Prompting Hierarchy in ABA: A Practitioner's Guide to MTL, LTM, Time Delay, and Prompt Fading

Most-to-least prompting (MTL)

MTL starts each trial, or each new step, with the most intrusive prompt likely to produce a correct response, often full physical or full verbal, then systematically fades to less intrusive levels (model, gestural, visual) as accuracy holds. It is an errorless approach: the learner is prevented from practicing the error before the correct response is well established.

Least-to-most prompting (LTM)

LTM inverts the sequence: the learner gets an opportunity to respond independently first, and prompting escalates only if that opportunity is missed, moving from the least intrusive level up to whatever is needed. Because the learner can attempt the response before any prompt is delivered, LTM tolerates more errors than MTL in exchange for more chances to catch and reinforce independent responding.

Prompt types across the hierarchy

Both MTL and LTM sequences are built from the same set of prompt topographies, ranked roughly by intrusiveness: full physical guidance, partial physical, verbal (full or partial), model, gestural, and visual or textual. These are the building blocks either hierarchy steps through; a verbal prompt requires receptive language for that instruction, and a model prompt requires imitation already in the repertoire, so the ladder is only ever built from topographies the learner can actually use.

Time delay: constant and progressive

Time delay pairs a controlling prompt with the natural instruction, then inserts a delay before the prompt is delivered so the learner has a window to respond independently. Constant time delay uses a fixed delay, commonly zero seconds for several trials, then a jump to a set interval such as four seconds, and stays there. Progressive time delay lengthens the interval in small, planned increments across sessions instead. Unlike MTL or LTM, the delay itself is the fading mechanism: the prompt topography never changes, only the window of time before it's delivered.

Graduated guidance

Graduated guidance is a physical-prompt-specific technique: the therapist provides only as much physical contact as needed at each moment within a single response, shadowing the learner's hand without full contact once movement begins, and fading contact within and across trials rather than stepping down a fixed prompt level. It differs structurally from MTL and LTM in that the fade happens continuously within a single response instead of in discrete steps between trials.

Prompt fading

Fading is the systematic, planned reduction of prompt intrusiveness or frequency over time. In MTL and LTM it means stepping down through the prompt levels described above; in time delay it means lengthening the pause. Every hierarchy on this page relies on it as the mechanism that moves a learner from prompted to independent responding.

The practical choice is errorless (MTL or time delay) versus a more error-tolerant least-to-most sequence, and three factors should drive it: the learner's error history on similar tasks, the skill type, and how much prompt dependence risk you're willing to carry. Before any of that, confirm every prompt topography in the planned sequence is one the learner can actually use, a verbal prompt only works with receptive language for that instruction, a model prompt only works with imitation already in the repertoire. Picking topographies the learner can't use is a more common point of program failure than the MTL-versus-LTM choice itself.

Choose an errorless approach, MTL or time delay, when the learner has a documented history of perseverating on incorrect responses, when an error is likely to be reinforcing in its own right (an attention-maintained wrong answer that gets a reaction, or a stereotyped incorrect response that's automatically reinforcing), or when the skill is brand new and there's no baseline data on partial competence. Errorless procedures trade a slower path to fully independent responding for a cleaner acquisition curve with fewer error patterns to undo later.

Choose least-to-most when the learner already shows partial or inconsistent independent responding, since LTM's structure exists specifically to catch and reinforce those independent attempts rather than prompt through them. It's also the more defensible choice when a rich verbal repertoire or generalized imitation is already present, because independent attempts are more likely to be correct than random.

Within an errorless approach, the choice between constant and progressive time delay is not as settled as it's often taught. Constant delay is simpler to run, easier to train staff on, and the more common default in practice. But at least one direct comparison found progressive delay winning outright, fewer errors and faster mastery teaching expressive labels, against both a 2-second and a 5-second constant delay (O'Neill et al., 2022). Start with constant delay for its simplicity when you have no reason to expect otherwise, but treat that as a default you're willing to abandon, not a rule: if early data show the jump to the full interval producing a burst of errors, or the learner's history says a gentler ramp is needed, switch to progressive delay rather than pushing through on the assumption that constant is the evidence-backed choice. It often isn't, at least not decisively.

Prompt dependence risk should factor into which hierarchy you pick, not just how you run the fading plan once you're in it. That risk runs higher under MTL, since the learner sits at a higher prompt level for more trials before independence is ever tested, so a learner with a known history of prompt dependence on prior programs is often a better fit for LTM or time delay than another round of MTL.

Skill type matters as much as error history. Motor and self-help chains lean toward graduated guidance or a physical-to-gestural MTL sequence, since the prompt needs to track resistance moment to moment (Ozen et al., 2022). Discrimination and academic targets, where the risk is a wrong stimulus-response pairing getting reinforced, lean toward time delay or a tight MTL sequence (Cariveau et al., 2023). Verbal and social targets, where independent attempts carry real diagnostic value, often do better under LTM.

When neither hierarchy fits because the learner has no version of the response anywhere in the repertoire, even a heavily prompted one, that's usually a shaping problem rather than a prompting problem. You're building a new response by successive approximations, not prompting an existing one into view.

The practical question for a working BCBA is rarely whether to use a prompting hierarchy but which format fits the skill and how aggressively to fade it once you're running it.

Stop treating blocked trials as a stimulus-control risk. A direct comparison found blocked trials acquire discriminations just as fast as mixed trials in early learners, without the faulty stimulus control clinicians worry about (McKeown et al., 2025). If a learner is struggling under mixed-trial presentation, moving to a blocked format for initial acquisition, then mixing once responding is fluent, is a defensible first move rather than a fallback.

Choose prompt topography by channel, not habit. Simultaneous prompting cut acquisition time substantially for early listener responses compared to a standard most-to-least sequence (Cariveau et al., 2023), and digital or video-based stimulus prompts can teach conditional discriminations effectively, provided you screen for basic tablet or device readiness first; skipping that screen is where digital prompting programs quietly stall (Niland et al., 2026).

Mind interference when you build a teaching set. Targets that share a first letter or another salient feature should be split across different teaching sets rather than taught together; clustering them in the same set measurably slows acquisition (Chotto et al., 2023). This is a five-minute fix at program-writing time that can prevent weeks of plateaued data.

Reconsider your mastery criteria before assuming a program is failing. Individual-operant mastery criteria with smaller trial doses per session can cut acquisition time meaningfully while improving maintenance, compared to blanket session-level criteria (Kim et al., 2023). If a target is stalling under a fixed criterion, the fix may be the criterion, not the prompt level.

On format selection more broadly: a conditional-only prompting sequence outperformed a modified simple-conditional approach for auditory-visual discriminations (Yuan et al., 2023), and progressive DTT generally outpaced equivalence-based instruction on speed, though both got learners to mastery (Ferguson et al., 2022), so staff fluency and learner preference can reasonably break the tie between the two.

The common error to correct in supervision: prompt fading treated as an afterthought rather than a written, criterion-based plan. Every study behind this grade paired a defined prompt level with an explicit fading rule. A prompting hierarchy without a fading schedule attached is not evidence-based practice; it is just prompting.