Rad N Bad: Your Prompting Hierarchy Sucks — Individualized Prompt Selection, Written Fading Rules, and Preventing Prompt Dependency

The Laminated Prompt Hierarchy on the Clinic Wall

Sean opens with a baseball story. Stand next to a hitter and fix his elbow before every pitch. He looks sharp in practice. On game day he freezes and waits for the coach. That is prompt dependency (waiting for help before responding). Our field often treats it as a flaw in the kid. Sean and Mike say we taught it.

The common setup is one hierarchy (a fixed order of prompts) for the whole clinic. Most programs pick most-to-least (start with the most help) or least-to-most (start with the least help). Then it runs on every learner and every skill. Sean points at the laminated sheet with the company logo on it. That is a system decision, he says. It is not an RBT decision.

Mike describes how staff actually learn it. New techs get a template that runs from point A down the alphabet. The plan assumes the BCBA will shape it for each learner. Often nobody does. Saying "we are using least to most" leaves the real questions open. Where do I start? What counts as the least intrusive prompt? How far do I go? With no answers, fidelity (running the plan as written) slips fast.

What we see is we're very quick to go to the prompt that works, not follow the system, which then feeds into that mentality of prompt dependency.

From the talk — Sean Yocum, MA, BCBA, LBA & Michael Carrero, BCBA, LBA

Prompt Type Versus Fading Procedure: Two Separate Decisions

A prompt type is the thing you do. It is the gesture, the model, the position of the item, the words, or hand over hand. A fading procedure is the plan for how those prompts change over time. Most-to-least is not a prompt. It is a plan. Sean says he loses his mind when a program note says only "use MTL." From where? Faded how? Nobody on the floor knows what to do.

Mike walks through a simple example. The room is dark, so he wants the blinds opened. He starts with an indirect comment about the light, then waits. Next he brings the string into view and holds it. Then he models the pull. Full hand over hand sits at the far end, and he says that one makes little sense for this skill. Notice what happened. He named the prompt types and the direction of the fade in the same breath.

Sean turns that into a writing rule. Line one is the prompt type, defined in plain words. Line two is the fading method, with any delay spelled out. Line three is the why for this learner and this skill. He also flags goals that say only "with prompts." By definition that goal cannot be failed, because it covers everything from full physical to independent.

And the first thing you have to do is identify the prompts that you're going to use so you can figure out how you're either going to fade in or out of support.

From the talk — Sean Yocum, MA, BCBA, LBA & Michael Carrero, BCBA, LBA

Staff and System Contingencies That Build Prompt Dependency

Ask what pays off for staff. A fast prompt gets a correct response. The trial finishes. The data sheet looks clean. Waiting a few seconds is slower, and it may produce an error on the graph. So prompts come fast and early. The learner then finds the smartest move in the room: wait for the adult.

Mike names trials per hour (teaching trials counted each hour) as his least favorite benchmark. The contingency is escape or avoidance. Staff want to stay out of trouble, and maybe get praise for a big number. He points out the catch. You can hit 842 trials and have all of them prompted. He also raises processing time. Repeating the instruction three seconds in can restart the learner, like a CD that skips.

Then there is reinforcement. If a prompted response and an independent response pay the same, waiting wins. Sean says that is not a learner deficit. That learner read the contingencies better than the team did. Mike adds a word on errorless teaching (teaching that prevents wrong answers). Avoiding errors matters, especially for learners who get frustrated. But failure is part of learning, and a learner who never tries never shows you what they know.

We prompt because it's efficient, not because it's effective.

From the talk — Sean Yocum, MA, BCBA, LBA & Michael Carrero, BCBA, LBA

What the Prompting Research Says About One-Size Hierarchies

The short version is that no single hierarchy wins for everybody. Libby and colleagues (2008) compared most-to-least and least-to-most for teaching play skills. Most-to-least produced fewer errors. Still, three participants learned faster with least-to-most. The team then added a delay to most-to-least, which gave the learner a chance to respond first. Learning was nearly as fast as least-to-most, with fewer errors. They used those results to propose guidelines for choosing.

Seaver et al. (2014) went further. They assessed different prompt types and fading procedures with 10 individuals. The results were reliable and they were individual. What worked best for one learner was not what worked best for the next one. A 2018 review pulled prompt fading work together with an eye on what is effective and efficient. Under all of it sits older science on delayed prompting and transfer of stimulus control (moving control to the natural cue) going back to 1984.

Sean draws two lines from this. First, a clinic-wide default is a guess. Second, there is no research showing that verbal prompts are the hardest type to fade, so drop that talking point. Mike adds the part people miss. It is not the prompt or the system alone. It is the combination, matched to the learner and to the specific skill.

So a clinic-wide default is basically a guess with a frame around it.

From the talk — Sean Yocum, MA, BCBA, LBA & Michael Carrero, BCBA, LBA

Running a Brief Prompt Comparison Without Running a Study

Mike starts with something free: watch. Hand the learner the item and see what they do, unless the situation is unsafe. People figure things out by acting and seeing what happens. He also brings the tech into the decision early. Both people take notes on the same session, then compare and build the plan together. That is buy-in you cannot train in later.

He leans on graduated guidance (help that adjusts moment to moment) in much of his work, because it fades toward a natural prompt. He also points out that prompts exist in daily life all the time. Independence will look different for each person, because learning histories differ.

Sean keeps the comparison small. A prompt assessment (a quick test of prompt types) does not have to look like a research study. Pick two or three similar targets. Teach one with model prompts and one with gestural prompts. See which reaches independence first. That fits inside a couple of weeks of regular teaching. When you still do not know, start with most-to-least plus a short delay. Then your job is to beat that default with data, not at the next re-auth (the next funding review).

Skill and learner both matter here. Mike uses rollerblading and bike riding. Same idea, different body parts, so the same teaching system will not transfer. Sean has two switch hitters on a team of ten year olds. Teach the swing on the other side with the old cues and the kid steps into the catcher.

There's no wrong way to just watch for a second, right?

From the talk — Sean Yocum, MA, BCBA, LBA & Michael Carrero, BCBA, LBA

Teaching Staff to Wait, and Reinforcing It

Waiting is a skill. It is a deficit for some learners, and for some staff too. Every instinct in human services says to help when a kid hesitates. So wait time will not show up just because it is written in the plan. It has to be trained, and staff have to be reinforced for doing it.

Mike's answer is BST (teach, practice, feedback with staff). Show the person what to do. Check that they can do it. Then support them in doing it the same way over time. He is clear that everyone makes mistakes, and that staff should feel safe saying so and asking for a fix.

Sean gives two ways to measure it. Watch a session and count the seconds between the instruction and the prompt. Under one or two seconds and the learner never had a chance to show you anything. Then try the wait on a single program for one week and track independent responses. If that number climbs, you just learned the learner was waiting on you.

Show them what to do, verify that they know what to do and support them in doing the right thing consistently.

From the talk — Sean Yocum, MA, BCBA, LBA & Michael Carrero, BCBA, LBA

Write the Fade Before You Write the Prompt

Here is the Monday rule from the episode. Write the fade first. The program should say how prompts come out, how many sessions at a level trigger a move, what the next move is, and what to do on a stall. Set that before teaching starts. Skip it and the learner sits at partial physical until someone reads the program again months later.

Sean offers a rule you can steal. Two straight sessions at 80 percent or better at this prompt level, move down one level. Three straight errors at a level, move back up one level and review it. You can write it in trials instead, such as three correct trials in a row at a level. Many data platforms can automate the flag. The exact numbers are a team call. The existence of the rule is not optional.

Mike lists the red flags that show up when the rule is missing. There is no clear process map for staff to follow. There is no real BST behind the plan. There is no procedural integrity or IOA check (two people scoring the same session) to catch gaps. And the plan may not be as individualized as the paperwork suggests, because a prompt in writing looks different in application.

But what I'm getting at is the numbers are a team decision, but that rule exists, right?

From the talk — Sean Yocum, MA, BCBA, LBA & Michael Carrero, BCBA, LBA

Graphing Prompt Level Next to Accuracy

Sean's biggest pet peeve is prompt level data that gets recorded and never used. He compares it to the gauges in his pickup. Nobody glances down until the engine stalls. His fix is a routine data audit. Pull the prompt data for the three oldest programs in the book. Count the sessions at the same prompt level. A month at one level with no plan change is not a learner problem. It is a decision nobody made.

Mike gives software some blame. Many systems graph prompted and independent responses on one line, which hides the story. His workaround is to code each prompt level as its own target: independent, partial physical, tactile, and so on. Staff see in live time what they used. Later you overlay the lines and see where the prompts actually sit.

Then put prompt level on the same chart as accuracy. Parents and RBTs can see support coming out while independence climbs. That matters when the graph looks spiky. A jumpy line may not mean the learner is stuck. The raw data may show prompts fading. Mike adds the other read on heavy variability: either the system is wrong or the implementation is wrong. Both are fixable, and both are on us. The goal of the whole plan stays the same, which is the natural cue with no prompt left to fade.

Don't let the data trick you on a graph with variable data.

From the talk — Sean Yocum, MA, BCBA, LBA & Michael Carrero, BCBA, LBA