How much is a ChatGPT mention worth?

Why nobody has a credible number

Nobody has published a credible figure because the path from an AI mention to a sale mostly leaves no trace, and we have no attribution data of our own to offer in its place.

Consider what usually happens. An assistant shortlists three businesses inside a conversation. The buyer reads the names, then searches one of them directly, or types the domain, or picks up the phone. Your analytics records a branded search, a direct visit or nothing at all. Nothing anywhere records the assistant that produced the shortlist. The recommendation did its work and left no referrer behind.

Even the clicks that do arrive are unreliable evidence. Referrer data from assistant surfaces is inconsistent, a share of it resolves as direct traffic, and the visit can land days later from a different device to the one the question was asked on.

Our own position here is a limit rather than a hedge. We have no access to anyone's real ChatGPT or Gemini sessions and no query volume data, which is stated plainly on the technology page. We can tell you how often a fixed set of buyer questions named you, on a dated run, and we can show you the transcripts. We cannot tell you what that mention converted at, so we are not going to price it.

When a proposal quotes you a value per AI mention, the useful question is where the denominator came from. We are aware of no published Australian dataset that supports one, and that includes ours.

What you can see, and what you cannot

You can capture a slice of assistant referrals in analytics, and you should, as long as you read the result as a floor rather than a total.

The setup itself is covered in how to track ChatGPT traffic in Google Analytics, so it is not repeated here. What matters for valuing a mention is what that setup misses, which is most of it.

Two instruments catch part of what analytics cannot see:

• An open field on your enquiry form. How did you hear about us, as free text rather than a dropdown. A dropdown with an AI option in it teaches people the answer you are hoping for; free text does not.
• The same question in your phone intake. Whoever answers the phone asks it and writes the reply down verbatim. For most Australian service businesses the phone is where the money actually arrives, and it is invisible to every analytics property you own.

Self-reported attribution is weak evidence, and it should be labelled as such when you present it. It is also the only instrument that catches the zero-click path, which is where the bulk of this value lives. Weak and present beats rigorous and absent.

The arithmetic that produces your own number

Four inputs you already have, multiplied in order, give you a figure you can defend in a meeting. Pick a fixed period first. A quarter is usually long enough to be stable and short enough that people still remember the enquiries.

1. A. The average gross profit of one closed job, from your accounts. Gross profit, not revenue, or you will overstate the channel by your entire cost of delivery.
2. B. Your close rate on enquiries over the same period, from your CRM or a manual count.
3. C. The enquiries in that period anyone attributed to an AI assistant, from the form field and the phone intake above.
4. D. The number of questions in your pinned scan set that named you, from a dated scan of a fixed question set.

The second line is the one that answers the question in the title, and it is worth noticing what it is: your number, from your figures, for your category and your geography. It does not transfer to anybody else, and no industry average would have told you it.

Treat the result as a floor, then put it next to the quote

Because input C undercounts, the figure the arithmetic hands you is a floor rather than an estimate, and you should say so out loud when you present it to anyone.

That is still enough to make the decision arithmetic instead of belief. Annualise the floor, put it beside what you are being asked to pay, and see whether the gap is wide enough to be worth the uncertainty. Published Australian prices, including ours, are on what AI visibility costs in Australia, and the decision framework sits on is AI visibility worth it.

If the floor comes out smaller than the quote, the honest conclusion may be that this is not worth buying yet. We would rather you reached that conclusion on this page than after paying for it.

The number that comes before this one

Before the value of a mention matters, you need to know whether you get one, and that part is free to establish and needs no attribution at all.

It is not a safe assumption. Across the 137 Australian businesses in our 30 July 2026 scan with complete records from both engines, 41% were never named once by ChatGPT or Gemini across every question asked. A further 26% of those same 137 were recognised when asked about directly and never named when a buyer asked who to use.

If you are in either group, the value of a mention is the wrong question for now, and whether you can earn one is the right one. The shape of that problem is in ChatGPT knows my business but never recommends it. The complete dataset, with every denominator, is in the study.

What would settle this properly

A credible industry figure would need a holdout experiment, and we are aware of none that has been published.

It would look like this: a set of comparable businesses in the same category and geography, some worked on and some deliberately left alone, the same pinned questions asked of both groups on the same dates, and closed revenue tracked over a fixed window with the attribution instruments above running the whole time. Expensive, slow, and the only design that separates the mention from everything else happening to those businesses at once.

We have not run it. If a firm tells you it has, ask for the sample size, the window, the question set and the raw data. Ours is published as a CSV under CC BY 4.0 so that it can be argued with, and anyone quoting a value per mention should be held to the same standard.