. If your site has good structure, clear writing, and current schema, you’ve already done most of the groundwork. The rest is measurement, which barely any tool does well yet.
Skip llms.txt. The llms.txt proposal from Jeremy Howard at Answer.AI in September 2024 is a tidy idea: a text file telling AI crawlers what matters on your site. In May 2026 Google stated flatly that its Search systems don’t use it, and Ahrefs looked at 137,000 sites and found that 97% of the llms.txt files they published got zero requests in May 2026. John Mueller compared it to the old meta-keywords tag. I’d file it under “don’t bother” for most sites, and I wouldn’t pay anyone to strategise about it.
Chasing every new “GEO tool.” Most tools show you mentions in some subset of AI tools, extrapolate wildly, and sell you a dashboard. After the SparkToro numbers above, I’d be wary of any tool that reports a single “AI ranking” for your business.
The two-hour fix list
Here’s what I’d actually do if you gave me a Saturday morning and a Melbourne small business website that came out badly in the check above.
1. Rewrite the homepage opening paragraph 15 min
One sentence that says what you do, who you do it for, and where, in plain facts an AI tool could quote rather than words like “passionate”.
Done when: the homepage opens with one sentence an AI tool could quote: what you do, who you do it for, and where.
2. Claim and fill in your Google Business Profile 20 min if you haven’t, 5 min to double-check if you have
Hours, categories, photos, description. For local queries this matters: Google says you can use your Business Profile to improve your local ranking. In my experience auditing Melbourne businesses, plenty of the sites I look at haven’t claimed their Google Business Profile properly, or they claimed it years ago and never touched it again. If that’s you, the local SEO guide goes deeper.
Done when: your profile is claimed, and the hours, categories, photos, and description are filled in.
3. Check your key pages are crawlable and indexable 10 min
No accidental noindex, nothing important blocked in robots.txt, and your main content sitting in the HTML rather than hidden behind JavaScript. Indexability is the gate Google’s own guide names for appearing in an AI answer, and content in the HTML also helps any crawler that doesn’t run JavaScript.
Try this: paste your homepage address into the URL Inspection tool in Google Search Console. It tells you whether Google has the page indexed, and if not, why. You’ll need Search Console set up for your site first.
Done when: no key page carries an accidental noindex, nothing important is blocked in robots.txt, and your main content sits in the HTML.
Stop if: a key page turns out not to be indexable. Fix that before the rest of the list.
4. Write one proper FAQ section 30 min
Pick five to eight questions real customers actually ask (the five from your check are a good start) and answer each one directly. The questions and answers are the value; FAQPage schema on top is optional polish now that Google has retired the FAQ rich result.
Done when: the page directly answers five to eight questions real customers actually ask.
5. Add a “last updated” date to every piece of content 10 min
Recency appears to be a signal for several AI engines.
Done when: every piece of content shows a “last updated” date.
6. Add LocalBusiness or Person schema 35 min
Schema markup is a small block of code that tells search engines, in a fixed format, who you are and where. Microsoft’s Fabrice Canel confirmed at SMX Munich in March 2025 that schema markup helps the LLMs behind Bing’s AI tools understand your content. Google’s May 2026 guide says structured data isn’t required for AI search and there’s no special markup to add, but it’s still worth using because it helps you qualify for rich results in regular search. So add it because it’s cheap, not because it’s a magic AI-citation lever. The only real downside is neglect: markup has to match what’s visibly on the page, and stale or wrong markup is worse than none.
Use LocalBusiness on the homepage and contact page if you have a shopfront or serve an area, or Person if you are the business. Drop it into the <head> of the page, inside <script type="application/ld+json"> tags, then test it with Google’s Rich Results Test and the Schema Markup Validator.
LocalBusiness, for a Melbourne small business:
{
"@context": "https://schema.org",
"@type": "LocalBusiness",
"name": "Your Business Name",
"image": "https://example.com/storefront.jpg",
"address": {
"@type": "PostalAddress",
"streetAddress": "123 Brunswick St",
"addressLocality": "Fitzroy",
"addressRegion": "VIC",
"postalCode": "3065",
"addressCountry": "AU"
},
"geo": {
"@type": "GeoCoordinates",
"latitude": -37.7963,
"longitude": 144.9780
},
"telephone": "+61-3-9000-0000",
"url": "https://example.com",
"openingHoursSpecification": [{
"@type": "OpeningHoursSpecification",
"dayOfWeek": ["Monday","Tuesday","Wednesday","Thursday","Friday"],
"opens": "09:00",
"closes": "17:00"
}],
"priceRange": "$$"
}Person, for a consultant or sole trader (this is roughly mine):
{
"@context": "https://schema.org",
"@type": "Person",
"name": "Nathan Schram",
"url": "https://nathanschram.com",
"jobTitle": "Digital Marketing Consultant",
"description": "Melbourne-based digital marketing consultant specialising in SEO, GEO, and Google Ads for small businesses.",
"address": {
"@type": "PostalAddress",
"addressLocality": "Melbourne",
"addressRegion": "VIC",
"addressCountry": "AU"
},
"sameAs": [
"https://www.linkedin.com/in/nathanschram",
"https://bsky.app/profile/nathanschram.bsky.social",
"https://github.com/littlebearapps"
]
}Done when: the Rich Results Test shows no red errors on the pages you added it to.
That’s about two hours, and none of it guarantees a citation. The writing steps are the part that lines up with what the Princeton research measured (clear, specific, well-sourced content); the schema and Business Profile steps are ordinary SEO and local hygiene that Google says still matter. Nothing here leaves you worse off than you are now.
Then leave it alone for a while. I’d give it at least a month before re-running the check, and when you do, compare how often you’re named, not where you appear in the list.
What won’t save you?
Schema alone isn’t magic. Schema is often credited with big citation lifts. The evidence is thinner than the claims: a Search Atlas analysis of schema coverage and citation rates across OpenAI, Gemini, and Perplexity found domains with extensive schema markup were cited no more often than domains with little or none.
The “AI-first” funnel is still tiny. Most small businesses I work with still get the majority of their traffic from Google, direct visits, and paid ads. AI referral traffic is real and growing, but for a typical Melbourne trade business it’s still a small slice of sessions, not half of them. Don’t take my word for the size either: GA4’s “AI Assistant” channel will show you your own number, read as a floor rather than the full picture. Don’t neglect your Core Web Vitals or your Google Ads setup chasing an AI-only strategy.
Tools & Resources
Free tools
- ChatGPT temporary chat: choose Unpersonalized before your first message so the check isn’t coloured by your own chat history.
- URL Inspection in Google Search Console: shows whether a page is in Google’s index, which is the gate for appearing in Google’s AI answers.
- Rich Results Test and Schema Markup Validator: check the schema on a page for errors before and after you change it.
Checklists & templates
The AI check record sheet. Copy it into a spreadsheet or a notepad, one row per run:
| Date | Question | Tool | Run | Named? (Y/N) | Who else was named | Anything wrong about you | Sources linked |
|---|---|---|---|---|---|---|---|
| ChatGPT | 1 | ||||||
| ChatGPT | 2 | ||||||
| Gemini | 1 | ||||||
| Google AI Overview | 1 |
Further reading
- Google’s guide to AI features in Search: what Google says you need (and don’t) to appear in AI Overviews and AI Mode.
- SparkToro’s study on how inconsistent AI recommendations are: the reason the check asks every question more than once.
Frequently asked questions
Do I need to do GEO if I already rank well in Google?
Yes, because ranking well in Google and being cited in Google AI Overviews are not the same thing. A page can be in position 1 for a keyword and still not appear in the AI Overview for the same query. That said, if your SEO fundamentals are strong, your GEO starting point is already better than most.
How do I measure if GEO is working?
It’s harder than SEO. Google Analytics now has a dedicated “AI Assistant” channel for visits from tools like ChatGPT and Gemini. Treat it as a floor rather than the full picture: it leaves out Google’s own AI Overviews and AI Mode. Search Console counts clicks and impressions from those Google AI features inside its normal Performance report, under the Web search type. Beyond that, re-run the check above every month or two. Nothing yet measures the people who read an AI answer and never click.
Can I just use ChatGPT to write my GEO content?
You can, and plenty of people do, but the Search Engine Land 16-month study of 2,000 AI-written articles, run on brand-new domains with zero authority, found that only 3% of them were still in Google’s top 100 after about three months (they clawed back some ground later, but most never returned). Write it yourself, or get a human to edit AI drafts properly. The Princeton research is very clear that specific, factual, well-cited writing wins. Generic AI copy is the opposite of all of that.
What’s the minimum budget for GEO?
Two hours of your time and zero dollars, if you follow the weekend plan above. If you want an expert audit and implementation, a decent local consultant (shameless plug: me) will do a thorough GEO pass on a small business site for a few hundred to a couple of thousand dollars, depending on size. Don’t pay ongoing $2,000-a-month “AI SEO” fees unless someone’s actually proving the traffic.
Rather not do it yourself?
If you’d rather not spend the Saturday on it, the free Basic Digital Health Check includes manual AI-search checks like these, and I’ll walk you through what came back on the call. For a proper GEO audit or help with the schema, get in touch.
Nathan Schram is a Melbourne-based digital marketing consultant with 14 years in tech marketing, sales, and business development across startups and ASX-listed organisations. He writes about Core Web Vitals, Google Ads, and local SEO for Melbourne businesses. Read more at About.
Originally published April 2026; updated June 2026 to fold in Google’s first official AI-search guidance (and to retire the llms.txt advice Google has since said it doesn’t use), July 2026 to tighten several statistics and claims against their original sources, and October 2026 to lead with a check of what AI tools say about your business and to trim the general AI-search statistics. AI search data moves fast, so check the linked sources for the latest figures.