How do Australian businesses track AI search citations?

A Perth tradesperson gets recommended by ChatGPT to someone searching for a plumber. The searcher never clicks a link, calls the number straight from memory of the name, and nothing in the business’s analytics ever registers what just happened. This is now routine, and most Australian businesses have no system built to see it.

I run Outercite, where we track how AI search engines cite Australian businesses across ChatGPT, Claude, Gemini, Perplexity, Grok and DeepSeek. The question we get asked most by businesses new to this is not "how do I get cited," it is "how would I even know." This is the answer.

What does it actually mean to track an AI search citation?

It means running real, repeated prompts against the AI models people actually use, ChatGPT, Claude, Gemini, Perplexity, Grok and DeepSeek, and recording whether your business is named, cited as a source, or absent, then doing it again on a schedule. Because the models pull from live web content and are non-deterministic, a single check tells you almost nothing; the same prompt run twice in one day can return a different answer.

What is the difference between a mention and a citation?

A mention is your name appearing somewhere in an AI answer. A citation is the model treating your content, or a page about you, as the actual source it built the answer from, often with a visible link or footnote. Counting mentions is easy and overcounts badly, because a keyword match in a saved transcript is not the same as being the source an answer relied on; we cover the mechanics of how models choose sources in how AI decides who to cite.

Why does this matter for Australian businesses specifically, right now?

Because the audience doing the asking is now large and growing fast. ChatGPT reached 13.8 million Australian users by mid-2026, ahead of Gemini at 9.1 million, Meta AI at 5.6 million and Copilot at 5.4 million, according to Telsyte research (B&T, 2026). On the business side, the National AI Centre’s SME AI Pulse survey put Australian small business AI adoption at 43 to 44% across the December 2025 to February 2026 quarter (National AI Centre, 2026). Both sides of the market are moving before the measurement tools have caught up.

Doesn’t ranking well on Google already cover this?

No, and the gap is bigger than most businesses expect. Ahrefs analysed 15,000 long-tail queries across Google, Bing and four AI assistants and found only 12% of AI-cited URLs also ranked in Google’s top 10 for the same query, and more than 80% of AI citations came from pages that did not rank in Google’s top 100 at all (Ahrefs, 2026). A business sitting at position one on Google can be entirely invisible in ChatGPT, Claude or Perplexity for the exact same question, because the two systems are largely evaluating different evidence.

Why can’t Australian businesses just check Google Analytics?

Because GA4 was built to attribute clicks by referrer, and most AI platforms either do not send a click at all or strip the referrer header when they do, so the traffic that does arrive is misclassified. It is not a settings problem you can fix by turning something on; the data most AI tools pass to your site genuinely does not carry the information GA4 needs.

Why does AI traffic show up as "Direct" in GA4?

Because a browser or app without a standard referrer header looks identical to someone typing your URL from memory, and GA4 buckets both as Direct. An analysis of 446,405 website visits found 70.6% of AI-driven traffic arrived with no referrer header at all, landing in that Direct bucket indistinguishable from bookmark traffic (Loamly, 2026). Mobile apps strip referrers more aggressively than desktop browsers, so the undercount is worse on the device most people actually use to ask a question on the go.

What should Australian businesses actually track?

Five things, and a dashboard that only shows one of them is showing you a fraction of the picture. Citation rate on its own tells you whether you show up; it does not tell you whether you are losing to a named competitor, whether the source engines are relying on is even accurate, or whether the answer people see is favourable.

• Citation rate, per engine. What share of the prompts that matter to your category actually name you, tracked separately for ChatGPT, Claude, Gemini, Perplexity, Grok and DeepSeek, because coverage differs sharply between them.
• Source of the citation. Whether the model is citing your own site or a third party writing about you. Earned, off-site sources dominate real citations far more than most businesses assume, which we go into in how AI decides who to cite.
• Competitor gap. Who gets named instead of you on the same prompts, and how consistently.
• Sentiment and framing. Being named as "a budget option" and being named as "the market leader" are both a mention; they are not the same outcome for the business.
• Change over time. A single snapshot is close to useless given how much model answers vary run to run; what matters is the trend across weeks, not one lucky or unlucky prompt.

How are Australian businesses actually tracking this today?

In practice, three tiers, and most businesses start at the cheapest one and outgrow it fast. The first is manual: someone on the team periodically types the same handful of prompts into ChatGPT and writes down what comes back. The second bolts an AI-visibility module onto an SEO tool the business already pays for. The third is a purpose-built monitoring platform that runs a structured prompt set against every engine on a schedule and verifies whether a name-drop is a real citation before it counts. We compare the specific tools in that third category, including where Outercite loses, in best AI visibility tool for Australian agencies.

Is manually prompting ChatGPT enough?

For a first look, yes; as an ongoing system, no. Model answers are personalised, time-sensitive and non-deterministic, so one person running one prompt once tells you what happened in that single instance, not what a real customer is likely to see across a week. Getting a reliable number means running the same prompt set repeatedly, across every engine that matters to your customers, and treating each individual check as one noisy data point rather than the answer.

How should a business start tracking this next week?

Start narrow and manual before you buy anything, because the exercise itself tells you whether the problem is even real for your category.

1. Write down the 10 to 15 questions a real customer would ask an AI assistant before choosing a business like yours, in their words, not your marketing copy.
2. Run each one against ChatGPT, Claude, Gemini and Perplexity at minimum; add Grok and DeepSeek if your category skews technical or price-sensitive.
3. Record, for each answer: were you named, were you the cited source or a passing mention, and who else was named.
4. Repeat the same prompt set a week later before drawing any conclusion from a single round.
5. If the pattern holds and you are consistently absent or losing to named competitors, that is the point to weigh a continuous monitoring tool against doing this by hand indefinitely.

Sources

• ChatGPT at 13.8 million Australian users, Gemini 9.1 million, Meta AI 5.6 million, Copilot 5.4 million, Telsyte research: B&T, 2026. Article.
• Australian SME AI adoption at 43 to 44%, December 2025 to February 2026 quarter: National AI Centre SME AI Pulse, 2026. Blog.
• Only 12% of AI-cited URLs rank in Google’s top 10 for the same query, more than 80% do not rank in the top 100 at all, from a study of 15,000 long-tail queries: Ahrefs, 2026. Study.
• 70.6% of AI-driven website visits (446,405 analysed) arrive with no referrer header and are misclassified as Direct traffic in GA4: Loamly, State of AI Traffic 2026. Report.