37% of consumers now start their search with an AI tool rather than Google. The two work completely differently. This article explains how AI search actually works and how to test whether your business is appearing in the recommendations that now drive buying decisions.
Something changed in the past two years in how buyers find businesses. Most business owners missed it.
A growing share of buying journeys now start with an AI tool — not a Google search. A 2025 Semrush study of over 1,000 consumers found that 37% now start their search process with an AI tool rather than Google. For professional services and B2B buyers, the proportion is higher.
Understanding what changed is the first step to knowing whether you have a problem.
How Google works versus how AI search works
These two systems work completely differently. Most business owners assume they work the same way.
How Google works:
Think of Google like a phone directory. You tell it what you are looking for. It shows you a list of options — usually ten links. You pick one to investigate. Then another. Then you decide.
The buyer does all the decision-making work. Google just gives them the list.
How AI search works:
Think of AI search like asking a well-informed friend for a recommendation. You describe your situation. Your friend thinks about it and names one or two specific businesses. You trust their judgment and contact those businesses directly.
There is no list. There is one answer — with names.
In Google, being seventh on the page still means you are visible. In AI search, if you are not in the answer, you do not exist in that conversation. Full stop.
What the numbers tell us
The volume difference between the two systems shows how differently they are being used.
ChatGPT handles about 12% of Google's total search volume. But AI tools send approximately 190 times less traffic to websites than Google does. People are getting their answer inside the AI tool and not clicking through to websites at all.
That sounds like bad news. For businesses relying on website traffic, it is.
But here is the counterintuitive part: the buyers who do contact businesses after an AI recommendation are a completely different quality of prospect.
Research from Bain & Company found that 98% of consumers verify an AI recommendation before making contact — but 45% verify by Googling the business name directly. The sequence looks like this:
- AI tool names your business
- Customer Googles your name to confirm you are real and credible
- Customer contacts you
These are not browsers. They are buyers. Semrush found that traffic arriving from AI recommendations converts at 14.2% — compared to 2.8% for people arriving from Google searches. These visitors are already decided. They are looking for confirmation, not discovery.
The test most business owners get wrong
Most business owners test their AI visibility by typing their own business name into ChatGPT.
That is the wrong test.
When you type your own name, the AI tool treats it as a request for information about your business. It tells you what it knows — not whether it would recommend you to a stranger.
Real customers do not type your name. They do not know your name yet. They type what they need.
"Best accountant for small business in Brisbane" is a recommendation query. ChatGPT then decides which businesses to name. If you are not named, you are invisible to that buyer — even if ChatGPT knows you exist.
How to run the right test
Write five queries in the language a motivated buyer would use. The format is: what the customer needs + their location or qualifier.
Five examples across different industries:
- "small business accountant Brisbane who specialises in hospitality"
- "employment lawyer Melbourne unfair dismissal small business"
- "web designer for tradies Sydney"
- "bookkeeper for medical practices Canberra"
- "financial planner for women nearing retirement Adelaide"
The specificity matters. Broad queries return generic answers. Specific queries reveal whether the AI tool can match you to a real customer need — which is exactly what your potential customers are actually typing.
Which AI tools to test
Six tools matter for Australian buyers right now:
- ChatGPT — the most used AI tool in Australia. Use the default chat mode, not the research or image tools.
- Perplexity — checks live websites, so it reflects your current online presence faster than other tools.
- Gemini — Google's AI tool. It connects directly to your Google Business Profile, Maps listing and reviews. A strong Google presence translates here.
- Claude — growing fast among knowledge workers and professional services buyers. Tends toward careful, considered answers.
- Grok — X's (formerly Twitter's) AI tool. Less relevant for most Australian businesses, but worth including for a full picture.
- DeepSeek — free and growing in adoption, particularly in Asia-Pacific.
Open each tool in a browser tab. Paste your five queries one at a time. Record the result for each. The full test across six tools takes 20 to 30 minutes.
What the results mean
Every result falls into one of three categories.
Recommended — the tool names you with positive context. Something like: "Smith & Co is a strong option — they specialise in X and are known for Y." This is the outcome that sends pre-qualified buyers to your website.
Mentioned — your name appears somewhere in the response without an endorsement. The tool knows you exist but is not confident enough to back you. Better than missing, but it does not drive the same action.
Missing — you do not appear at all. The tool either cannot identify you clearly, cannot find outside confirmation of you, or simply has better-described alternatives it prefers.
What three common result patterns mean
Missing on one or two tools, visible on others.
This is a tool-specific gap. Missing from Perplexity often means your live website presence is weak — few mentions on other sites, or outdated content. Missing from Gemini often points to an underdeveloped Google Business Profile.
Missing across all tools.
This almost always means the tools cannot clearly identify who you are, or cannot find outside sources that confirm you. The fixes are foundational: rewrite your homepage main headline, add background identity code to your website, create a guide file for AI tools, and get your business listed in credible directories.
Mentioned but not recommended on every tool.
The tools know you exist but lack the confidence to endorse you. Common causes: thin or stale reviews, very few mentions on other websites, or not enough written content in your field for the tool to cite.
What to do next
Work in order. Fix your identity signals first — the headline on your homepage, the background code on your website. Then build trust signals — reviews, mentions on other sites. Content depth comes last.
The article on why ChatGPT doesn't recommend your business walks through each cause with specific fixes. The AI visibility audit article covers how to check all four layers in order.
Start with the test. Know where you stand. Then fix the layer that failed first.
Frequently asked questions
Is AI search replacing Google?
Not entirely, but it is changing how buyers use Google. Research from Bain & Company found that around 80% of search users rely on AI summaries at least 40% of the time. The bigger shift is in where decisions are made — AI tools are increasingly where buyers decide which specific business to contact. Google is becoming a verification step rather than a discovery tool.
Why does testing my own business name not work?
When you type your own business name into ChatGPT, it treats your query as a request for information about your business — and tells you what it knows. That is not how a potential customer uses it. Customers ask need-based questions, like "who should I use for X in Y," and the engine decides which businesses to name. Testing your own name checks recognition, not recommendation. Those are two very different things.
How long does it take for AI engines to reflect changes to my website?
It depends on the tool. Perplexity checks live websites and can reflect changes within days. ChatGPT relies more on its original training data and typically takes two to six weeks for changes to show up. Background code changes tend to surface faster than written content changes because they are processed differently.
Should I test in incognito mode?
It makes little difference for AI tools. Unlike Google, ChatGPT and Perplexity do not personalise results based on your browsing history. What matters is testing with queries written in customer language rather than your own business name.
How often should I re-test?
Once a month when you are actively making changes. Once your core fixes are in place and results have stabilised, quarterly is enough. Testing the same queries each time gives you a useful comparison — testing at random makes it hard to know whether your changes helped.
