Prompt Hierarchies and Prompt Fading: A Working Guide for ABA Teams

A prompt is any added stimulus that increases the likelihood of a correct response. Prompting is trivially easy. Prompt fading is where teaching programs succeed or quietly fail, because a skill that only occurs under prompting is not a skill the learner has.

The two families of prompt

Response prompts

These act on the learner's behavior directly.

• Verbal — telling the learner what to do, or part of it. Ranges from a full instruction to a partial vocal cue such as the first phoneme of a word.
• Model — demonstrating the response. Requires the learner to have generalized imitation, otherwise a model prompt is just an interesting thing to watch.
• Gestural — pointing, tapping, or a directional look. Generally less intrusive than a model.
• Physical — guiding the learner's movement. Divided into partial (light touch, a nudge at the elbow) and full (hand-over-hand through the whole response).

The conventional intrusiveness order runs full physical, partial physical, model, gestural, verbal, independent — though the ranking is not universal. For some learners a verbal prompt is far more intrusive and harder to fade than a gesture, and the hierarchy should be arranged around the individual rather than the textbook.

Stimulus prompts

These act on the materials rather than the learner.

• Positional — placing the correct item closer to the learner.
• Movement — tapping or pointing to the correct stimulus.
• Within-stimulus — exaggerating a feature of the correct stimulus, such as size or colour.

Stimulus prompts are often easier to fade than response prompts because the fade can be made gradual and mechanical — an item moved two centimetres closer each session, or a colour desaturated in steps.

The two directions

Most-to-least

Begin with a prompt that guarantees success, reduce intrusiveness as the learner succeeds.

Use it when the skill is new, errors are costly, or the learner has a history of frustration during instruction. It is the standard errorless approach.

Cost: more prompted trials, more staff time, and the highest prompt-dependence risk if fading is not scheduled.

Least-to-most

Give the instruction, wait, and escalate through the hierarchy only as needed.

Use it when the learner has partial mastery, acquires skills readily, or is in maintenance. It maximizes opportunities for independent responding.

Cost: errors occur, and for some learners those errors are aversive enough to trigger escape behavior.

Graduated guidance

A physical variant where the practitioner shadows the learner's movement, applying only as much guidance as needed moment to moment, and fading proximity over time — hand on hand, then wrist, then forearm, then shadowing without contact.

Well suited to chained motor tasks: handwashing, dressing, toothbrushing.

Time delay

Technically a fading procedure rather than a hierarchy. Present instruction and prompt simultaneously at first, then insert a delay so the learner can respond independently before the prompt arrives. Constant time delay uses a fixed interval; progressive time delay lengthens it gradually.

Efficient, and the easiest to run consistently across staff because the rule is a number rather than a judgment call.

How to fade without losing the skill

Fading fails in two directions. Fade too slowly and you build a prompt-dependent learner. Fade too quickly and you lose the response, which usually means backing up and re-teaching from a lower prompt level.

What works:

1. Define the advancement criterion numerically. Two consecutive sessions at or above 90% at the current prompt level, for example. Written into the program, applied by every staff member the same way.
2. Define the back-up criterion too. Two consecutive sessions below 60% means returning to the previous prompt level. Without this, a struggling learner accumulates failed trials while everyone waits to see if it improves.
3. Record prompt level on every trial. This is non-negotiable and it is the most commonly skipped step. Percent correct without prompt level is uninterpretable — 90% correct at full physical prompting is not progress, it is compliance with a prompt.
4. Differentiate reinforcement by prompt level. Independent responses earn more — higher quality, higher magnitude, or a preferred item reserved for unprompted responding. Prompted responses earn acknowledgement. If both earn the same, independence has no advantage.
5. Watch latency. Rising latency under time delay is the earliest indicator of prompt waiting, and it shows up before accuracy declines.

What the data has to show

A prompt-fading program produces a graph with two dimensions: accuracy and prompt level. A single accuracy line is not enough, and this is the most common data design flaw in skill acquisition programs.

The clean way to present it is accuracy plotted with prompt level shown as a condition change, so a reader can see accuracy holding while prompt intrusiveness decreases. That pattern — flat accuracy, descending prompt level — is what successful fading actually looks like, and it is invisible on an accuracy-only graph.

Practically this means the trial-level record has to carry prompt level as a field rather than as a note. Systems that only capture correct and incorrect force the BCBA to reconstruct prompt level from memory at review time, which is how programs end up advancing on data that does not support it.

Common failures

• The invisible prompt. A staff member has drifted into an unrecorded verbal cue or a subtle gaze shift toward the right answer. The learner is at 100% independent on paper and cannot do the task with anyone else. This is what fidelity checks are for.
• Fading on the wrong dimension. Reducing physical guidance while increasing verbal support is not fading, it is substitution.
• Prompt hierarchies that skip levels inconsistently. One RBT going model to independent while another goes model to gesture to independent produces data that cannot be pooled.
• Never fading at all. Programs that sit at partial physical for eight months because nobody set an advancement criterion.

The summary

Response prompts act on the learner, stimulus prompts act on the materials, and both need a planned fade. Most-to-least for new or costly skills, least-to-most for partial mastery, graduated guidance for motor chains, time delay when you want consistency across staff. Write numeric criteria for advancing and backing up, record prompt level on every trial, reinforce independence more richly than assistance, and treat rising latency as the early warning it is.