Ask almost any physician, nurse practitioner, or PA to name the conversation they felt least prepared for in training, and a common answer surfaces: telling someone they, or someone they love, has a serious or terminal diagnosis. It's one of the most emotionally demanding tasks in clinical practice, and historically one of the least practiced. Most clinicians report learning to break bad news largely on the job — through trial, error, and the discomfort of getting it wrong in front of a real patient.
That gap exists for a structural reason, not a lack of will. Breaking bad news is difficult to teach because it's difficult to practice safely. You can't rehearse a cancer diagnosis conversation on a real patient. Standardized patient programs have offered one solution for decades, but they're expensive, hard to schedule repeatedly, and — because the same actor can only deliver so many emotionally intense encounters in a day — limited in how much repetition a single student can realistically get. AI patient simulation is changing that math, giving learners a way to rehearse this specific, high-stakes conversation as many times as it takes to build genuine competence.
Why the SPIKES Protocol Needs Repetition, Not Just Instruction
SPIKES — Setting, Perception, Invitation, Knowledge, Emotions, Strategy/Summary — is the most widely taught framework for breaking bad news, and for good reason: it breaks an overwhelming task into a sequence a learner can actually follow. But knowing the six steps of SPIKES and being able to execute them in a live, emotionally charged conversation are two entirely different skills. A student can recite the protocol perfectly on a written exam and still freeze, over-explain, or rush past a patient's emotional reaction the first time they have to deliver it in real time.
That's the core argument for simulation-based SPIKES training: it's not a knowledge gap, it's a performance gap, and performance gaps close through repetition under realistic conditions — not through additional reading. Research on breaking bad news training consistently finds that learners report low confidence and insufficient practice opportunities as the primary barrier, not insufficient instruction on the framework itself.
Where AI Roleplay Changes the Equation
AI-driven patient simulation addresses the two constraints that have always limited SPIKES practice: cost and repeatability.
Unlimited attempts, without actor fatigue. A human standardized patient delivering repeated bad-news encounters across a full day of student sessions experiences real emotional fatigue — and it shows, in ways that make later sessions less realistic than earlier ones. An AI patient doesn't tire, which means every student gets a consistent, fully-present scenario, whether they're the first learner of the day or the fiftieth.
Practice at the moment of need, not just the scheduled lab slot. Traditional SPIKES training typically happens once, in a single scheduled simulation session per semester. AI roleplay lets a learner rehearse the specific conversation — say, disclosing an abnormal biopsy result — the night before a rotation where they know it's likely to come up, not just months earlier in a general communication skills course.
Variation across scenario types. Breaking bad news isn't one conversation; it's dozens of different ones, each with its own emotional texture — a new cancer diagnosis, a pregnancy loss, a poor prognosis update, a pediatric case delivered to a parent. AI scenario authoring makes it realistic to build and assign a library of these variations, rather than the single generic "bad news" scenario most programs have historically had the resources to produce.
Believable emotional response. The instructional value of SPIKES practice depends heavily on the "patient" reacting the way a real person would — with shock, denial, anger, or a flood of questions — so the learner has to actually apply the Emotions step of the protocol, not just deliver information into a void. Well-built AI avatars are specifically designed to carry that kind of emotional range and respond dynamically to how a learner navigates the moment.
What This Looks Like in Practice
Asher Marks, at Yale, described the value of this kind of practice directly: "Foretell AI is allowing us to rethink soft skills training during post-graduate education. Our trainees have the opportunity to practice delivering difficult news, receive feedback, and iterate on their approach before the crucial moment of disclosure." That last phrase captures the entire argument for simulation-based SPIKES training: the goal isn't a single successful rehearsal, it's iteration — the chance to try, get specific feedback, adjust, and try again, until the approach holds up under real emotional pressure.
Designing a Strong AI-Based SPIKES Scenario
Programs building breaking-bad-news scenarios into an AI roleplay platform should structure them around the protocol itself, so both the conversation and the assessment reinforce the same framework:
- Setting: Does the learner establish privacy, sit down, and check who the patient wants present, before launching into clinical information?
- Perception: Does the learner ask what the patient already understands about their condition, rather than assuming?
- Invitation: Does the learner check how much detail the patient wants, rather than delivering a uniform script regardless of the patient’s preference?
- Knowledge: Does the learner deliver the information in plain language, in small chunks, checking for understanding — or does it come out as a single dense paragraph?
- Emotions: Does the learner pause and respond to the patient’s emotional reaction before moving forward — the step most novice learners skip under pressure?
- Strategy/Summary: Does the encounter end with a clear next step, so the patient isn’t left with information and no path forward?
A rubric built around these six checkpoints gives both the learner and the instructor a precise picture of where the conversation succeeded and where it needs more work — far more actionable than a single overall "how did that go" impression.
An Honest Note on Limitations
Text- and voice-based AI simulation cannot fully replicate every dimension of an in-person bad-news conversation — physical presence, a hand on a shoulder, a box of tissues within reach are part of how these conversations happen in real life, and no software replaces that. What AI roleplay does replace is the scarcity: the fact that most learners, historically, got one or two chances to practice this conversation before doing it for real. Turning that into a dozen chances, with structured feedback after each one, is a meaningfully different starting point for a new clinician's first real disclosure conversation.
The Bottom Line
Breaking bad news well is a skill, not an instinct, and skills are built through repetition — something the traditional standardized patient model was never resourced to provide at scale. AI patient simulation doesn't remove the need for the SPIKES framework or for faculty guidance on delivering it. It removes the artificial ceiling on how many times a learner gets to practice before the conversation matters for real.
Foretell AI gives healthcare programs a scalable way to build SPIKES-aligned breaking bad news scenarios, with realistic emotional response and rubric-based feedback after every attempt. Schedule a consultation to see how it fits your communication skills curriculum.