Customers increasingly take their service questions to a generative AI tool of their own choosing rather than to the chatbot a company provides. A Gartner survey of 3,566 customers, published on 8 July 2026, found they were about three times more likely to use third-party generative AI than company-provided chatbots for customer service. The company's channel is no longer competing only with the phone line and the branch. It is competing with an assistant the customer already trusts more.
What Gartner found#
The source is a Gartner press release dated 8 July 2026, reporting a survey of 3,566 B2B and B2C customers fielded in February and March 2026. The headline finding is a ratio: respondents were roughly three times more likely to use external generative AI tools than company chatbots to deal with customer service matters. The release is about digital self-service and does not single out the telephone.
| Item | What the release states |
|---|---|
| Publisher | Gartner, press release |
| Date | 8 July 2026 |
| Sample | 3,566 B2B and B2C customers, February–March 2026 |
| Headline finding | About 3× more likely to use third-party GenAI than company-provided chatbots for customer service |
| Channel scope | Not phone-specific |
| Type | Customer survey, not a forecast |
Source: Gartner press release, 8 July 2026. Figures as published; we have not seen the underlying dataset.
The same customer survey also fed two other Gartner releases in 2026: that only 27% of customers would try a chatbot again after a negative experience, and that 87% say companies using generative AI for service must provide access to a human agent. They come from one data source, so they are three views of the same customers rather than three independent studies.
What the ratio does and does not tell us#
A ratio says nothing about absolute levels. "Three times more likely" could describe a small minority compared with an even smaller one, and the press release, as we read it, does not state the underlying shares. Nor does it tell us whether customers resolved their issue with the external tool, or simply researched it before contacting the company anyway.
What it does show is a preference. When customers want help with a service problem, many reach for a general-purpose assistant first. We suspect they trust it, know how it behaves, and it does not belong to the company they are trying to deal with. That last point may matter more than the technology.
Why customers prefer their own assistant#
Gartner's own explanation is a mismatch with customer expectations: many company chatbots are still designed mainly to answer questions, while customers increasingly want AI to act for them. Beyond that we can only reason. A company chatbot is often built to deflect: to answer the question it was designed for, and to keep the customer away from more expensive channels. A customer who has met such a bot learns that it serves the company first. A general-purpose assistant appears to serve the customer. The previous finding from the same survey, that only 27% would retry a chatbot after a bad experience, fits that reading.
In 2019 we said in public that a bot must be a real channel customers want to use. That remains the test. In our view, if customers prefer an external tool, the problem is rarely that the company lacks AI. More often the company's channel is narrower, slower or less honest than the alternative.
Does this reach the telephone?#
Not directly, and we should not pretend otherwise. Gartner measured chatbots and generative AI tools, not phone conversations. But the telephone is affected in two ways.
The customer arrives better prepared#
A customer who has asked an assistant about a late payment, a policy clause or a tariff change may call with a firmer view of what they want, and sometimes with wrong information. In regulated sectors, the conversation then has to correct without arguing and confirm what applies to that customer's contract. That favours processes built on the client's own data and approved wording rather than general knowledge.
The customer's agent may one day make the call#
Forrester predicted in October 2026 that around 10% of consumers may use a personal AI agent as their main service interface in 2027, with rising bot-to-bot traffic. That is a forecast, not a measurement, and we treat it as such. If it materialises even partly, companies will need to decide how they recognise, verify and respond to an agent acting for a customer. Ping Identity's 2026 consumer survey, run by Talker Research among 11,000 consumers in 12 countries, found that 71% are concerned an AI agent acting on their behalf could misrepresent their personality or values, so trust cuts both ways.
For regulated processes, the principle we already apply is the starting point. In collections, debt details are never disclosed before identity is verified; that is a rule of the platform, not a line in a script. Whoever or whatever is on the other end, verification comes before information.
Other sources point the same way, with limits#
Two other publications help place Gartner's ratio, without adding to it. In September 2026 Ofcom, the UK communications regulator, reported that about one in 12 online UK adults already use AI to help manage their telecoms services, and that human contact is still valued. That is a UK finding under UK rules, about one sector, but it shows the behaviour Gartner describes reaching an everyday service relationship.
In June 2026 EY's Future Consumer study framed the next step as a permission economy: whether people delegate actions to AI agents depends on trust and on consent they can revoke. EY's work is not about telephone service, and we cite it only for the principle. If customers come to rely on assistants they control, companies will be judged on how clearly they handle the boundary between helping a customer and acting for one.
None of these sources measures what happens on a phone line in a bank or an insurer. Together they describe a customer who is getting used to AI on their own terms, and who will compare every company channel against that experience. We think that comparison will become harder for weak automated channels to survive, in text and in voice alike.
What it means for regulated enterprise contact#
Banks, insurers, healthcare providers, telecoms and utilities have something an external assistant does not: the customer's actual account, contract, appointment or meter, and the authority to change it. A company channel that uses those advantages, by completing a task rather than answering a general question, gives the customer a reason to use it. A channel that merely restates what is on the website does not.
- Act on the customer's real data. Confirm a payment date, record a promise to pay, rebook an appointment, schedule a technician visit, and write the result back into the client's systems.
- Be transparent. Vee always introduces itself as AI, in line with Article 50 of the AI Act, which has applied since 2 August 2026. We think customers who already use assistants are not fooled by pretence and resent it.
- Offer the route out. Customers who ask for a person, or whose situation needs one, are transferred to the client's own staff along routes agreed before launch.
- Keep regulated replies approved. In regulated processes, what the conversation says comes from a closed, versioned library, so the company's channel stays accurate where a general assistant might improvise.
Outbound contact is a separate case. When a bank calls a customer in early delinquency or an insurer calls before a renewal, the customer is not choosing between channels. The call reaches them. That makes the first seconds decisive: the second rung of our KPI ladder is continuation, the inverse of hang-ups after the company name is heard. A customer who has learned to distrust company bots will hang up quickly unless the conversation is clearly useful from the start.
You don't buy phone calls or AI from us. Those are only means to an end. You get the highest conversion we can achieve, exactly where and exactly when you need it. – Dawid Wójcicki, Founder & CEO of Vee
Questions to ask about your own channel#
- What can a customer complete in our channel that they cannot complete with a general assistant?
- Do we know which questions customers research elsewhere before they contact us, and do our call recordings show it?
- How does our process verify identity before disclosing anything, whoever is on the line?
- Is the AI disclosed at the start of every conversation?
- What share of customers ask for a person, and what happens next?
Gartner's ratio is a warning about trust, not about technology. Customers will use the channel that serves them. In regulated contact the answer is a channel that knows the customer's case, says what it is and gets the task done. That is the standard we hold our own service to.
Follow The First Second
Occasional issues on the mechanics and economics of AI-powered customer contact. No noise, no vendor news.
Follow via RSS