AI is 30× harder than Google. Most retailers haven't noticed yet.

I run a small SEO agency in Canberra. We work with about 42 businesses across Australia, most of them in retail, hospitality, and trades. For years the playbook was simple. First, get the Google Business Profile right. Then build citations. Earn reviews. Show up in the 3-pack. Win.

And that playbook still works. But it's now covering maybe two-thirds of the game.

So the other third is something I've been calling the local visibility crisis, and I think most retailers and service businesses are walking into it blind.

What the numbers actually say

SOCi published their 2026 Local Visibility Index in March. They tracked 350,000 business locations across 2,751 brands and ran them through Google, ChatGPT, Gemini, and Perplexity to see who got recommended.

The gap is brutal. Google's local 3-pack surfaces 35.9% of those locations. Gemini drops that to 11%. Then Perplexity at 7.4%. And ChatGPT? Just 1.2%.

That's the headline stat. But the line that should worry every multi-location operator is this one: less than half of the businesses winning in Google local search also appear in AI recommendations. Winning one game does not mean you win the other.

Meanwhile BrightLocal's consumer survey backs it up from the demand side. Back in 2025, only 6% of consumers said they used AI to find a local business. By 2026, that number jumped to 45%. So AI is now the third most popular local discovery channel in less than 12 months.

In other words, you've got demand surging and selectivity tightening. The maths is unforgiving.

Why this is happening

Each LLM is grounded differently, and that changes everything about how you optimise for it.

So Google's Gemini pulls almost entirely from Google Maps. If your GBP is solid, you're in decent shape there. But ChatGPT is a different animal. According to BrightLocal's source analysis, ChatGPT pulls 58% of its local recommendations from business websites, 27% from press mentions, and 15% from directories. Notably absent: Yelp, Facebook, and Google Maps itself.

So that means everything you spent the last decade building for Google can be largely irrelevant to ChatGPT. Instead the platform is reading your website, your unstructured citations, your editorial mentions, and your presence on niche directories like Three Best Rated and Expertise.com.

And here's the kicker. An Ahrefs study from August 2025 found that 80% of LLM citations don't even rank in Google's top 100 for the same query. Now the traditional SEO winners and the GEO winners are often two different sets of websites entirely.

The review threshold nobody is talking about

I've been studying review patterns across our client base, and the data lines up with what MyPlace found in their 2026 restaurant analysis. ChatGPT-recommended brands have on average 3.6 times more reviews than non-recommended ones. Specifically, the recommended group averaged 3,424 reviews. Non-recommended? 955.

Now the threshold seems to sit around 2,000 reviews for competitive urban categories. Below that, you can have a 4.9-star average and still not get cited. The model just doesn't have enough signal to feel confident about you.

So for a Canberra cafe with 180 reviews, this is a wake-up call. But for a multi-location retailer with 30 locations averaging 200 reviews each, it's an existential one.

Meanwhile Whitespark put review recency in their top 5 ranking factors for 2026, up from #20 in 2023. The rule of thumb their team is using now is simple. First, match your top competitor's review pace. Then add one per period. Stop, and rankings actively decay.

What's actually working

We've been testing different GEO interventions across clients for about 14 months, and a few patterns are holding up.

First, Wikipedia and Wikidata entity establishment matters more than people think. Wikipedia accounts for roughly 39% of business mentions cited in ChatGPT, according to BrightLocal's tracking. So if your brand has no entity, you're invisible at the entity layer.

Second, unstructured citations beat structured ones for AI. Whitespark's research shows that mentions in local news, podcasts, and "best of" expert lists drive AI visibility more than NAP citations in directories. Now the directory game still matters for Google. But for ChatGPT and Perplexity, editorial mentions punch harder.

Third, passage-level content engineering is real. Aleyda Solis has been pushing this for a year now. LLMs don't read your page the way Google does. Instead they chunk it and grade individual passages for relevance. So a 2,000-word page with one strong 80-word answer will often outperform a meandering 4,000-word piece. Mike King at iPullRank calls this Relevance Engineering, and the framework is gaining traction fast.

Then the Princeton GEO paper from KDD 2024 quantified what works. Adding statistics, citing authoritative sources, including direct quotes, and writing in a fluent authoritative tone lifted citation rates by up to 40%. Meanwhile keyword stuffing and traditional SEO tactics did basically nothing for LLM visibility.

The Australian angle

Australia is behind the US on AI adoption, but not by much. Roughly half of Australian adults have used generative AI in the past year. So far ChatGPT is the most-named tool by a wide margin.

But what concerns me about the Australian agency landscape is how few players have productised GEO offerings. I've been watching the market closely and I count maybe a dozen agencies nationally with a real GEO product. Meanwhile most are still selling traditional SEO retainers and Google Ads management. Still, the whitespace is enormous.

Then the other Australian dynamic worth flagging is the Square Neighbourhood Nation data from August 2025. Today 73% of monthly dining spend goes to neighbourhood venues rather than CBDs. Around 46% of employed Australians work from home at least sometimes. Here in Canberra, the CBD WFH rate sits at 61%, among the highest in the country.

So this means the local discovery question is not abstract. It's where 73 cents of every restaurant dollar is being decided. And increasingly, it's being decided inside an LLM that pulls from sources most retailers have never heard of.

Where to start

If you operate a retail or service business and haven't tested your AI visibility, do this today. First, open ChatGPT. Ask it for the best three options in your category in your suburb. See if you appear. Then do the same in Gemini, Perplexity, and Google AI Mode.

If you're not there, you have a problem that won't fix itself. Meanwhile reviews keep flowing to your competitors. Editorial mentions accumulate elsewhere. And Wikipedia entries get written about other brands.

So the retailers winning the next five years will be those whose digital presence matches the operational reality of their physical catchment. Right now, very few do. That's both the warning and the opportunity.