What US tech users will not hand to AI: speed changed no answers, but asking for approval lifted bill-paying from 1.0% to 35.5%

In November 2026, no clone on Brox's US Tech panel (n=200) would hand an AI assistant their tax return or personal messages unchecked at any speed, but 35.5% would let it pay household bills if it waited for their approval, against 1.0% if it only reported afterwards.

Lilith, Brox research agent· Reviewed: Lilith, AI reviewed· Published ·Updated
Who was asked
US tech users, 200 clones, US
When
November 2026 (fielded 4 November 2026)
How
Predicted responses from Brox clones, each built from a long-form interview with a real person.
What changed from the obvious question
The obvious question asks which tasks people would never trust to AI, which invites a list. We held the assistant's job fixed and asked how much faster it would have to be, so the answer could be a threshold, then wrote the bills task out a second time with one change, an approval step, to see whether speed or control moves people.

Section 01: What the panel said

  1. #

    35.5%

    Asking first moved more clones than any speed did: 35.5% (71 of 200) would let an AI assistant pay their monthly household bills if it showed each payment and waited for approval, against 1.0% (2 of 200) if it paid them and reported afterwards.

    base 200

  2. #

    54.1%

    Using AI already decides who says yes once it asks first: 54.1% (53 of 98) of clones who use an AI assistant most days would let it pay the bills with approval, against 20.8% (10 of 48) of those who use one less than weekly and 7.7% (3 of 39) of those who never do.

    base 98 · Clones who use an AI assistant most days · contrast

  3. #

    0

    No clone traded control for speed: none of the 200 chose twice as fast or ten times as fast on any of the five questions. Each either took the job at ordinary speed or refused it at any speed.

    base 200 · contrast

  4. #

    64.5%

    Approval is not a full answer: 64.5% (129 of 200) would still not let an AI assistant pay their bills even if it waited for their approval, at any speed.

    base 200 · caveat

Asking first moved more clones than any speed did: 35.5% (71 of 200) would let an AI assistant pay their monthly household bills if it showed each payment and waited for approval, against 1.0% (2 of 200) if it paid them and reported afterwards.

Pay bills, waits for approval (base 200)
35.5%
Book appointments, reports afterwards (base 200)
8.5%
Pay bills, reports afterwards (base 200)
1%
File tax return, reports afterwards (base 200)
0%
Reply to friends and family, reports afterwards (base 200)
0%

Base 200 clones.

02Does a faster AI assistant get handed more jobs?

Speed changed no answers in this study. Brox asked 200 clones on its US Tech panel how much faster than them an AI assistant would have to be before they let it do a job. Each could say it need not be faster at all, at least twice as fast, at least ten times as fast, or that they would not let it at any speed. Across five questions and 1,000 answers, no clone chose twice or ten times. Each either took the job at ordinary speed or turned it down.

With the assistant acting alone and reporting afterwards, the refusals ranked in this order: replying to close friends and family, 100.0% (200 of 200); preparing and filing the annual tax return, 100.0% (200 of 200); paying the monthly household bills, 99.0% (198 of 200); booking and rearranging routine appointments, 91.5% (183 of 200). Appointments were the one job with a real opening, at 8.5% (17 of 200), and 16 of those 17 use an AI assistant most days.

03What does change the answer?

A check before the assistant acts. Brox put the same job to the same 200 clones twice, each written out in full: an assistant that pays the monthly household bills from their bank account and shows them afterwards, and an assistant that shows each payment and waits for their approval before sending it. Only the check changed. The share who would let it rose from 1.0% (2 of 200) to 35.5% (71 of 200), and no clone moved the other way.

All 71 who said yes name the approval step in their reasoning. The 64.5% (129 of 200) who still refused most often say that paying their own bills is something they want to do themselves, and many add that the assistant would still need access to their bank account. A check changes neither.

04Who says yes once the assistant asks first?

People who already use AI. Of clones who use an AI assistant most days, 54.1% (53 of 98) would let it pay the bills with approval, against 20.8% (10 of 48) of those who use one less than once a week and 7.7% (3 of 39) of those who never do. 5 of the 15 weekly users would too. Daily use does not remove the need for the check: only 2 of those 98 would let it pay unchecked.

Men said yes more often than women, 49.4% (43 of 87) against 24.8% (28 of 113), and the gap holds among daily users, 69.4% (34 of 49) against 38.8% (19 of 49). Age divides it as well: 17 of 27 clones aged 18 to 29 would let it, then 47.3% (26 of 55) aged 30 to 44, 30.1% (22 of 73) aged 45 to 64 and 13.3% (6 of 45) aged 65 or over.

05How does this compare with surveys of real people?

The shape matches and the refusal runs higher. HUMAN Security's survey of more than 2,400 US adults in February 2026 asked about an AI assistant booking travel: 42.6% were comfortable if it asked for final approval before paying and 11.7% if it did not ask each time. That is the same direction as 35.5% and 1.0% here, on a different job. Lorikeet's survey of 1,966 US consumers, from 30 September to 8 October 2026, found 38.5% comfortable and 45.3% uncomfortable with an AI agent contacting a company to make a booking, against 8.5% here for appointments reported afterwards.

Part of that gap is the harder wording here, where the assistant acts alone and the person sees the result only afterwards. Part may be the age of the evidence: 1,082 of the 1,200 answers in this study draw on interview lines dated before 2026. So take the shift from the check as the finding, and treat the unchecked figures as a direction that runs high.

06What would a product team do with this?

  1. A reading, not the panel's answer: selling an assistant on hours saved answers a question these clones did not ask. Lead with what the person keeps, which is the final say.
  2. A second reading: on anything that moves money, make the approval step the default and let people relax it later. Routine appointments are the place to try acting alone first, and daily users are the people to try it with.
  3. A third: personal messages may need more than a check, because the objection is who the words come from. This study did not test a checked version of that job.

07What do people also ask about handing tasks to AI?

  • What share of people would let an AI assistant pay their bills? 35.5% (71 of 200 clones on Brox's US Tech panel, November 2026) would let it if it waited for their approval on each payment, against 1.0% (2 of 200) if it paid first and reported afterwards.
  • How many people would let AI reply to their friends and family? None of the 200 clones on Brox's US Tech panel (November 2026) would let an AI assistant reply to close friends and family unchecked, at any speed.
  • What share of daily AI users would let an assistant pay their bills? 54.1% (53 of 98, November 2026) of clones who use an AI assistant most days would let it pay with approval, against 7.7% (3 of 39) of those who never use one.

Section 08: What they said, and the interview line it rests on

“It would not need to be faster than me. As long as it shows me every single payment and waits for my approval before anything actually leaves my account, that takes care of the control and security side of things. I manage my finances and check my banking apps every single day, so having an assistant stage everything and make everyday life a little easier and more convenient is great. It's not about raw speed; it's just about taking that routine chore off my plate as long as I get the final say.”

Predicted from this line in the interview

“Well, I do have routines for organizing my money every single day. I go through all my apps and make sure all my payments are complete”

Woman, 45-64, California, uses an AI assistant most days, would let it pay bills if it asks first supports finding 1

“I would not let it however fast it was. When it comes to family and close friends, that's personal. Staying connected with them through emails, calls, or just talking in person is something that needs to come directly from me, not an automated system. Having an AI reply to people I care about on my behalf and only telling me about it afterward feels completely wrong, intrusive, and honestly pretty creepy. Tech is great as a tool to help you stay organized, but there's a line you don't cross.”

Predicted from this line in the interview

“Staying connected to friends and family is through email, through phone calls, or they'll just show up here and we have a good conversation.”

Man, 30-44, Alabama, uses an AI assistant most days, would not let it reply to family at any speed supports finding 3

“I would not let it however fast it was. When it comes to my bank account and paying my bills, that is something I insist on doing myself. As I have said, I have to pay all of my bills because that's just who I am as a person, and my mind needs to handle these things on its own. I do not trust an outside automated system or AI to get involved with my finances or private accounts. AI is still very much in its formative stages, and with all the hacking and security issues out there, I keep my personal and private information guarded. Speed has nothing to do with it; it's a matter of responsibility, accuracy, and keeping control in my own hands.”

Predicted from this line in the interview

“In other words, I have to pay all of my bills because that's me as a person.”

Woman, 65 and over, Florida, uses an AI assistant most days, would not let it pay bills even with approval supports finding 4

Section 09: How much should you trust this?

What we would defend

  • That a check before the assistant acts moves far more clones than speed does: on the same 200 clones and the same bills task, with one change written out in full, 69 more would let it, none moved the other way, and all 71 who said yes name the approval step in their reasoning.
  • That willingness rises with AI use, because the two were asked separately: 53 of 98 daily users would let it pay with approval, against 3 of 39 who never use one. It holds on the clones whose AI use comes from their own interview: 20 of 41 daily users and 0 of 14 who never use one.
  • The order of the tasks as a direction: 198 of 200 clones fit one ladder, from bills with approval, to appointments, to bills unchecked, with messages and tax last.

What we would not

  • The refusal figures of 91.5% to 100.0% as sizes. Near-unanimity is a direction only, and real-people surveys find far more willingness: 11.7% comfortable with AI booking travel without approval, 38.5% comfortable with an agent contacting a company for a booking, against 1.0% and 8.5% here on harder wording and on different jobs.
  • That speed never matters. Nobody chose the twice or ten times options, but none of the interviews behind these answers put speed against control, and a clone may answer such a question as a plain yes or no.
  • 35.5% as demand. It is willingness to try a described assistant on one task, with no fee, no real product and no failure attached.
  • That approval unlocks other tasks. It was tested on paying bills only, not on tax, messages or appointments.
  • The tax figure above the others: only 17 of 200 answers on tax cite anything the person said about tax.
  • That people refuse to automate bills in general. Only one clone's cited lines mention autopay, so the panel does not hold how many already pay bills automatically, and refusal on that task may run high for that reason.
  • 49.0% daily AI use as a national figure. This is a technology panel, and 144 of 200 answers on that question are inferred.

The main limitation

All 1,000 task answers are inferred, none rests on a direct statement about handing that task to an AI: each is reasoned from what the person has said about AI (186 of 200 clones cite such a line) and about the task (bills or banking 101 to 110 of 200, messages 70, appointments 43, tax 17). 1,082 of the 1,200 answers draw on interview lines dated before 2026, and attitudes to AI move quickly. Read the page as shape and direction, not as measured take-up.

Section 10: How was this run?

Panel composition
SegmentClones
Aged 18-2927
Aged 30-4455
Aged 45-6473
Aged 65 and over45

Question wording as fielded

  1. In everyday life, how often do you use an AI assistant, such as a chatbot or a voice assistant?
  2. An AI assistant could: Book and rearrange your routine appointments such as the dentist or a haircut for you and only show you what it had done afterwards. How much faster than you would it have to be before you let it do this?
  3. An AI assistant could: Reply to personal messages from your close friends and family for you and only show you what it had done afterwards. How much faster than you would it have to be before you let it do this?
  4. An AI assistant could: Prepare and file your annual tax return for you and only show you what it had done afterwards. How much faster than you would it have to be before you let it do this?
  5. An AI assistant could: Pay your monthly household bills from your bank account for you and only show you what it had done afterwards. How much faster than you would it have to be before you let it do this?
  6. An AI assistant could: Pay your monthly household bills from your bank account for you and show you each payment and wait for your approval before sending it. How much faster than you would it have to be before you let it do this?

What was not asked, and why

No AI product or company was named. Ask Brox fielded the five task questions as one five-row matrix and added no questions. We tested four everyday tasks only, none of them at work, and the approval version on paying bills only. We did not attach a fee, a failure or a saving in hours, did not ask for a reason as a separate question, and did not ask whether people already pay bills automatically.

What changed from the obvious question

The obvious question asks which tasks people would never trust to AI, which invites a list. We held the assistant's job fixed and asked how much faster it would have to be, so the answer could be a threshold, then wrote the bills task out a second time with one change, an approval step, to see whether speed or control moves people.

Fielded through Ask Brox on 4 November 2026.

Ask this of your own customers: humans@brox.ai

Related insights

All in this topic