In Customer Support We Trust (Maybe)

When your kid has a life-threatening food allergy, every meal outside your home becomes a high-stakes test of trust. It’s intense, and the very opposite of those corporate retreat “fall-backwards-and-Greg-from-accounting-will-catch-you” trust exercises. When it comes to eating out, I’ve found there are two paths to (relative) safety. One is to eat at the hugest corporate chains, which have incredibly consistent suppliers and proper allergy sheets, and pick the safe items.

The other is to find a little local place with staff who will listen carefully, understand your needs and what’s required, and spend the time to make it safe. Two forms of trust, both important, though I know which one I’d rather eat at. It’s much the same for trust in customer support, though with fewer chicken nuggets.

Cognitive vs. affective trust

Researchers Johnson and Grayson found that same split in their 2005 work, "Cognitive and Affective Trust in Service Relationships." They label cognitive trust as confidence in a company’s competence and reliability, based on its track record, and contrast that with affective trust, which is trust built from feeling cared for and having perceived shared values. You might think of it as the difference between “head” trust and “heart” trust.

When ATMs were first introduced, they had a trust problem to overcome. People knew their bank teller would give them the correct money, and who to shout at if they did not. Not everyone was willing to extend that same trust to a machine embedded in a wall.

Curt Miller, writing in The New York Times in 1973 about the arrival of ATMs, claims that “the younger and highly mobile white collar population” was open to using the new machines, but that “Older people, however, seem to have a basic mistrust of machines that banks are finding difficult to overcome.” 

We trust ATMs now — mostly in the form of cognitive trust that is built on decades of typically successful transactions and solid service recovery when things went wrong. We don’t expect people to feel affective trust toward them, because an ATM is a constrained machine. You press the buttons, the money comes out (depending on your budgeting skills), and you can count the notes in your hand to verify the results. You don’t need emotional trust in the machine’s values or intentions to use it.

The same cannot be said of large language model (LLM)-based artificial intelligence tools in customer service transactions. They are (for the most part) non-deterministic, which means they can produce different results even given the same set of prompts, and it is impossible to figure out specifically why a given answer was generated. They aren’t verifiable in the same way.

That’s not necessarily a problem in practice. In many service situations, LLM-based chat interfaces have made getting help faster and more accessible than ever before, and a machine that regularly provides useful answers creates a trustworthy track record. That is building the cognitive trust necessary to make AI-mediated support work for everyone. 

What’s happening in parallel, though, is the deeply human temptation to anthropomorphize anything that seems like a mind. A computer that can talk feels like an emotional connection, and it can even claim to have its own feelings. It does not, and it cannot, any more than a brass octopus can feel drunken rage.

AI can deliver fast, competent help, given the right prompts and structures around it. It will earn that cognitive trust. But it can’t earn affective trust, because that requires a real person you can have some shared values with. 

You can’t automate affective trust

Bryan S. Gaakman submitted his artwork to the Royal Academy’s summer show, but didn’t make the cut. Bryan isn’t his real name … but neither is Banksy, which is the usual name on his art. The judges didn’t pick Bryan’s work, but they did reach out, unknowingly, to ask Banksy to contribute a piece. He submitted a version of the same work, which was accepted. 

The same art, without the name and the context attached to it, produced a very different feeling in the judges. They were evaluating the art, but also the artist. Johnson and Grayson’s research found that “perceived similarity” was the only trust-building factor unique to affective trust. That is, people felt trusting toward a business when they thought that business had similar values to their own. Other factors contributed to both cognitive and affective trust, but perceived similarity only built up the affective connection.

An AI bot, which can change its entire personality faster than a drama kid moving to a new school, doesn’t have a persistent set of values that customers can align with.

Johnson and Grayson also showed that service providers who act against their own interests inspired more trust. Their example was a financial advisor who recommends a product that earns them less income but is a better fit for the client. An AI bot doesn’t have interests of its own to sacrifice, and so can never build trust in the same way, even if it applies a rule that generates the same recommendation.

AI isn’t “bad” at generating affective trust in customers … it just cannot do it by definition. Given that it can generate other forms of trust though, it is fair to ask whether that limitation matters.

The cost of lost trust

Johnson and Grayson put it this way: “A customer’s cognitive and affective trust in a service provider are positively related to a customer’s anticipation of future interactions.” Or as you and I might say it: You want to go back to the businesses you trust.

In the research, cognitive trust seems to have a stronger impact on people’s intention to return, but affective trust is an important element that only people can provide. Restaurateur Danny Meyer, in his book "Setting the Table," says “Service is the technical delivery of a product. Hospitality is how the delivery of that product makes its recipient feel.” That’s the difference between cognitive and affective trust. 

What builds relationships over the long term, and creates the sort of emotional attachment that lasts beyond the next Black Friday sale deal, is the alignment of people with the values the service provider represents. No person aligns their spending with their values perfectly, but when people have enough choice, they consistently choose the companies that match the way they think, and make them feel good about it.

If we were to abandon affective trust building and rely purely on cognitive trust, we’d be giving up a powerful connection tool right at the moment when widely accessible generative AI tools are making differentiation more difficult. Instead, build affective trust through deliberate choices.

Customer trust-building in practice

When you ask Help Scout’s support team a question, you won’t get a ticket number back. That’s not an accident; it’s a deliberate choice. It is a message to our customers and to our internal team that we’re here to help people, not merely shuffle text around. It’s a signal that then must be backed up with matching behaviors from support staff that build trust over time. If people are treated like a number, instead of an individual, calling it a conversation is actively unhelpful.

Most financial advisors claim to care about your financial health — it’s the financial advisor mentioned earlier, sacrificing their own short-term bonus for a long-term benefit to their client, who is truly investing in trust and building a longer-term relationship.

So building (and rebuilding) trust can be intentional, as in those examples, but it also happens as a by-product of every decision made by the business and its staff. For frontline support staff, one practical approach is to think explicitly about the two types of trust when talking to individual customers. Yes, they probably need specific information, and delivering that will build cognitive trust. But they may also need to feel like someone is on their side, even (and perhaps especially) if they can’t get the outcome they hoped for. That emotional, empathetic, affective trust component pays off in the long term, well beyond anything a satisfaction survey could fully capture. Identifying the particular form of trust needed in the moment will improve your effectiveness.

Could this be automated? To some degree, yes. You can certainly write scripts and documentation in an empathetic and trust-building manner. Your customers will benefit from them, and AI can be trained on them too. AI, though, can be made to sound empathetic but can’t actually be empathetic, and your customers will be increasingly aware of the difference. Think of the difference between a pre-scripted “we’re sorry for any inconvenience” response and a person who hears your frustration and engages with you in it. 

Humans are not off the hook for building cognitive trust either, of course. Even when AI or self-service tools are doing the work, there are implementation decisions to be made, quality to be checked, and service recovery to roll out. Consider carefully the impact of your decisions on trust, and not just on the financial bottom line or even the short-term customer satisfaction. Billy Joel gets it. 

It’s always been a matter of trust

If you’ve gone full Sapolsky, believing that there is no free will and all of life is completely determined, then perhaps trust isn’t relevant to your life. For the rest of us, and even Robert Sapolsky in practice, we need a degree of trust to get through our days.

I need to trust that the local owner of the cafe, who knows our family well, will not mix utensils and accidentally cross-contaminate the food. I can do that because they have shown me they care about it. Our customers, who reach out for support in their time of need, want to be able to trust in our service, to feel cared for and listened to.

AI can absolutely build trust, and will continue to do so, reinforcing cognitive trust through consistent actions and reliable responses, but it can never create the affective trust so vital for long-term relationship building. That’s a job for humans living out their values, making a genuine difference you can feel.