Marketing teams already have a professional skill for finding out what someone actually needs before you respond. Most rollouts never point it at the team itself.
Marketing teams resist AI rollouts for reasons that get flattened into one word: resistance. In practice it’s usually one of six things a person feels they’re protecting: security, competence, autonomy, craft and identity, accountability, or a value about what should stay human. Each needs a different response, and answering the wrong one wastes the conversation. The skill for telling them apart is the same discovery skill marketers already use on customers: ask open questions, listen past the first answer, don’t pitch before you’ve diagnosed. This guide shows how to run that conversation, what it looks like differently for a content marketer, a performance analyst and a community manager, how to stay honest instead of manipulative about it, and how to tell if it actually worked instead of just measuring who logged in.
A marketing leader told me some version of this story recently, and I hear enough versions of it that the pattern has become familiar to me. The company had bought the AI tools. The marketing team had gone through training, the session had gone well, and the feedback scores were strong. A few people left talking about things they wanted to try.
Three weeks later, usage was all over the place. A couple of people had gone all in. Several had tried the tool once or twice. Others had barely touched it, and one of the people using it least was exactly the person the leader expected to pick it up fastest: an experienced writer who was smart, curious and normally comfortable learning new things.
So her manager asked the reasonable question: “Have you had a chance to try that drafting workflow we covered?” The writer said she had been busy. And, to be fair, she was busy. But that was not the whole answer.
This is one of the places I think AI adoption gets misunderstood. When somebody does not use a tool after training, our first instinct is often to treat the problem as logistical or educational. Maybe they need more time. Maybe they forgot. Maybe the training did not stick. Sometimes that is exactly the problem. And sometimes the person knows perfectly well how to use the tool and still has no desire to open it, which is a completely different problem to solve.
That gap between “did the training happen” and “did anyone find out what is actually stopping someone” is where a lot of AI rollouts quietly stall. It is not unique to this one writer. About half of U.S. workers, 52%, say they feel worried about how AI may be used in their workplace, and only 6% believe workplace AI will create more job opportunities for them personally in the long run.[1] That is not a training gap. Training fixes “I don’t know how.” It does nothing for “I know how, and I still don’t want anyone to see me try it yet.”
Marketers already know how to work with an answer like that. We would never run customer research by asking “do you want our product,” hearing no, and writing down “they are resistant.” We would ask what is making them hesitate, what they have tried before, what would need to be true for this to feel worthwhile. We know the first answer is often only the beginning of the answer. Yet when the audience becomes our own team, that discovery muscle sometimes disappears, and “I’ve been too busy” gets written down as “needs more time” instead of investigated.
I am careful with the word resistance. It makes the person sound like the problem: someone is resisting the transformation, someone needs more buy-in. Sometimes people do resist a sensible change. But in AI adoption work, hesitation often tells you something useful about the rollout itself, whether that is a lack of clarity about where AI belongs in the job, a fear that the workflow creates more review work than it saves, or a legitimate risk everyone else is moving too quickly to notice.
Prosci’s change management research backs this up. It names lack of clarity about why a change is happening, changes to a person’s role, fear of the unknown, lack of trust in leadership, and exclusion from the decision as common drivers of resistance, and breaks fear of the unknown into more specific strands: fear of job loss, insecurity about one’s own ability to perform well with new tools, and a sense of losing power, status or identity.[2] AI adds another layer because the technology can touch work people associate closely with their professional value: writing, analysis, design, strategy, customer relationships, decision-making. That can make the reaction far more personal than “I don’t like the new software.”
The question I have found more useful than diagnosing “resistance” is simpler: what does this person feel they are protecting? Usually the answer sits somewhere around six areas, and I want to be upfront that this list is a practical framework we use at Future Factors to guide a conversation, not a validated psychological assessment.
| What they may be protecting | What might be running through their head |
|---|---|
| Security | “If AI can do more of this work, will the company still need me?” |
| Competence | “Everyone else seems further ahead. I don’t want people seeing me struggle.” |
| Autonomy | “Is this another system that tracks how I work and gives me less control?” |
| Craft and identity | “The skill I spent years becoming good at is the thing AI is now doing in seconds.” |
| Accountability | “If AI makes the recommendation and the decision goes badly, am I still the one blamed?” |
| Values and authenticity | “Using AI here feels wrong for the kind of relationship or work I’m responsible for.” |
These are prompts for a better conversation, not labels to pin on a person. Built on Prosci’s resistance research and Future Factors’ own extension of it for AI-specific work.
These are not psychological diagnoses. I would not look at an employee and decide she has craft fear. They are prompts for a better conversation, because the response that helps one person can make another person trust you less. If someone is worried about competence and you announce that AI is going to transform the whole department, you just made the concern bigger. If someone is worried about job security and you give them another prompting workshop, you answered a question they were not asking. That is what empathy looks like in an AI rollout. It is not being nicer about the rollout. It is understanding the problem accurately enough to respond to the problem the person actually has.
And these concerns are not imaginary. More than half of employees, 53% in Mercer’s 2026 workforce research, believe new technology will affect their own job security, which reflects a real, current business environment and not just individual anxiety.[3] A person can be interested in AI and worried about AI at the same time. I see people make the mistake of sorting a team into pro-AI and anti-AI, but human reactions are rarely that tidy. The same writer can be fascinated by how Claude helps research an article and deeply uncomfortable watching it generate a competent first draft. Those are not contradictions. They are boundaries, and your rollout needs to understand them.
Answering the fear someone didn’t name is the same as pitching a feature nobody asked for.
This is my favourite part of this whole conversation, because marketing teams are unusually well equipped for it. Think about what a good marketer does with a customer objection. A customer says a product is too expensive. A weak marketer starts defending the price. A stronger marketer gets curious: compared with what, which part feels hard to justify, what would make the investment make sense. “Too expensive” might mean I don’t see enough value, I don’t trust you yet, or I like the product but need a reason to say no. Completely different problems hiding behind the same sentence.
AI adoption works the same way. When someone says “I haven’t had time to try it,” do not assume that sentence explains the behaviour. One of the biggest lessons from years in marketing is that the first objection is often useful data, but it is rarely the root objection. AI adoption needs the same discovery discipline we already use on customers, and it is not new. It is the same skill, aimed at the wrong audience.
So what does an empathetic AI conversation actually sound like? Not “why aren’t you using Copilot,” which already implies there is a correct answer. Not “you know AI isn’t replacing your job, right,” unless you genuinely know that and are in a position to promise it. I would start much closer to: “I’ve noticed you haven’t used the workflow much yet. I’m not worried about the number. I’m more interested in whether something about it isn’t working for you.” Then stop talking. That pause matters, because managers have a habit of asking a question and then filling the silence when it feels awkward. Do not rescue the person from answering.
If they say they have just been busy, you can stay curious without turning the conversation into an interrogation: “That makes sense. If you did have an hour with it tomorrow, which part would you be comfortable trying, and which part would you still rather do your current way?” That question tells you far more. Maybe they say they don’t mind using it for research but don’t want it writing for them. Do not jump straight to “but it writes great first drafts.” That is the moment to ask what bothers them about that part, and the answer underneath might be that writing is the thing they are good at, and they do not want their role turning into fixing machine-written copy all day. Now you are having the conversation that was sitting underneath “I’ve been busy.”
Empathy without changing anything becomes a nice conversation. The point of the last step is to use what you heard to design a better path into the work.
This is where generic AI adoption advice tends to fall apart. “Marketing” contains jobs built around completely different kinds of value, so the same rollout can feel completely different depending on where someone sits.
A senior content marketer may have spent twelve years becoming an excellent writer. She understands pacing, she can hear when copy sounds wrong, and she knows which customer insight deserves a whole article and which one belongs in a single sentence. When the company introduces generative AI by showing that anyone can produce an 800-word draft in thirty seconds, “look how much faster you can write” may not land as the motivational message leadership intends, because you have just demonstrated speed against the exact skill she built her reputation on. The empathetic response is not a blanket promise that AI will never replace good writers, since nobody can know what every writing role looks like five years out. I would make the work split clearer instead: AI takes more of the research assembly, repurposing and blank-page friction, and she still owns the idea, the argument, the claims and whether anything is good enough to publish. That is more useful than reassurance.
A performance marketer or analyst may have almost the opposite reaction. AI doing more analysis does not necessarily threaten the part of the job they enjoy. Their concern is more likely to be accountability: what happens when the model recommends moving a quarter of a million dollars and they cannot fully explain how it got there. That is not a confidence problem, it is a governance question, so I would answer it with governance. Define the handoff clearly: AI surfaces the recommendation, the analyst tests the assumptions, a person approves a material budget shift. Nobody pretends the system is autonomous when a human still carries accountability for the outcome.
A community manager may happily use AI to summarize comments, identify recurring questions and prepare a weekly sentiment report, then hesitate the moment you suggest using AI to draft replies. It is tempting to call that resistance. But ask another question and you might hear that if somebody is genuinely upset, she believes they deserve a person, which is not an AI literacy problem, it is a professional judgment about authenticity. Slack’s research with desk workers found a similar tension: people were uncomfortable with AI-generated communication feeling inauthentic to send or receive, and privately weighed whether using AI on a task would make a manager think less of their competence or their ability to be replaced.[4] The solution here is usually a boundary, not a mandate: routine information requests can use AI-assisted drafts, complaints and sensitive customer moments stay human. That is not overcoming resistance. That is improving the operating model.
| Role | What may sit underneath the hesitation | A better first response |
|---|---|---|
| Content marketer | Craft and identity | Clarify which parts of the work still depend on editorial judgment and let AI start around the edges |
| Performance / analytics marketer | Accountability | Define where AI recommends, where a person verifies, and who approves consequential decisions |
| Community manager | Values and authenticity | Decide which interactions can be AI-assisted and which moments remain entirely human |
A manager may be protecting autonomy, a junior employee may be protecting competence, someone who survived three restructurings may be protecting security. The right response depends on what you actually hear.
This deserves its own section. Empathy is not “how do I figure out the person’s fear so I can neutralize it and get my adoption number up.” People can feel the difference. If the answer to every concern is still predetermined to be “interesting, anyway, everyone needs to use AI,” the discovery conversation was theatre.
Sometimes the person is right. The workflow might be poor, the tool may produce unreliable output, the review burden might make the process slower than the old way, or the team may have a legitimate privacy question. Prosci’s guidance makes this point well: resistance often surfaces unmet needs, unclear role impacts or real gaps in the change approach, and the useful response is listening for the root cause rather than simply suppressing the resistant behaviour.[2] Empathy should leave open the possibility that the rollout needs to change, not only the employee. That is where leaders earn trust.
This also means being careful about what you promise. If someone asks whether AI is going to reduce headcount, the fastest response, “no, this is here to help you, not replace you,” sounds empathetic and may also be untrue. If leadership has not made a staffing commitment, do not make one on leadership’s behalf. A more credible answer sounds like: “I understand why you’re asking. I don’t have information suggesting a headcount change from this rollout, and I don’t want to promise something I’m not in a position to promise. What I can tell you is which parts of your role we’re changing in this phase, which parts we’re not, and how we’ll communicate if that changes.” Less comforting, maybe. More trustworthy, and employees remember when confident reassurance turns out to have been invented.
Employees experience AI adoption locally, through their own job, their own workload and their own manager, which is why I would put empathy training into manager enablement rather than only into the central change plan. Gallup’s research on this makes the manager effect hard to ignore: employees who strongly agree their manager actively supports AI use are 2.1 times as likely to use AI a few times a week or more, and 8.8 times as likely to say AI gives them more opportunities to do what they do best every day.[5] I would make sure the manager can answer what changes in this person’s role, what does not, where AI stops, what still requires human judgment, and whether the manager can hear a concern without immediately turning the conversation into a sales pitch for AI. That last one is a capability worth training on its own.
There is a real role for AI inside an empathetic rollout, just not the role people usually reach for first. Use AI to analyze anonymous survey comments and cluster recurring concerns, to draft a first-pass FAQ, or to pressure-test an announcement by asking it to read the draft from the perspective of an employee worried about job security and flag which lines create more uncertainty. All of that is useful. What I would not outsource is the relationship itself. If an employee tells you they are scared that the thing they built their career on is becoming worthless, that is not the moment to send them an AI-generated reassurance. The person needs a manager.
| AI can help with | The manager still owns |
|---|---|
| Finding themes across anonymous feedback | Having the individual conversation |
| Drafting communication options | Deciding what can truthfully be promised |
| Generating FAQs | Answering personal role questions |
| Role-playing a difficult conversation in advance | Listening without becoming defensive in the real one |
| Identifying possible concerns from patterns | Checking whether the concern is real for this specific person |
| Suggesting a small experiment | Agreeing on a fair next step with the person |
AI can help you prepare for an empathetic conversation. It cannot care on your behalf.
A good conversation does not necessarily end with “great, I feel much better about AI now.” Someone may still be worried, and if the concern was legitimate, I would expect some concern to remain. What I would measure instead is whether the conversation changed the person’s ability to engage with the work, using the same use, persistence, and impact lens we apply to adoption generally, just pointed at one person’s specific worry.
Use asks whether the person tried the specific piece of work you agreed on, not whether they logged in. Persistence asks whether they are still using that workflow a few weeks later without their manager chasing them, since a burst of usage right after a 1:1 may only mean their manager is watching now. Impact asks whether the underlying friction reduced enough for the person to work differently: maybe the writer still does not love AI-generated prose but happily uses AI for research, maybe the analyst still wants human sign-off on significant budget calls but now trusts the recommendations inside that boundary, maybe the community manager still refuses AI-written complaint responses but saves hours on triage and reporting. That is adoption. The goal is not converting everyone into an AI enthusiast. It is helping people find a safe, valuable and honest way to work differently.
Then ask one more question: what have you learned about where AI should and shouldn’t sit in your work? That answer tells you far more than a license-utilization dashboard.
I would be careful not to oversell this. A team does not fail to adopt AI only because leaders lack empathy. Sometimes the use case is poor, the workflow is clunky, the tool does not integrate with existing systems, training is weak, or the policy is so vague that nobody knows what is safe. Gallup’s comparison of frequent AI users against holdouts found workflow fit, supported experimentation and clear policy each make a large, measurable difference: 88% of employees who strongly agree AI integrates well with their existing systems use it frequently, against 55% of those who don’t; 72% of those who feel supported experimenting with AI are frequent users, against 44% of those who don’t; and 68% of those with clear AI guidelines use it frequently, against 47% of those without.[6]
So I think about adoption as several pieces working together. Training answers can I do this. Workflow design answers where this belongs in my work. Governance answers what I am allowed to do. Manager support answers will my team help me work through this. Reinforcement answers will the behaviour survive after the launch. Empathy answers what is getting in the way for this specific person, and it is what stops the rest of the system from treating everyone as if the same intervention will work.
Whether someone crosses from having sat through training to genuinely changing how they work week to week is close to Future Factors’ own adoption framework, Tool times Workflows times Behavior, covered in full in our explainer on what makes an AI-powered professional. Fear is exactly the kind of thing that keeps the Behavior variable at zero even after Tool and Workflows are both in place. If you are designing the rollout itself and not just running the 1:1s, our guide to change management for AI adoption covers sequencing the announcement before anyone panics. If the gap on your team is people hiding their AI use rather than resisting it outright, our guide to psychological safety for AI adoption covers the leader behaviors that make honesty about AI use feel safe. And if the question is less how does this feel and more who does what now, our breakdown of how marketing teams split the work with AI, role by role covers that directly.
You do not need to survey the entire company. Start with one person, ideally someone whose AI usage surprises you: someone capable who is barely using the tool, someone using it constantly but hiding that use, someone who did well in training but never changed their workflow, or someone whose reaction seems stronger than the use case itself.
Do not begin by diagnosing them. Begin with curiosity: “I’ve noticed this hasn’t become part of your work in the way I expected. What am I missing?” Then listen, ask one question deeper, say back what you think you heard, and let them correct you. Then change something. Maybe it is a smaller first task, a clearer approval point, private practice space, a boundary around customer communication, or an honest conversation about what parts of the role are actually changing. Maybe the workflow itself needs fixing. Listening should change the next move, or you were collecting feedback rather than using it.
Marketing has spent decades getting better at understanding customers. We build personas, run interviews, map journeys, and listen for what people mean beneath what they say. We know that “I’ll think about it” does not necessarily mean someone needs twenty-four more hours. Then AI arrives inside our own department, someone says “I’ve just been busy,” and suddenly we take the sentence at face value. That is the missed opportunity. The skill AI adoption needs, curiosity, good questions, listening without rushing to the pitch, is already sitting inside many marketing teams. It may just need to be pointed in a new direction.
It means understanding what someone feels they are protecting before choosing how to respond, rather than treating hesitation as one generic feeling called resistance. A person avoiding AI may be protecting their sense of security, competence, autonomy, craft and identity, accountability, or a value about what should stay human. Each of those needs a different response, and answering the wrong one wastes the conversation.
Job security is a real and common concern, but it is not the only one. Pew found 52% of U.S. workers feel worried about future workplace AI use, and Mercer’s 2026 research found 53% of U.S. employees expect new technology to affect their job security specifically. Employees can just as easily be worried about looking incompetent in front of peers, being monitored more closely, losing the skill they are known for, or being blamed for a decision an AI model helped make.
Treat that as useful information rather than something to talk them out of. If a workflow creates more work than it saves, introduces a real risk, or crosses a boundary a customer relationship depends on, change the workflow instead of the person’s mind. The same honesty applies to reassurance: only promise what you actually know, since unsupported comfort about job security costs more trust later than an honest “I don’t know yet” costs now.
Start with an observation instead of an accusation, ask what is making the workflow hard to adopt, and let the silence sit after you ask. Listen past the first answer, test your interpretation by saying back what you think you heard, and agree on one small next step that respects the specific thing they are worried about rather than pushing broader adoption.
Do not measure it by whether the person says they feel better. Check whether they tried the specific workflow you agreed on, whether they are still using it three or four weeks later without a reminder, and whether the concern they named has actually stopped getting in the way of the work, rather than just going unmentioned.
This guide combines external workforce research with the patterns we see helping professionals and teams adopt AI through role-based training and workflow design. It draws on Pew Research Center’s February 2025 survey of U.S. workers’ views on workplace AI, Prosci’s published research on why employees resist change, Mercer’s 2026 Inside Employees’ Minds research on job security concerns, Slack’s Workforce Lab interviews with desk workers about how AI use feels day to day, and two of Gallup’s current workplace AI studies on manager support and on what separates frequent AI users from holdouts. A widely repeated claim that a large share of employees expect new technology to make their job more frustrating in the near term could not be verified on Mercer’s own published pages and is excluded here. Future Factors’ “what are they protecting” framework is a practitioner tool for guiding conversations, not a validated psychological assessment, and the role-based scenes in this article are illustrative examples rather than documented case studies from a named company.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.