How do I check AI visibility for an Australian small business?

You check AI visibility by running a fixed set of 10 buyer-style questions across six AI engines, ChatGPT, Claude, Gemini, Perplexity, Grok, and DeepSeek, for 60 measurements per scan. Then score each answer. Note whether your business shows up, whether the facts are right, and whether the reader gets a clear next step. The work is simple. It takes about two hours the first time. You end up with a baseline you can re-test in 30 days.

If you have asked ChatGPT about your own industry and watched it skip your business, you are not alone. Most Australian small businesses were built for Google in 2018. They were not built for AI search in 2026. The good news is the check is easy. The fix takes longer.

Why this matters now

Australian small businesses are buying AI tools faster than they are being found by them. The Department of Industry, Science and Resources reports that about 40 percent of SMEs were using AI by Q4 2024. The smallest firms, those with up to four staff, jumped from 25 to 34 percent in a single quarter. The Reserve Bank's November 2025 Bulletin shows AI spend rising across nearly every sector.

That is the supply side. On the demand side, Australians use AI tools to find businesses too, not just to write captions. Pew tracking shows Google AI Overviews now appear in about one in five searches. When an Overview shows up, clicks on the normal blue links drop by about half. Deloitte Australia's November 2025 report puts the gap from underused AI at 44 billion dollars across the small-business sector.

Here is the part that stings. MYOB's bi-annual Business Monitor finds 41 percent of Australian SMEs still have no online presence at all. If a business is invisible to plain Google search, it is invisible to AI search too. The AI engines pull from the same web Google reads.

You do not need to fix every problem this week. You just need to know where you stand.

What "AI visibility" actually means

AI visibility is the share of buyer questions where an AI engine names your business, links to it, or cites it as a source. It is close to Google ranking, but it is not the same.

Three things make it different.

First, AI engines summarise rather than list. A Google search shows ten links. A ChatGPT answer often names two or three businesses and skips the rest. The cut-off is sharper.

Second, the same question asked twice can give different answers. Search Engine Land's 2026 LLM tracking shows brand mention rates can swing by more than 40 percent across runs of the same prompt. You need more than one sample to trust the result.

Third, structured data matters more. Princeton's peer-reviewed GEO paper (Aggarwal et al., KDD 2024) found that adding clear stats and citations to a page lifted the chance an AI engine would quote it by 37 percent. Schema markup, named expertise, and clear numbers are signals AI engines weight heavily now.

The four-step check

Use this order the first time. It takes about two hours.

1. Pick a prompt set that mirrors how a real buyer would ask.
2. Run each prompt on each engine, more than once, in fresh sessions.
3. Score what you find across five simple lenses.
4. Find the gaps an AI engine would need closed before it could name you.

Each step is below.

Step 1: Pick your prompt set

The working baseline used in standard AI visibility audits is 10 buyer-style prompts. Ten is enough to see a real pattern. Fewer than that and you are drawing conclusions from anecdote. Many more and you are paying for repetition without learning more.

Mix three intent levels.

• Top of funnel. Broad questions a stranger might ask. "What are the best bookkeepers for tradies in Sydney?" "Who should I hire for a small fit-out in Brisbane?"
• Mid-funnel. Compare and shortlist questions. "Compare three local accounting firms in Newcastle for a sole trader." "Which marketing agencies in Hobart work with cafes?"
• Bottom of funnel. Direct brand checks. "Tell me about [your business name]." "What does [your business name] do and where are they based?"

A balanced set looks like 4 top-of-funnel, 4 mid-funnel, and 2 bottom-of-funnel.

Write the prompts in plain words. Write them the way your customer would type them, not the way you would. If you sell a service, name the suburb or city. AI engines weight local context heavily.

Save the 10 prompts in a notes app or a sheet so you can re-run the same set in 30 days. The number stays at 10. Only the content of the prompts changes if your services change.

A bookkeeper in Wagga ran her first 10-prompt scan. She showed up on Perplexity twice, on ChatGPT once, and on the other four engines not at all. Her gut said she was "doing fine online." The data said the opposite. That gap, between gut feel and a real baseline, is the whole point of the exercise.

Step 2: Run each prompt on each engine

Use six chat engines. Together they cover the bulk of AI search behaviour. They act differently and pull from different data, so a business can be named on one and missed on the others.

• ChatGPT (free or Plus). Sends the largest share of AI chatbot traffic to websites in 2026.
• Claude (claude.ai). Anthropic's chat product. Strong on careful reasoning and citations.
• Google Gemini (gemini.google.com). Google's standalone chat product. Distinct from AI Overviews.
• Perplexity. Search-and-cite model. Always shows its sources.
• Grok (x.com). xAI's chat product, currently inside the X platform.
• DeepSeek. Open-weights model with a free chat interface, growing share in technical and B2B queries.

10 prompts on 6 engines gives you 60 measurements per scan. That is the working baseline.

Two related surfaces are worth a separate quick check, even if they sit outside the six-engine baseline:

• Google AI Overviews, the AI summary that appears inside Google search results.
• Microsoft Copilot, embedded in Bing and Microsoft 365.

For each prompt, on each engine, do this:

1. Open a fresh session or an incognito tab. AI engines remember context.
2. Ask the prompt as written. Do not soften it.
3. Run the same prompt at least twice. Variation matters.
4. Save the answer. A screenshot is fine.
5. Note whether your business is named, linked, or cited as a source.

Most owners are surprised by how often the answer is "no." Take a breath. The point of the check is to know.

Step 3: Score what you find

Five lenses cover the ground. Tally each prompt against each one.

• Visibility. Are you named at all? Yes or no.
• Favourability. When named, is the tone positive, neutral, or negative?
• Factuality. Are the basics right? Address, services, hours, founder name.
• Actionability. Did the engine give the reader a clear next step (a link, a phone number, a citation)?
• Consistency. Does the same prompt give a similar answer across runs and across engines?

A simple grid works. Prompt down the side, engine across the top, score in each cell. Colour the misses red so the pattern jumps out.

You are not chasing perfection. You are looking for patterns. A common one is "named on Perplexity, missing on ChatGPT." That tells you the issue is likely weak citations, not weak basics. Another is "named on ChatGPT, but with the wrong address." That is a factuality problem driven by stale third-party listings.

Step 4: Identify the gaps

Three gaps cover most cases.

Gap one: nothing for the AI to find. This is the MYOB 41 percent. No website, or a site with thin pages. The fix is foundational. A clear Home, About, Services, and Contact page, written as real text rather than baked into images, is the minimum.

Gap two: a website exists but the structured data is missing. Open one of your pages, view the source, and search for application/ld+json. If you find nothing, you have no schema. Google's own Search Central docs name Organization schema as the baseline for any business. They name LocalBusiness schema for any business with a shopfront. AI engines lean on this markup hard for entity matching. A small business site without LocalBusiness schema is, in effect, telling AI engines almost nothing about itself in machine-readable form.

Gap three: few or no citations from sources AI engines trust. AI engines pull from real media, peer-reviewed research, official directories, and well-cited articles. If your business is not named in any of those, you have what GEO researchers call a citation desert. The fix is slow but it lasts. Get listed in real industry directories. Write one or two original pieces for outlets your customers read. Make sure your About page names a real person with real credentials.

What good AI visibility looks like for an Australian small business

There is no official benchmark yet. Here is a working one based on what shows up in healthy first scans.

• Named in 60 percent or more of bottom-of-funnel prompts. If "tell me about [business name]" cannot find you, the brand has no AI footprint.
• Named in 20 to 40 percent of mid-funnel prompts. Compare questions are tough. A spot in this range means you are in the consideration set.
• Named in 5 to 15 percent of top-of-funnel prompts. This is the hardest tier. A small business should not expect to win broad queries.
• Factuality at or near 100 percent when you are named. A wrong address or a wrong service line is worse than a miss.

Re-test in 30 days, then again at 90. A baseline you cannot compare to is just a snapshot.

Tools and resources

You can do the whole check by hand with a notes app and the six engines above. That is the cheapest option and it teaches you the most.

If you want to automate the prompt runs and scoring, a few products track AI visibility for a monthly fee. Tools in this space include HubSpot's AEO grader, Otterly, Peec AI, and Get Recommended. Choose by which engines and signals you want covered, and by which region the tool understands. Prices vary and the space is moving fast.

For the schema side, Google's Rich Results Test at search.google.com/test/rich-results checks Organization and LocalBusiness schema for free. Schema.org's own docs are the source of truth for which fields the major engines accept.

For the citation side, the work is human. Local trade journals, Chamber of Commerce listings, regional industry bodies, and named-author guest posts on respected outlets carry weight. AI-written content marketing blogs do not.

A few quirks worth knowing if you are in Australia

• ABN and trading-name mix-ups are common. AI engines often blur two ABNs that share a similar name. Make your legal name and trading name clear on your About page and in your schema.
• Suburb-level intent is strong. Australians search by suburb more than by postcode. Local pages should name the suburbs you serve, not just the city.
• Trade press punches above its weight. The AU directory map is thinner than the US one. A single AFR mention or trade-magazine piece can shift visibility more here than the same piece would in a larger market.
• Regulated industries face a higher bar. In financial advice, legal services, or health, the factuality lens matters most. Wrong info about a regulated service can be a compliance issue, not just a marketing one.

Pulling it together

The honest version is short. Pick 10 prompts. Run them across six engines for 60 measurements. Score what you see. Write down your three biggest gaps. Pick one you can close this month. That is the whole method.

The first scan will tell you more than any marketing report you have read in the last year. It will likely be uncomfortable. That is fine. Discomfort is the start of clarity.

Run the check. Write down your three biggest gaps. Pick one to fix this month. Re-test in 30 days and look for movement, not perfection.