How to measure your brand in AI search

Discovery has moved to AI. Your measurement hasn't. Buyers now research, compare and decide inside AI answers, and most companies have no way to see how their brand shows up there. This is how to fix that.

Has discovery moved to AI?

900 million people use ChatGPT every week. That is more than the combined populations of the United States, the European Union, Australia and New Zealand. Google's AI Mode has passed 1 billion monthly users. And this year, for the first time in internet history, machines generated more web traffic than people.

This is the biggest change in how people find, compare and choose that we have seen in a long time. A buyer asks an AI which option to trust, what to compare, whether you are worth it, and often decides before they ever land on your site. The behaviour has moved. Most measurement has not. If your most important visitor is now an AI answer you never see, the first job is to make that path visible.

Why doesn't the old SEO playbook work in AI search?

Early on, this looked easy to game. Ask a model for the "top 10 best agencies in New Zealand" and it would happily build the list, so people rushed to write that page. The newer models already rewrite the query to route around the trick. The shortcuts are dying fast.

Anyone who has done SEO for 15 years has seen this film. Every hack works until the system gets better, because the system is rewarded for serving the best answer, not the most optimised page. If you are trying to hack the system but you are not the best answer to the question, the model improves and the hack stops working. There is a human at the end of the screen who spends real money. Being genuinely the best answer is the only durable position.

A word on honesty too. In a space this noisy, a citation or accuracy lift is sometimes the model upgrading, not your work. Measure carefully enough to tell the difference between a change you caused and a change the platform shipped.

What should I measure? Build a prompt library around the customer journey

You cannot get a "search console" for ChatGPT. There is no report of how many people prompted for you or how you performed, responses change by person and context, and models change often. So the foundation of measurement is a prompt library you build yourself, structured by where the buyer is in their journey, not by keyword volume.

We hang the whole thing on the customer journey, because that is where clarity comes from, and it works across three layers.

The first is category entry, the moment someone first enters the market, like "things to do with a toddler in Queenstown." Broad, early, high in the funnel. The second is consideration, where they compare options: who is on the list, in what order, and why those brands keep appearing. The third is conversion, the exact questions buyers type before they commit, like "is [brand] worth it?" That tells you what AI says about you when the buyer is closest to spending. Start there and work backwards.

Run those prompts through ChatGPT and Gemini and read the answers for citation, accuracy, sentiment and share of voice. Every person gets a slightly different response, but the responses cluster into recognisable sequences, and the same brands keep recurring, because the models mimic the same strategic sources: reviews, reputation, consistent signals. That clustering is what makes the measurement reliable enough to act on.

What is entity diagnostics, and why does it come first?

An AI model builds a picture of who you are from your whole digital footprint: who you serve, what you promise, how you are different. Entity diagnostics tests that picture. Ask the models what your brand is good for, who they would compare you to, and how you differ. It is a surprisingly honest mirror of your positioning.

If you don't get your entity diagnosis right, all the other optimisation doesn't work, because AI doesn't know who you are. This is the crossover of PR, brand and technical work: consistent language about what you do, schema markup, and a clean organisation and Wikidata presence. Land this first, or the rest is wasted effort.

How do I connect AI visibility to revenue?

Never look at a number without context. "Share of voice: 12%" means nothing on its own. The value comes when you place AI metrics next to the rest of the picture. Filter Search Console and GA4 by your category terms, layer in bot traffic so you can see how often AI systems are reading your content, and set a benchmark.

Now the picture sharpens. Impressions may be softening while your AI citation is climbing, which tells you the journey is shifting instead of failing. You go from a blur of numbers to something you can act on: where you are cited, where you are invisible, and which stage of the journey moves revenue. That is where budget decisions should be made, not in a monthly argument over a single vanity metric.

What kind of content wins now?

Google's own guidance is blunt. It calls generic "7 tips for first home buyers" pieces commodity content: "often based on common knowledge, which could originate from anyone, and typically has little unique insight for readers." The machine already knows the basics. It does not need your version of them.

The value now sits in your brand's specific point of view and in real proof. Prompts have changed too. The average Google search is a few words. In ChatGPT people give the whole context: "I have a one-year-old and a three-year-old, I want this, not that." That rewards depth and specificity, content built for the real scenario instead of the generic category.

Proof matters more than polish. No one cares what the machine summarises. They care what other people think, because we are social creatures. So embed genuine social content, video and first-hand experience into your strongest pages. Plan one shoot that feeds organic social, a practical guide and your AI-facing content at once. One capture, many places.

Where do I start this week?

Open ChatGPT and run the exact prompts your buyers run. "Is [your brand] worth it?" and "best [your category] in [your city]." Read how you are described, or whether you appear at all. That is your baseline, and the reason to build the rest.