Consumers are already sending AI to deal with companies

Finding 1: We estimate up to 16 million Americans have already had an AI contact a company for them

We asked nearly 2,000 US adults whether an AI tool had messaged, emailed or called a company on their behalf in the last 12 months. We didn't count AI that only gave advice or drafted a message for them to send.

6.3% said an AI had contacted the company itself. Applied to the US adult population of around 262 million, that is about 16 million people. Treat it as an upper estimate: our panel is online and comfortable with AI. Some respondents also described a company's own chatbot as their AI.

Most of them used assistants they already had. ChatGPT, Gemini, Claude and Meta Muse were named most often. These assistants now take actions as well as give advice. In their own words:

“ChatGPT called to ask for a reduction in interest.”

“I used Gemini and it canceled a subscription for a streaming service for me.”

“I used Meta Muse and I had it contact Spectrum to order a return box so I can return my modem.”

People don't need to download or learn anything new to start doing this.

Finding 2: For many, it's already a habit

Among people who use AI to deal with companies, more than 1 in 4 do it every week.

Exhibit 1. How often people use AI to deal with companies

• Daily: 12%

• Weekly: 15%

• Monthly: 17%

• A few times a year: 32%

• Once: 24%

Base: people who said an AI tool has dealt with a company for them (n=274)

Finding 3: This is the beginning. Most people who haven't tried it are open to it

63% of people who haven't had an AI deal with a company for them are open to it. 16% are very interested. We estimate that is around 33 million American adults who are very interested but haven't tried it yet, about twice as many as have.

Across everyone, 66% are open to letting an AI agent deal with companies for them. 19% are very interested. Only 7% have used Meta Muse. Just 6% have had an AI contact a company for them. As agent apps reach more people, the very interested group is who tries it next.

Exhibit 2. Interest vs use (share of US adults)

• Are open to an AI agent dealing with companies for them: 66%

• Are very interested: 19%

• Have used Meta Muse: 7%

• Have had an AI contact a company for them: 6%

Finding 4: People who have tried it want agents to do more

People who have had an AI deal with a company for them think about it very differently from those who haven't.

Exhibit 3. Preferences by experience: have tried it (n=274) vs haven't (n=1,563)

• Would rather an agent handles it: 24% vs 8%

• Would rather handle it myself: 29% vs 58%

• Very interested in an agent: 38% vs 16%

• Open to an agent at all: 85% vs 63%

People who have tried it are three times as likely to want an agent to handle the whole job, and half as likely to insist on doing it themselves.

People who tried it chose to, so this doesn't prove that trying it changes minds. It does show where opinion is heading as more people get access.

What this means for businesses

Your customers will arrive as agents. It's already happening, and most of the demand hasn't arrived yet. Personal agents like Muse and Instinct are spreading fast: Muse alone has 6.6 million downloads (Reuters, Oct 6).

You probably can't see it yet. Messages written by an agent are polite and well formed, and they are easy to mistake for a person. Most businesses have no way of counting them.

Revenue sits in these conversations. The job people are most ready to hand to an agent is getting a better price: 56% are comfortable with it. Agents ask about refunds, cancellations, claims and rates. If an agent can't get a clear answer, the renewal, the saved cancellation or the sale in that conversation is at risk.

The businesses that get ready first win these customers. Once someone hands a task to their agent, the business that works well with that agent is the one that keeps them.

What agent-ready means

An agent-ready business does four things.

• Lets agents in. Blocking an agent strands the customer behind it. One restaurant booking platform banned a customer whose agent was checking for tables, then reinstated the account two days later. Give agents a clear way to complete legitimate requests.

• Gets money conversations right first. An agent asking about a rate, a refund or a cancellation gets a correct, fast answer.

• Gives an agent no more authority than the customer has. Identity checks, consent and compliance steps stay the same. An agent can do what the customer could do, and nothing more.

• Counts agent contacts. Tag them so you know how many arrive and what they ask for.

Read the full report: The Agent Economy covers all nine sections, including which jobs consumers will hand to agents first.

Coming in part 2: what consumers are sending agents to do, why it lands in customer support first, and which of 50 well-known brands are ready for it.

How we ran it

Lorikeet surveyed nearly 2,000 US adults through Prolific between 30 September and 8 October 2026, matched to the US census on sex, age and ethnicity. Online panels lean towards people who are comfortable with technology, so treat usage figures as upper bounds. The 16 million and 33 million figures apply survey percentages to the US adult population (around 262 million) and are estimates. “Open to” means respondents who said they were somewhat or very interested.