The Short Answer
People ask an AI version of an expert the questions they would feel awkward asking the expert twice. Across 4,670 questions in 1,810 live conversations, 70% were substantive questions about the subject itself, and only 3% asked to be put through to a human.
The single most common pattern was members asking for the expert's own specific positions and numbers, on things they had already been told and could no longer remember. The second was using the clone as a search engine over years of back catalogue. Neither is a beginner behaviour, and neither is what a clone gets demoed for.
What People Ask, by Share of Questions
Grouped by what the person was trying to do, not by topic.
| What the person wanted | Share of questions |
|---|---|
| A substantive answer about the subject itself | 70% |
| To find a specific piece of the expert’s content | 15% |
| Pricing, membership or joining | 8% |
| To reach a human | 3% |
| Everything else, including new-member orientation | The remainder |
The finding that matters
Only 3% asked to speak to a human.
That is the objection raised in every conversation about putting an AI version of yourself in front of paying members: they will just ask for you anyway. In this dataset it is the smallest category apart from orientation. People used the thing in front of them.
How far conversations go
Of conversations included at least one follow-up
Ran to five questions or more
Questions in the longest single conversation
The average conversation ran to 2.6 questions. A single-question interaction is someone checking whether the thing works. A follow-up is someone using it, and most conversations had one.
How We Classified It
Stated plainly, because the method is the reason the numbers are worth anything.
We looked at 4,670 questions people asked AI clones across 1,810 conversations, all from live sites rather than tests, and grouped them by what the person was trying to do.
Classification was by keyword over the full set of messages. Two caveats worth stating, because a number without them is not worth quoting.
The 70% is a residual. It is everything that was not clearly retrieval, pricing, contact or orientation. We describe it as substantive questions about the subject rather than claiming a finer classification than keyword matching can support.
Seeded prompts are excluded. New clones ship with suggested starter questions. None of those are counted here, and none of the quotes below are generated. Every one is a real question typed by a member in a live member area.
"I Can't Remember, and I Don't Want to Ask Again"
The strongest finding, and the one we did not have a category for.
Members ask the clone for the expert's own specific positions. Not general principles, but the actual numbers and frameworks they were taught, on things they were told once and have since forgotten.
- “What are the coach's percentages that he looks at on a P&L? I can't remember.”
- “What does the coach teach about a, b & y clients?”
- “What percentage of lender work do you do in your own business, and what percentage is non-lender private work?”
- “What specific online platform do you use to hire VAs in the Philippines?”
- “If I am going to go through all the trouble to develop a niche, market myself, and do my own sales, why not just become a real estate agent and get into the sales side of it??”
Two things stand out. The first is that "I can't remember" is the entire value proposition in three words. The expert already answered the question. The member simply cannot find where.
The second is that the last one is not a question, it is a challenge. People argue with a clone in a way they might not argue with the person, which means objections surface where you can see them instead of quietly costing you a renewal.
Searching Years of Back Catalogue
The clone as a search engine over a body of work.
The requests in this category are strikingly vague, and the clone still resolves them. Members cannot remember the title, the month, or even the format.
- “I don't know the year and month. I would like you to find it. The title says something about 'Easy SOPs...'”
- “Can you help me find the virtual conference that has something to do with getting on the VA panel”
- “I am looking for a recording that talks about OpenClaw memory”
- “There was an AI training in July. Is that recording available?”
- “Anything that talks about SOP in the title”
- “Can you give me a direct link to that archived recording”
This is the argument for loading everything you have ever produced, including the material you assume is dead. A recording from three years ago that nobody can name is still being asked for, and a keyword search box on your member portal would not have found it either.
"That was perfect! Now can you find a Virtual Conference recording that talks about Social Media."
Front-Door Questions
The questions a new member would otherwise send you directly.
A smaller share, but the one that most obviously saves the expert time, because these are the questions that arrive one at a time and always need the same answer.
- “I'm new, where should I start?”
- “What are the different memberships you offer?”
- “How can you help me build or structure my business?”
See your AI clone in 30 seconds
Paste your website. Watch it learn what you do, who you help, and how you sound.
Free, no signup needed to preview.
What This Means for Your Clone
Two practical consequences, both of which change what you feed it.
Load the whole back catalogue, not the highlights
Fifteen percent of all questions were someone trying to find a specific thing you already made. Every recording you leave out is a question your clone cannot answer, and members are searching for material they cannot even name.
Load your positions and numbers, not just your philosophy
The most common behaviour is members asking what you specifically said about a specific thing. A clone trained on your general approach will produce something reasonable and useless. It needs the actual percentages, the actual thresholds, the actual frameworks.
Expect to be argued with, and treat that as the feature
Some questions are challenges rather than requests. Those objections were previously invisible, resolved privately by the member deciding not to renew. Seeing them is worth more than the answer itself.
This is what members want from a clone. For why the experts themselves build one, which turns out to be a different answer entirely, see why coaches are building AI versions of themselves. If you want the accuracy question instead, we published how accurate an AI clone really is.
Frequently Asked Questions
What do people actually ask an AI clone?+
Across 4,670 questions asked of AI clones in 1,810 live conversations, 70% were substantive questions about the subject the expert teaches. 15% were people trying to locate a specific piece of the expert's existing content, 8% were about pricing or membership, and 3% asked to be put through to a human. The remainder were orientation questions from new members.
Do people ask to speak to a human instead?+
Rarely. Only 3% of questions were a request to reach a person. That is the objection most often raised about putting an AI version of yourself in front of paying members, and in this dataset it is the smallest category apart from orientation.
Are the questions basic, or genuinely useful?+
Mostly the latter. The largest single behaviour is members asking for the expert's own specific positions and numbers on things they were told before and have forgotten. The second is using the clone to find a recording or document in a back catalogue when they cannot remember the title, the month or the format.
How many questions does one person ask?+
The average conversation ran to 2.6 questions. 55% of conversations included at least one follow-up rather than stopping after a single question, 14% ran to five or more, and the longest single conversation reached 21 questions.
What should I put in an AI clone's knowledge base?+
Everything you have produced, and your specific positions rather than only your philosophy. The two most common uses in this dataset both fail without it: members asking what you said about a particular number or framework, and members searching for a recording they cannot name. A clone trained only on your general approach cannot answer either.
How was this measured?+
Keyword classification over 4,670 questions across 1,810 conversations, taken from live member areas rather than tests or demos, grouped by what the person was trying to do. The largest category is a residual: everything that was not clearly retrieval, pricing, contact or orientation.
Put your back catalogue to work
Point Personify at everything you have already made and see what it can answer. Free tier to start, then from $29/month billed annually.
Free tier available. No credit card needed.