The Human Side of AI
Automate the Transaction. Protect the Relationship.
By: Kara Maddox
AI can make customer communication faster than ever. But research suggests that what customers expect from technology changes with the task.
AI is making it remarkably easy for businesses to communicate at scale.
Appointment confirmations can be automated. Customer questions can be answered instantly. Orders can be tracked. Information can be retrieved. Follow-up messages can be drafted in seconds.
For consumer-facing businesses, that's an extraordinary opportunity.
But there's a mistake we can make before we even begin implementing the technology:
treating every customer interaction as though it asks the same thing of us.
It doesn't.
Consider two messages from the same company:
Your appointment is confirmed for Tuesday at 2:00 p.m.
And:
I've reviewed what happened, and I'd like to talk about how we're going to make this right.
Both can be generated by AI.
Technically, they're both text.
Relationally, they're nothing alike.
The first primarily requires accurate information.
The second carries expectations of attention, judgment, and responsibility.
And the research gives us good reason to take that distinction seriously.
Trust in AI is task-dependent
One of the clearest findings from the broader AI-trust literature is that people don't have one fixed level of "trust in AI."
Their willingness to rely on AI changes depending on what AI is being asked to do.
Alasmari and colleagues examined trust across cognitive domains and found distinct patterns across tasks, including considerably greater skepticism in higher-stakes contexts such as healthcare. Familiarity with AI also predicted trust, reinforcing that neither the task nor the user can be ignored (Alasmari et al., 2025).
That finding sits within a longer tradition of research on task-dependent algorithm aversion. Castelo, Bos, and Lehmann showed that people's willingness to rely on algorithms varies systematically with the nature of the task rather than reflecting a universal rejection of algorithmic decision-making (Castelo, Bos, & Lehmann, 2019).
For businesses, that's an important correction.
The question isn't:
Do customers trust AI?
It's:
Do customers trust AI to do this?
Sensitive communication changes the equation
The distinction becomes even clearer when communication moves into a sensitive domain.
Brandtzaeg, Skjuve, and Følstad asked 440 young people to evaluate mental-health advice produced by ChatGPT and health professionals.
When participants didn't know who produced the responses, ChatGPT's answers actually received higher ratings across validation, relevance, clarity, and utility.
But when authorship was revealed, participants favored the health professionals and rated their responses higher for validation. Qualitatively, human responses were viewed as more credible, empathetic, and tailored (Brandtzaeg, Skjuve, & Følstad, 2026).
The underlying communication hadn't suddenly become worse.
The context surrounding it changed what its source meant.
That should matter to any business designing customer-facing AI.
Automate the transaction
There are many interactions where customers primarily want a company to be:
fast
accurate
available
and
easy to work with.
Appointment confirmations.
Order status.
Scheduling.
Routine reminders.
Basic FAQs.
Straightforward product information.
Simple administrative updates.
Making a customer wait for a human to perform work that technology can handle instantly doesn't necessarily make the experience more human.
Sometimes it simply makes it worse.
Use AI.
Protect the relationship
Now consider another category.
A customer is angry.
Someone has experienced a serious service failure.
A client is making an important decision.
Someone needs advice rather than information.
A company needs to apologize.
The interaction requires judgment.
Those moments contain something the first category doesn't:
an expectation of human involvement.
AI may still belong in the workflow.
It can retrieve information.
Summarize history.
Identify options.
Prepare an employee.
Draft possibilities.
But AI supporting a relationship and AI replacing the human inside that relationship are different design choices.
We need to start treating them that way.
Start with the customer journey
Before automating another customer-facing workflow, ask:
What is the customer actually asking from us here?
Information?
Speed?
A recommendation?
Judgment?
Responsibility?
Understanding?
The answer should influence the role AI plays.
Because somewhere between:
Your appointment is Tuesday.
and:
We understand what happened and we're going to make this right.
the nature of communication changes.
That's the boundary I'm interested in understanding.
Research note
This article is Part 1 of KJMdigital's The Human Side of AI series. Existing research supports the proposition that AI trust and acceptance are context- and task-dependent. The application of that principle to a continuum of transactional versus relational customer communication is an emerging practical interpretation—not yet a validated model.
References
Alasmari, A. A., Alruwaili, R. F., Alotaibi, R. F., Youssef, I. K., & Asklany, S. A. (2025). Demographic influences on trust in artificial intelligence across cognitive domains: A statistical perspective. PLOS ONE, 20(11), e0331003.DOI / article
Brandtzaeg, P. B., Skjuve, M., & Følstad, A. (2026). AI aversion? Effects of author disclosure on young people's perceptions of mental health advice. Cyberpsychology: Journal of Psychosocial Research on Cyberspace, 20(2), Article 1.DOI / article
Castelo, N., Bos, M. W., & Lehmann, D. R. (2019). Task-dependent algorithm aversion. Journal of Marketing Research, 56(5), 809–825.DOI / article
Comments