Key Takeaways
- Accept that you are estimating, not measuring. Nobody can count the clients who never heard of you. What you can do is bound the number with four inputs and act on the range.
- Start from your own baseline, not an industry figure. How often assistants name you is the one input you can check yourself in an afternoon, and it moves the answer more than anything else.
- Report a range and never a single figure. A number with three decimal places is a tell. Two of the four inputs are guesses, so the honest output is a spread.
- Compare the range against the cost of fixing it, not against zero. Some gaps are real and small. The decision is whether the range clears a weekend of your time, not whether it is greater than nothing.
- Re-run the same arithmetic in ninety days. The absolute number matters less than whether the gap narrows once you have changed something. Direction is the signal.
This question usually arrives in a slightly angrier form: how am I supposed to know what I never got? It is a fair objection, and the honest answer is that you cannot know. You can only bound it.
Bounding it takes four numbers, two of which you already have. The other two are estimates, which is why the output is a range and not a figure you would put in a business plan.
What follows is the arithmetic, a worked example with illustrative numbers, and the three ways this calculation misleads people who like the answer it gives them.
Is being invisible to AI costing you clients?
Some, yes, and the size depends almost entirely on how many of your buyers now start by asking an assistant. If that share is small in your niche, so is the loss. If it is large, the loss is large and invisible at the same time.
The reason it stays invisible is structural. Someone asks an assistant for three coaches, gets three names that are not yours, and hires one of them. Nothing about that sequence touches your website, your inbox, or your analytics. There is no bounce to look at, because there was never a visit.
So the useful move is not to hunt for evidence of the loss. It is to estimate its size from things you can observe, decide whether that size justifies any work, and then re-run the same estimate later to see whether it moved.
The four numbers you need
Two come from your own records and two are judgment calls. Keep them separate in your head, because the two you can look up are worth arguing about and the two you cannot are worth widening.
In text, because nothing may live only inside a figure. Input A is the enquiries you get in a year, illustratively 60. Input B is the share of buyers who now start by asking an assistant, illustratively 1 in 4. Input C is how often assistants fail to name you, illustratively 21 of 25 answers. Input D is the share of enquiries you turn into clients, illustratively 1 in 3. The illustrative output is 3 to 6 clients a year.
Input C is the interesting one, because it is the only estimate in the set that you can replace with a measurement. Asking twenty-five buyer questions and writing down who gets named is an afternoon of work, and the fifteen-minute version of that test is enough to get a first reading.
A worked example, with illustrative numbers
Here is the whole calculation in sentences, using the figures above. Every number in this section is illustrative and none of it is client data.
You get 60 enquiries a year. Suppose 1 in 4 buyers in your niche now starts by asking an assistant. Your 60 enquiries are then coming mostly from the other three quarters of the market, which implies that quarter should be worth roughly 20 enquiries a year if you were as visible there as you are everywhere else.
You are not. Your baseline says assistants name you in 4 answers out of 25, so you are capturing something like a sixth of that quarter, call it 3 enquiries. The shortfall is around 17 enquiries a year. At your usual close rate of 1 in 3, that is 5 or 6 clients.
Round that down, because the inputs do not deserve precision. The honest output is 3 to 6 clients a year. Multiply by what a client is worth to you over their whole time with you, and you have a number you can actually weigh against a weekend of work.
| Input | Illustrative value | Where it comes from | How much to trust it |
|---|---|---|---|
| A. Enquiries in a year | 60 | Your own records | High, if you count consistently |
| B. Buyers who ask an assistant first | 1 in 4 | Judgment, informed by your own client conversations | Low. This is the soft input |
| C. Answers that never name you | 21 of 25 | A baseline run you do yourself | Good, once you have run it twice |
| D. Enquiries you close | 1 in 3 | Your own records | High |
| Output | 3 to 6 clients a year | The four above, multiplied | Directional only. Never quote it as a fact |
The table is deliberately unflattering about input B. There is real research on how widely assistants are used, including Pew's 2026 survey of American AI use, but none of it tells you what share of people hiring a strength coach in your city opened an assistant first. That gap between good research and your specific market is the honest limit of this arithmetic.
Where this estimate goes wrong
First, double counting. Some of the people who ask an assistant also search normally, click your site, and enquire. They are already in input A. If you treat the whole assistant-using quarter as lost, you have counted the same person twice and inflated the answer.
Second, the naming rate is not the same as the losing rate. Being named in 4 of 25 answers is not the same as losing 21 clients, because most of those questions were never going to end in a hire. Some are research questions with no buying intent behind them at all, which is why the share of buyers who ask before they visit is worth thinking about separately from the raw count.
Third, motivated arithmetic. This calculation is easy to nudge. Move input B from 1 in 4 to 1 in 2 and the answer doubles, and nothing stops you doing that except the discipline of writing the assumption down before you like the result. Write B down first, then do the multiplication.
There is a fourth failure that is less about arithmetic and more about timing. Assistants give slightly different answers to the same question on different days, so a single baseline run can overstate or understate input C by a few points. Run it twice a week apart before you build anything on it.
What size of gap is worth acting on
Compare the low end of your range against the work, not against zero. The first fixes are roughly a weekend for the base and half a day per page you publish, which is a real cost even though no money changes hands.
If the low end of your range is under one client a year, do nothing about this and go back to whatever produces clients today. That is a genuine outcome of the calculation and it is not a failure. Plenty of businesses run on referral and repeat work where none of this applies.
If the low end is a few clients a year and a client is worth a few thousand dollars to you, the arithmetic clears a weekend comfortably, and the first thing to do is the baseline rather than the fixing. Google's own guidance on its AI features and the Princeton study GEO: Generative Engine Optimization both point at the same unglamorous work: clear, specific, sourced pages that answer the question a buyer actually asked.
Then re-run the arithmetic in ninety days with the same four inputs. If input C has improved and the rest held steady, the range narrows, and that movement is worth more than the original number ever was. Keeping that record costs nothing, and tracking it without buying software is a spreadsheet and an hour a month.
Our free check does the input C part for you, at $0, and emails a plain-English readout within the hour. It runs as soon as you ask for it.
Questions we hear the most
Is being invisible to AI costing me clients?
Almost certainly some, and you can bound the number without being able to measure it. Four inputs get you there: your current enquiry count, how many buyers now ask an assistant first, how often answers name you, and your close rate.
How do I calculate how many clients AI invisibility costs me?
Take your yearly enquiries, work out what share of buyers now start with an assistant, apply how often assistants fail to name you, then apply your normal close rate. The output is a range of clients per year, not a precise figure.
Which number in that calculation is the least reliable?
The share of buyers who start by asking an assistant. It varies by niche, age, and price point, and no public figure is specific enough to your business to trust. Treat it as a range you test rather than a constant.
Can I measure lost clients directly instead of estimating?
No. A person who asked an assistant, got three names that were not yours, and hired one of them leaves no trace in your analytics. That absence is exactly why the arithmetic is worth doing.
How big does the estimated gap have to be before it is worth fixing?
Compare the low end of your range against the hours the first fixes take, usually a weekend plus half a day per page. If the low end is under one client a year, your attention is better spent elsewhere.
Does this arithmetic work for a local business as well as an online one?
Yes, with one change. Local buyers ask location-shaped questions, so your baseline has to use questions that name your city or area, otherwise the naming rate you plug in will describe a market you do not serve.
How do I find out how often ChatGPT names my business?
Write down a fixed list of the questions your buyers actually ask, put each one to the assistant, and record who gets named. The list has to stay identical between runs or the comparison means nothing later.
What does the AnswerHalo free check include?
It costs $0. We ask ChatGPT 25 real buyer questions from your niche, then email a plain-English readout within the hour: how many answers named you, who was named instead, and the first fix. It runs as soon as you ask for it.
Your buyers are already asking. Find out what AI tells them.
The free check asks ChatGPT 25 real buyer questions about your niche and sends you the report within the hour, at $0. If it shows you are already getting named everywhere, we will say so in plain words.
Get my free check →