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

What a prompting hierarchy actually is in 2026 practice

The BACB Task List, 5th ed., names "stimulus and response prompts and fading (e.g., errorless, most-to-least, least-to-most, prompt delay, stimulus fading)" as a single G-4 competency, signaling that practitioners are expected to fluently choose among formats rather than default to one (BACB Task List, 5e G-4). The applied literature backs that up: across 40 mand-training studies covering 118 participants, prompt selection and fading varied widely (echoic prompts were the modal choice at 51%, physical at 27%, model at 6%, with 23% of learners receiving two or more prompt modalities), and only about half of studies tied prompt delivery to behavioral indicators of motivation, suggesting the field still leans on clinician judgment more than on data-based algorithms (McCammon et al., 2024). The takeaway for practice is not the percentages themselves but the architecture: a defensible hierarchy specifies (1) the prompt topographies in use, (2) the order in which they are delivered, (3) the criterion for fading to the next level, and (4) the mastery rule for prompt removal — all written down before instruction begins Cowan et al. (2023).

MTL vs. LTM: the directional decision

The two directional models work in opposite directions. MTL begins each new target with the most intrusive prompt necessary to evoke a correct response (commonly full physical or full echoic) and systematically decreases intrusiveness across successive trials or sessions, on a fixed criterion such as three consecutive correct prompted responses (Thakore et al., 2024) (McCammon et al., 2024). LTM begins with the least intrusive prompt — usually the natural SD or a brief response interval — and only escalates to a more intrusive level (gestural → model → partial physical → full physical) when the learner fails to respond or errs (Torelli et al., 2026) (McCammon et al., 2024). The clinical rule across the corpus is consistent: MTL is the default for new, error-prone, or safety-relevant skills because it minimizes errors during acquisition, while LTM is the default for established skills, maintenance probes, generalization, and contexts where prompt dependence is the dominant risk (Thakore et al., 2024) (Torelli et al., 2026). When parents are taught all three (MTL, LTM, and progressive time delay) and then allowed to choose which to continue using via a concurrent-chains preference assessment, they consistently prefer LTM and rate it as more acceptable post-training — useful information for parent-implemented programs, where buy-in determines fidelity Halbur et al. (2020).

Time delay is the third workhorse — and PPD often beats CPD

Time delay fades prompts by inserting a latency between the SD and the prompt, giving the learner an opportunity to respond independently before support arrives O’Neill et al. (2022). Constant prompt delay (CPD) holds the latency at a fixed value (e.g., 2 s or 5 s) across trials; progressive prompt delay (PPD) starts at 0 s and increases the delay in small steps (0 → 1 → 2 → 3 → 4 → 5 s) as the learner demonstrates accurate prompted responding O’Neill et al. (2022). An alternating-treatments design across four children with autism and intellectual disability compared 5-s PPD, 2-s CPD, and 5-s CPD on expressive labeling and found PPD reached mastery in fewer trials and less total instructional time, while all three formats led to mastery — making PPD the most efficient single-procedure choice when picking a fading method without committing to MTL or LTM O’Neill et al. (2022). The architecture also translates across populations: a single-case study with a 77-year-old with chronic aphasia used 0 → 5 → 10 s PPD with token reinforcement to teach functional spelling, advancing only after three consecutive independent correct responses Ritchie et al. (2021), and a telehealth study with a 6-year-old autistic child used the same architecture during tact and intraverbal instruction with instructive feedback (Anderson & Wiskow, 2025).

Errorless learning rationale

Errorless learning — typically operationalized as a 0-s prompt delay paired with a high-intensity prompt at the start of training — exists to prevent error patterns from being practiced and to short-circuit prompt dependence by ensuring early reinforcement is contingent on correct (prompted) responding before fading begins (Halbur et al., 2025) (Waits & Gilroy, 2025). A multiple-baseline across-participants tact-training study with three young autistic children layered a 0-s constant prompt delay (errorless) with echoic prompts, then thinned to a 5-s delay; the format produced rapid tact acquisition and the untrained emergence of intraverbal-tacts without direct training (Halbur et al., 2025). A bilingual AAC mand-training study used the same architecture — a 0-s full-physical prompt for errorless responding, faded to a 5-s constant time delay — to transfer stimulus control from the therapist to language-specific queries, with differential reinforcement (30-s item access plus praise for unprompted selections; praise only for prompted) to strengthen unprompted bilingual mands (Waits & Gilroy, 2025). The shared design pattern is worth memorizing: errorless start, criterion-based step to a delay, differential reinforcement that favors independence.

Stimulus prompts vs. response prompts vs. prompt fading

Prompts fall into two families that get faded with overlapping but distinct procedures (BACB Task List, 5e G-4). Response prompts (echoic, gestural, model, physical, textual) are delivered in addition to the SD to evoke a specific response and are typically faded via MTL, LTM, or time delay (BACB Task List, 5e G-4) (McCammon et al., 2024). Stimulus prompts modify the SD itself — making the target more salient (size, color, position) — and are faded by gradually equating the target with comparison stimuli (stimulus fading) or by superimposing and shaping the modified stimulus into the natural one (stimulus shaping) (BACB Task List, 5e G-4). A multiple-baseline study taught receptive identification to six autistic children using only positional prompts — the target stimulus was placed progressively farther from a salient location — producing mastery without gestural or physical response prompts and demonstrating that a hierarchy can be built entirely out of stimulus-prompt fading Leaf et al. (2016). Prompt selection precedes hierarchy selection: choose the prompt family the SD can carry, then choose how to fade it (BACB Task List, 5e G-4).

Within-trial prompt sequences

The classic LTM within-trial sequence runs natural SD → wait response interval (typically 3–5 s) → gestural → wait → model → wait → partial physical → wait → full physical, with the trial terminating at the lowest level that produces a correct response (Torelli et al., 2026) (Radogna et al., 2024). A multiple-baseline study with three Italian adults with autism in a community vocational program used a 5-s gestural to vocal within-trial sequence (gesture → partial model → full model → vocal) embedded in a BST package and produced reliable increases in job-related social initiations, with the explicit logic that starting at gesture preserves naturalness in workplace settings (Radogna et al., 2024). A kindergarten special-education study used a three-step verbal → model → full physical LTM sequence during whole-group life-skills instruction and contrived play-center practice, reaching criterion on group-taught skills and generalizing to play centers with descriptive praise (Torelli et al., 2026).

Specific prompt-fading strategies

Four fading strategies show up repeatedly across the corpus and tend to combine. Time delay (constant or progressive) lengthens the latency between SD and prompt, transferring control from the prompt to the SD O’Neill et al. (2022) Ritchie et al. (2021). Prompt fading by component decreases the intensity of a single prompt topography step-wise — full echoic → partial echoic → first phoneme → no echoic — rather than swapping prompt families Kahlow et al. (2019) (Olaff & Holth, 2025). Model-based fading decreases the salience of a model prompt across trials (live full model → live partial model → video model → no model) within an LTM frame (Mattson et al., 2024) (Torelli et al., 2026). Graduated guidance uses physical contact that varies in real time with learner performance, fading from full hand-over-hand to shadowing as the learner takes over the response (Yarzebski & Dickson, 2024) Preas & Mathews (2021). The combinatorial rule is consistent: pick one within-trial control structure (MTL, LTM, or pure time delay) and one fading lever (component, modality, or contact intensity), and document both in the program Cowan et al. (2023).

Prompt timing inside FCT

Prompt timing also functions as a level in the hierarchy when the program is functional communication training. A multiple-baseline study with four preschoolers with autism or developmental delay tested whether the FCR prompt should be delivered immediately after problem behavior or delayed; the immediate-prompt arrangement kept problem behavior low while maintaining high FCR use for two of four children, while the delayed-prompt arrangement prevented problem-behavior bursts in some cases but slowed acquisition in others Landa et al. (2022). The clinical implication is that within FCT, "where the prompt sits in the behavioral stream" is itself a hierarchy decision — and the choice should be data-driven and individualized rather than fixed by protocol Landa et al. (2022).

Comparative effectiveness — three systems, broadly equivalent for tacts

The closest the corpus comes to a head-to-head test is a randomized clinical trial that assigned 27 autistic children to constant time delay, most-to-least prompting, or flexible prompt fading during discrete-trial tact instruction; pre-post gains, sessions to mastery, generalization, and teaching-session responding were essentially equivalent across the three systems Cihon et al. (2020). A parallel-treatments single-case comparison (n = 3) between MTL, no-no prompting, and a novel "responsive prompt delay" — in which prompts are delivered only after the child errs — found all three reached mastery, with responsive prompt delay achieving comparable or slightly faster acquisition without an artificial prompting ladder Foran‐Conn et al. (2021). For tact-style discrete-trial targets in autism services, no single hierarchy has a durable acquisition advantage — so selection shifts to learner profile, staff training feasibility, and skill type Cihon et al. (2020) Foran‐Conn et al. (2021). Within-session arrangement matters too: an alternating-treatments study comparing prompt-only, target-only, and full rotation while teaching receptive labels found that an MTL hierarchy paired with full stimulus rotation produced the fastest acquisition and lowest error rates Leaf et al. (2018).

Exposure history shapes which prompt is most efficient

A finding the field underweights: prompt efficiency is plastic. An alternating-treatments study with three boys with autism systematically increased exposure to either visual or auditory prompts before intraverbal training, then compared trials-to-criterion across the two prompt types — the prompt with more recent exposure reliably required fewer trials, regardless of which one started off "more efficient" Roncati et al. (2019). The clinical implication is that the order-of-effectiveness ranking that anchors most LTM hierarchies (gestural < model < partial physical < full physical) is not fixed by topography; it shifts with the learner's recent prompt exposure, so before assigning a fading plan, probe or briefly manipulate exposure to ensure your "least intrusive" prompt is also the most efficient starter for that learner Roncati et al. (2019).

A decision tool exists — and it works

For practitioners who want a written algorithm rather than a clinician-by-clinician judgment call, a multiple-baseline-across-participants study (n = 3 adults with developmental disabilities in community day programs) developed and tested a decision-making tool that walks the practitioner through learner profile, skill type, and procedural-integrity considerations to select between LTM, MTL, and time-delay variants Cowan et al. (2023). The tool produced differentiated outcome data across all three participants and embedded systematic data collection so teams could compare hierarchies inside treatment without long pilot phases — a working example of how to convert "clinical judgment" into a procedural variable Cowan et al. (2023).

Caregivers can run a hierarchy with brief training

A multiple-probe single-case study trained three caregivers to deliver MTL with a 2-s prompt delay and differential praise during ADL teaching at home, with hand-over-hand as the entry-level prompt and praise withheld for prompted responses to strengthen independence Preas & Mathews (2021). A separate multiple-baseline study taught three caregivers a graduated-guidance LTM physical hierarchy via a 3–4-minute video model plus brief practice; all three reached mastery within 9–11 sessions and generalized to untrained play activities (Yarzebski & Dickson, 2024). The shared training architecture — short asynchronous content followed by brief in-vivo coaching with a written mastery criterion — is the same one that scales TBFA training Halbur et al. (2020).

Mand training is a special case

Mand training puts unique demands on the hierarchy because the controlling variable is the establishing operation (EO), not a discriminative stimulus, so the prompt must be delivered when the EO is at strength rather than on a fixed schedule. A clinical tutorial recommends a layered hierarchy that begins with EO-based interrupted-chain procedures (contrive an EO by withholding a needed item), adds an immediate intraverbal prompt at the first error, then fades to time-delay prompting (0 → 5 s), with non-vocal model or video prompts interspersed and AO trials inserted to confirm EO control rather than rote stimulus control (Frampton et al., 2024). A companion tutorial argues that observing an indicating response (IR — pointing, hand-leading, reaching) is the most reliable way to confirm that an EO is present before prompting, and provides a flowchart in which the prompt sequence only begins after an IR has been observed (Frampton et al., 2024). Without the IR, the prompted response may function as a tact rather than a mand, and prompt fading will not transfer control to the EO because there is no EO in effect (Frampton et al., 2024).

Stimulus and rule-based prompts are not always sufficient

A counterweight from the corpus: a small (n = 3) computer-task study with university students manipulated the explicitness of verbal rule statements ("Select the car/clock/chair") under immediate vs. delayed reinforcement contingencies and found that participants almost always reverted to the immediately reinforced response, despite the rule (Alonso-Vega et al., 2025). The practical message is that rule statements and verbal prompts cannot consistently override an immediate competing contingency — when problem behavior is on a dense schedule of reinforcement, simply prompting the adaptive response will not produce durable change unless the contingencies are also restructured (Alonso-Vega et al., 2025). The hierarchy is one input among several; reinforcement architecture is the other.

Environmental modifications can act as antecedent prompts

A multiple-baseline study with three at-risk kindergarteners showed that alternative seating (stability stool or scoop rocker) reduced average teacher prompts per session from 1.1–2.1 at baseline to 0.6–0.8 during intervention — a 30–60% reduction in prompt frequency — by altering the physical antecedent rather than escalating verbal or gestural support (Bloom-Williams et al., 2024). A separate withdrawal-design case study with a 20-year-old with autism and intellectual disability in a postsecondary classroom paired a low-intensity picture prompt (OK / Not OK lanyard card) with brief teacher proximity every 5 minutes (cued by a Motivaider) and produced an immediate jump from 38% to 92% on-task, generalizing to a community work site with a job coach (Haydon & Carnahan, 2024). The first hierarchy decision should sometimes be to modify the antecedent stimulus rather than escalate the response prompt (BACB Task List, 5e G-4).

A practitioner page should be honest about its evidence base Cowan et al. (2023). The prompting hierarchy literature lives almost entirely at the single-subject experimental design (SCED) layer, with one randomized clinical trial, a small handful of literature-synthesis pieces, and a band of clinical tutorials around the edges Cihon et al. (2020) (McCammon et al., 2024) (Frampton et al., 2024).

Randomized clinical trials. Cihon and colleagues' RCT (n = 27) is the closest the corpus comes to a head-to-head test of three named prompting systems for tact instruction — constant time delay, MTL, and flexible prompt fading — and reports broadly equivalent acquisition, generalization, and mastery Cihon et al. (2020). This is the page's strongest single evidence point, but it covers tact relations only; generalization to other operants and skill domains has not been tested at the same rigor Cihon et al. (2020).

Systematic and narrative reviews. McCammon and colleagues' review of 40 mand-training studies aggregates prompt type and fading data across 118 participants; it documents what the field uses (51% echoic, 27% physical) but cannot adjudicate which hierarchy works best because included studies rarely justified prompt choice (McCammon et al., 2024). Read it as practice-pattern data, not effectiveness evidence (McCammon et al., 2024).

Single-subject experimental designs. Most of the actionable evidence sits here O’Neill et al. (2022). O'Neill and colleagues' alternating-treatments comparison of 5-s PPD vs. 2-s and 5-s CPD across four autistic children is the cleanest demonstration that PPD is the most efficient time delay variant for expressive labeling O’Neill et al. (2022). Foran-Conn and colleagues' parallel-treatments comparison (n = 3) of MTL, no-no, and responsive prompt delay shows broad equivalence with a slight efficiency edge for responsive prompt delay Foran‐Conn et al. (2021). Halbur and colleagues' multiple-baseline tact-training study (n = 3) and Waits & Gilroy's bilingual AAC multiple-probe study (n = 3) anchor the errorless 0-s + delay architecture (Halbur et al., 2025) (Waits & Gilroy, 2025). Leaf and colleagues' positional-prompt study (n = 6) and rotation comparison (n = 3) show that stimulus prompts and within-session arrangement interact with the directional model Leaf et al. (2016) Leaf et al. (2018). Roncati and colleagues' exposure-history study (n = 3) is the most underused finding in the corpus — prompt efficiency rank is plastic with exposure Roncati et al. (2019). Landa and colleagues' FCT prompt-timing study (n = 4) demonstrates that prompt position is itself a hierarchy variable inside FCT Landa et al. (2022). Halbur and colleagues' parent-preference SCED (n = 3) provides the only direct preference data across MTL, LTM, and PPD Halbur et al. (2020). Cowan and colleagues' decision-tool study (n = 3) is the best existing attempt to convert hierarchy selection into a procedural variable Cowan et al. (2023).

Setting- and population-extension SCEDs. Torelli and colleagues (n = 6 kindergarteners) extend LTM into gen-ed inclusion (Torelli et al., 2026). Preas & Mathews (n = 3) and Yarzebski & Dickson (n = 3) demonstrate parent-implemented MTL and graduated guidance after brief asynchronous training Preas & Mathews (2021) (Yarzebski & Dickson, 2024). Anderson & Wiskow (n = 1) shows telehealth delivery of PPD layered on instructive feedback produces incidental-learning gains (Anderson & Wiskow, 2025). Ritchie and colleagues (n = 1) extend PPD into adult aphasia rehabilitation Ritchie et al. (2021). Radogna and colleagues (n = 3) demonstrate a 5-s LTM hierarchy in adult vocational programs in Italy (Radogna et al., 2024). Haydon & Carnahan (n = 1) and Bloom-Williams and colleagues (n = 3) show that environmental and stimulus prompt modifications can outperform escalating response prompts in postsecondary and kindergarten settings (Haydon & Carnahan, 2024) (Bloom-Williams et al., 2024). Thakore and colleagues (n = 1) demonstrate the MTL plus time delay architecture for a child with chronic stereotypy (Thakore et al., 2024). Frampton & Axe (n = 3) and Kahlow and colleagues (n = 3) extend MTL and echoic-fading hierarchies to adolescents using apps and to mand-information instruction (Frampton & Axe, 2025) Kahlow et al. (2019). Olaff & Holth (n = 9) and Glodowski & Rodriguez (n = 4) demonstrate component fading and scenic picture prompt fading in verbal-behavior programs (Olaff & Holth, 2025) Glodowski & Rodriguez (2019).

Theoretical, tutorial, and analogue work. Frampton and colleagues' clinical tutorials on EO-based mand prompt sequences and on indicating responses are conceptual but high-utility, specifying procedural sequences rather than testing them experimentally (Frampton et al., 2024) (Frampton et al., 2024). Alonso-Vega and colleagues' rule-explicitness analogue (n = 3 adults) is small-N and lab-based but provides the clearest demonstration that prompts and rules cannot consistently override an immediate competing contingency (Alonso-Vega et al., 2025).

BACB anchor. The 5th-edition Task List names errorless, MTL, LTM, prompt delay, and stimulus fading as the family of techniques behavior analysts are expected to fluently choose among under G-4 — the regulatory floor for the page's terminology (BACB Task List, 5e G-4).

Bottom line. The convergent picture is strong for the operational claims this page makes: MTL, LTM, and time delay are the canonical directional models; PPD is generally more efficient than CPD as a stand-alone fading method; errorless 0-s entry plus time delay is the dominant errorless architecture; and hierarchy selection should be matched to skill type, learner profile, and implementer O’Neill et al. (2022) (Halbur et al., 2025) Cowan et al. (2023). It is weaker for any claim that one directional model produces durably better client outcomes than another in head-to-head trials — there, the field has one RCT and a band of SCED with broadly equivalent results Cihon et al. (2020) Foran‐Conn et al. (2021). The corpus is also nearly silent on prompting hierarchy effects in adolescents and adults outside autism services (Radogna et al., 2024) Ritchie et al. (2021).

Hierarchy selection is not "MTL or LTM" so much as "what skill, what learner, what implementer, what is the failure mode I am most worried about." A defensible logic, drawn directly from the corpus:

1. New, error-prone, or safety-relevant skill. Default to MTL with a 0-s prompt delay (errorless entry) and a fixed criterion (e.g., three consecutive correct prompted responses) before fading to the next prompt level. This is the standard architecture in tact-training, AAC mand acquisition, and feeding/play replacement programs (Thakore et al., 2024) (Waits & Gilroy, 2025) (Halbur et al., 2025).
2. Established skill, generalization probe, or maintenance check. Default to LTM starting with the natural SD and a 3–5-s response interval, escalating only on error or non-response. This is the standard across kindergarten gen-ed inclusion, adult vocational programs, and ADL teaching (Torelli et al., 2026) (Radogna et al., 2024) Preas & Mathews (2021).
3. Picking a fading method without committing to a directional model. Use progressive prompt delay (0 → 1 → 2 → … → 5 s) — it reaches mastery in fewer trials and less total instructional time than 2-s or 5-s constant prompt delay across at least one head-to-head SCED O’Neill et al. (2022).
4. Parent- or caregiver-implemented program. Default to LTM with explicit teaching of a 2–5-s response interval and differential praise for unprompted responses; parents reliably prefer LTM after mastery training, which improves fidelity over time Halbur et al. (2020) Preas & Mathews (2021). If physical guidance is part of the program, use a graduated-guidance LTM physical hierarchy taught via a brief video model plus practice; mastery in 9–11 sessions is realistic (Yarzebski & Dickson, 2024).
5. Mand training. Build the hierarchy around the EO, not the SD: confirm an indicating response (IR) is present before prompting; start with an immediate intraverbal/echoic prompt; fade via time delay (0 → 5 s); rotate AO trials to confirm EO control rather than rote stimulus control (Frampton et al., 2024) (Frampton et al., 2024).
6. FCT inside a behavior reduction program. Treat prompt timing — immediate vs. delayed delivery of the FCR prompt — as a level of the hierarchy and individualize it; some learners benefit from immediate prompting after problem behavior, others from delayed prompting to avoid chaining problem behavior to the FCR Landa et al. (2022).
7. Stimulus prompts feasible (e.g., positional, salient color/size). Consider building the entire hierarchy out of stimulus fading rather than response prompts; receptive identification can reach mastery using only positional prompts that are progressively faded Leaf et al. (2016).
8. The "least intrusive" prompt is not actually the most efficient. Probe or briefly manipulate exposure history before fixing the order. Recent exposure can invert which prompt topography functions as the most efficient starter Roncati et al. (2019).
9. Within-session arrangement matters as much as the directional model. Pair an MTL hierarchy with full stimulus rotation (target and prompt) rather than prompt-only or target-only rotation when teaching receptive labels; this consistently produces the fastest acquisition and lowest error rates in head-to-head SCED Leaf et al. (2018).
10. Prompts are not enough on their own. When the target behavior is competing with a denser schedule of immediate reinforcement (e.g., problem behavior), restructure the contingencies first; rules and verbal prompts cannot consistently override an immediate competing contingency (Alonso-Vega et al., 2025).

Clinic and outpatient (1:1)

Outpatient and clinic-based ABA is where most of the prompting hierarchy literature lives, and the operational pattern is consistent: an MTL or errorless PPD architecture for new skill acquisition, written criterion-based fading rules, and full stimulus rotation within session (Halbur et al., 2025) Leaf et al. (2018). The Cihon RCT on tact relations sits here and provides the strongest single comparison — CPD, MTL, and flexible prompt fading produce broadly equivalent tact acquisition, so clinics should select on training feasibility, learner profile, and program consistency rather than on assumed superiority Cihon et al. (2020). Progressive prompt delay is the strongest single fading method when a directional model has not been pre-committed O’Neill et al. (2022). For automatic-reinforcement-adjacent skill replacement (e.g., functional play replacing chronic hand mouthing), MTL with physical-to-gesture fading followed by progressive time delay is well-supported (Thakore et al., 2024).

School (K-12 inclusion and special education)

In gen-ed inclusion, LTM is the default because the SD is the natural classroom contingency, the goal is to fade the BCBA/RBT out, and prompt dependence in front of typically developing peers is itself a stigma risk (Torelli et al., 2026) (Haydon & Carnahan, 2024). The kindergarten LTM verbal-model-full-physical sequence used during whole-group life-skills instruction generalizes to play centers and produces criterion performance with descriptive praise (Torelli et al., 2026). Within-trial 5-s delay-to-model hierarchies fit cooperative play and activity schedules without disrupting the larger instructional flow (Mattson et al., 2024). Environmental modifications often come first: alternative seating reduced average teacher prompts per session by 30–60% in at-risk kindergarteners — a hierarchy step that lives in the antecedent rather than in the response prompt (Bloom-Williams et al., 2024). Postsecondary inclusion benefits from low-intensity stimulus prompts (OK/Not-OK lanyard cards) paired with brief teacher proximity, which preserves dignity and avoids stigmatizing the learner (Haydon & Carnahan, 2024). For learners receiving special education services, the same MTL with criterion architecture used in clinic translates cleanly when life skills are taught in groups with embedded play-center practice (Torelli et al., 2026).

Home and parent-implemented

Parent- and caregiver-implemented hierarchies do best when (a) the directional model is the one parents prefer after experiencing all options (LTM), and (b) the training architecture is short, asynchronous, and has a documented mastery criterion before live implementation Halbur et al. (2020) (Yarzebski & Dickson, 2024). Three caregivers reached mastery on graduated-guidance LTM physical prompting in 9–11 sessions after a 3–4-minute video model plus brief practice, generalizing to untrained play activities (Yarzebski & Dickson, 2024). Three other caregivers learned MTL with a 2-s prompt delay and differential praise during ADL teaching at home, producing reliable increases in independent ADL performance Preas & Mathews (2021). The shared design pattern: short content first, written mastery criterion, in-vivo coaching after Halbur et al. (2020).

Telehealth

Asynchronous training plus brief telehealth coaching is the dominant model for hierarchy training in resource-limited regions, and the same architecture works for direct learner-facing telehealth instruction (Anderson & Wiskow, 2025). A telehealth case used 0 → 5 → 10 s PPD with instructive feedback to teach a 6-year-old autistic child both directly trained and incidental targets without an on-site clinician (Anderson & Wiskow, 2025). The corpus is thin on direct telehealth hierarchy comparisons, but procedural compatibility is clear: time delay and component fading translate to telehealth with no special accommodation, while physical-prompt levels of MTL or graduated guidance require an on-site implementer coached remotely (Yarzebski & Dickson, 2024).

Adult and vocational settings

Adult vocational and postsecondary settings push toward LTM with low-intensity entry levels because the social validity cost of intrusive prompting at a community job site is high (Radogna et al., 2024). A 5-s gestural to vocal LTM hierarchy embedded in a BST package produced reliable increases in job-related social initiations across three Italian adults with autism in community vocational programs (Radogna et al., 2024). A postsecondary case study used a picture prompt plus proximity scheduled every 5 min via a Motivaider, generalizing to a community work site with the same routine implemented by a job coach (Haydon & Carnahan, 2024). Aphasia rehabilitation in older adults uses the same PPD architecture (0 → 5 → 10 s) that works in pediatric programs, advancing only after three consecutive independent correct responses Ritchie et al. (2021).

• Picking a directional model by ideology, not by skill type. Both MTL and LTM are well-supported; the corpus is essentially silent on durable acquisition advantages between them at the group level, so the relevant question is not "which is better" but "which fits the skill, learner, and implementer for this program" Cihon et al. (2020) Foran‐Conn et al. (2021).
• Defaulting to constant prompt delay when progressive prompt delay would be more efficient. A direct head-to-head comparison shows PPD reaching mastery in fewer trials and less instructional time than 2-s or 5-s CPD; if you are choosing a stand-alone fading method, default to PPD O’Neill et al. (2022).
• Treating "least intrusive" as fixed by topography. Prompt efficiency is plastic with exposure; the prompt the learner has practiced most recently is often the most efficient starter, regardless of where it sits on a topography ladder. Probe before fixing the LTM order Roncati et al. (2019).
• Mixing directional models within a session. A program that drifts between MTL on some trials and LTM on others — without a written rule for the switch — produces erratic prompt delivery, slows acquisition, and makes prompt fading nearly impossible to track Cowan et al. (2023). Pick one directional model per program and document the criterion for transition Cowan et al. (2023).
• No written fading plan. Hierarchies fail when the criterion for fading to the next level is unwritten; staff drift, fading gets erratic, and prompt dependency calcifies. Specify the within-trial sequence, the fading lever, and the mastery rule in the program Cowan et al. (2023) (Thakore et al., 2024).
• Fading too fast. Advancing from 0-s to 5-s prompt delay before the learner has demonstrated stable accurate prompted responding produces error patterns; the corpus consistently uses three consecutive correct prompted responses as the entry criterion before increasing the delay Ritchie et al. (2021) (Thakore et al., 2024).
• Letting prompt dependency calcify by reinforcing prompted responses identically to independent responses. Differential reinforcement is the lever that moves the learner toward independence: prompted responses should receive a thinner reinforcer (praise only, lower-quality access) while unprompted responses receive the full reinforcer (Waits & Gilroy, 2025) Preas & Mathews (2021).
• Prompting before confirming the EO in mand training. A prompted response in the absence of an EO is a tact, not a mand; without the EO, prompt fading will not transfer control to the motivating operation, and you will get a faulty mand repertoire that doesn't generalize (Frampton et al., 2024) (Frampton et al., 2024).
• Treating prompts as a substitute for contingency change. When the target behavior is competing with a denser schedule of immediate reinforcement, prompts and rules cannot reliably override the immediate contingency; restructure reinforcement first (Alonso-Vega et al., 2025).
• Skipping environmental modifications. Sometimes the most efficient hierarchy step is structural: alternative seating, picture cards, or environmental rearrangement can reduce the need for response-prompt escalation by 30–60% in classroom settings (Bloom-Williams et al., 2024) (Haydon & Carnahan, 2024).

• The hierarchy is not producing acquisition after a defined trial period (e.g., 2 weeks of high-fidelity implementation with no upward trend). Re-examine the EO/SD assumptions, the prompt topography fit, and the differential reinforcement schedule before assuming the hierarchy is wrong; if the program still fails after one rotation, refer for senior BCBA review Cowan et al. (2023).
• Persistent prompt dependency despite documented fading. When the learner reliably waits for the prompt across multiple sessions and differential reinforcement is in place, escalate to a different fading lever (e.g., switch from MTL-by-step to time-delay-based fading; add stimulus rotation; manipulate exposure to alternative prompt topographies) Roncati et al. (2019) Leaf et al. (2018).
• The hierarchy is teaching a skill that cannot succeed against a competing contingency. If problem behavior is on a denser reinforcement schedule than the target behavior, the hierarchy alone will not produce durable change; the program needs a function-based contingency restructure, not a prompt change (Alonso-Vega et al., 2025).
• Mand training that produces prompted responding without EO control. When AO trials reveal that the response is under prompt or stimulus control rather than EO control, restart with IR assessment and EO contrivance before adding prompts (Frampton et al., 2024) (Frampton et al., 2024).
• Caregiver fidelity is below 80% after two cycles of brief asynchronous + in-vivo training. The training architecture is well-supported and short; if a parent or paraprofessional is not reaching mastery on a graduated-guidance or MTL hierarchy in 9–11 sessions, the issue is rarely the hierarchy itself — refer for direct supervision or shift the directional model to one the implementer can actually deliver (Yarzebski & Dickson, 2024) Preas & Mathews (2021).

The operational claims on this page sit on solid SCED evidence with one supporting RCT for tact relations and a band of clinical tutorials around the edges; the comparative-effectiveness layer is thin and largely confined to autism services in pediatric populations Cihon et al. (2020) Foran‐Conn et al. (2021). A randomized comparison of MTL vs. LTM vs. PPD across non-tact skill domains (chained ADLs, intraverbal repertoires, social initiations) and across non-autism populations would clarify whether the broad equivalence reported in the Cihon tact-relation RCT generalizes Cihon et al. (2020). A prospective replication of the Roncati exposure-history finding at scale would convert "probe before fixing the LTM order" from a useful single-case observation into a routine clinical step Roncati et al. (2019). The Cowan decision-making tool (n = 3) needs a multi-site replication in school and community settings before it can reasonably be embedded in agency-wide programming standards Cowan et al. (2023). The Landa FCT prompt-timing finding — that immediate vs. delayed FCR prompting interacts with individual learner profile — needs follow-up that produces a written decision rule rather than a case-by-case judgment call Landa et al. (2022). Telehealth comparisons of PPD vs. MTL vs. LTM are essentially absent at scale; given the procedural compatibility of time-delay fading with remote delivery, this gap is closeable (Anderson & Wiskow, 2025). The Alonso-Vega finding that rule statements cannot consistently override an immediate competing contingency is a small-N analogue, but a clinical replication in FCT, demand fading, or compliance training would be high-yield (Alonso-Vega et al., 2025). The corpus is largely silent on prompting hierarchy effects in adolescents and adults outside tact and mand instruction, so the field needs more work in vocational, postsecondary, and adult services settings (Radogna et al., 2024) Ritchie et al. (2021).

1. Pick the directional model by skill and learner, not by ideology. Default to MTL for new, error-prone, or safety-relevant skills with a 0-s prompt delay (errorless entry) and a written criterion before fading to the next level. Default to LTM for established skills, generalization, and maintenance, starting with the natural SD and a 3–5-s response interval (Thakore et al., 2024) (Torelli et al., 2026).
2. When choosing a stand-alone fading method, default to progressive prompt delay. PPD reaches mastery in fewer trials and less total instructional time than 2-s or 5-s constant prompt delay across head-to-head SCED O’Neill et al. (2022).
3. Use the errorless 0-s + delay architecture for early acquisition. A 0-s prompt delay paired with a high-intensity prompt at the start of training, faded to a 5-s delay, is the dominant errorless architecture across tact, intraverbal, and AAC mand programs (Halbur et al., 2025) (Waits & Gilroy, 2025).
4. Document the hierarchy in writing before the first trial. Specify the within-trial sequence, the fading lever (component, modality, contact intensity, time delay), and the mastery rule (e.g., three consecutive correct prompted responses before advancing). Hierarchies fail when the criteria are implicit Cowan et al. (2023) (Thakore et al., 2024).
5. Use differential reinforcement to move the learner toward independence. Prompted responses get praise only or a thinner reinforcer; unprompted responses get the full reinforcer (e.g., 30-s item access). Without this lever, prompts calcify (Waits & Gilroy, 2025) Preas & Mathews (2021).
6. For parent-implemented programs, default to LTM and use brief asynchronous + in-vivo training. Parents who experience all three options reliably prefer LTM after mastery, and a 3–4-minute video model plus brief practice gets caregivers to graduated-guidance mastery in 9–11 sessions Halbur et al. (2020) (Yarzebski & Dickson, 2024).
7. In school inclusion settings, default to LTM and prioritize stimulus-prompt and environmental modifications first. Alternative seating cuts teacher prompts per session by 30–60%; picture-prompt + proximity preserves dignity in postsecondary inclusion (Bloom-Williams et al., 2024) (Haydon & Carnahan, 2024).
8. In mand training, build the hierarchy around the EO, not the SD. Confirm an indicating response before prompting, contrive EOs via interrupted-chain procedures, and rotate AO trials to confirm EO control rather than rote stimulus control (Frampton et al., 2024) (Frampton et al., 2024).
9. Treat FCT prompt timing as a hierarchy variable. Whether the FCR prompt is delivered immediately after problem behavior or delayed is itself a level of the hierarchy and should be individualized with data, not fixed by protocol Landa et al. (2022).
10. Probe exposure history before fixing your "least intrusive" prompt order. Recent exposure can invert which prompt topography is most efficient; a single exposure manipulation can change the LTM order for a given learner Roncati et al. (2019).
11. Pair MTL with full stimulus rotation for receptive-label acquisition. Full rotation (target and prompt rotated together) outperforms prompt-only or target-only rotation across head-to-head SCED Leaf et al. (2018).
12. When the SD can carry it, build the entire hierarchy out of stimulus prompts. Receptive identification reaches mastery using only positional prompt fading, eliminating gestural and physical response prompts Leaf et al. (2016).
13. Restructure contingencies before escalating prompts. When the target behavior is competing with a denser schedule of immediate reinforcement, no hierarchy will reliably produce durable change; the program needs a contingency change (Alonso-Vega et al., 2025).

Primary research synthesized in this guide. DOIs link to the original source.

• McCammon, M. N., Wolfe, K., & Check, A. R. (2024). A review of the environmental variables included in mand training interventions. The Analysis of Verbal Behavior, 40, 345–378. https://doi.org/10.1007/s40616-024-00211-9 https://doi.org/10.1007/s40616-024-00211-9
• Haydon, T., & Carnahan, C. (2024). Implementing a picture prompt and proximity intervention in a classroom with an adult learner: A case study. Behavior Analysis in Practice, 17, 638–642. https://doi.org/10.1007/s40617-024-00924-2 https://doi.org/10.1007/s40617-024-00924-2
• O’Neill, S. J., McDowell, C., & Leslie, J. C. (2022). A Comparison of Variations of Prompt Delay During Instruction on an Expressive Labeling Task. Journal of Behavioral Education, 31(2), 404-422. https://doi.org/10.1007/s10864-020-09407-0 https://doi.org/10.1007/s10864-020-09407-0
• Halbur, M., Kodak, T., & Reidy, J. (2025). A comparison of training procedures on the emergence of intraverbal-tacts. The Analysis of Verbal Behavior, 40, 379–402. https://doi.org/10.1007/s40616-024-00214-6 https://doi.org/10.1007/s40616-024-00214-6
• Noell, G. H., Gansle, K. A., Sliman, H. M., & Lucia, D. M. D. (2026). Establishing Hierarchical Classification Responding: Directly Trained and Emergent Responses. Behavioral Interventions, 41(2). https://doi.org/10.1002/bin.70090 https://doi.org/10.1002/bin.70090
• Bloom-Williams, C. E., Crosland, K. A., & Fuller, A. A. (2024). Evaluating the use of alternative seating with kindergarteners at risk for emotional and behavioral disorders. Behavior Analysis in Practice, 17, 500–513. https://doi.org/10.1007/s40617-023-00852-7 https://doi.org/10.1007/s40617-023-00852-7
• Frampton, S. E., Davis, C. R., Meleshkevich, O., & Axe, J. B. (2024). A clinical tutorial on methods to capture and contrive establishing operations to teach mands. Behavior Analysis in Practice, 17, 1270–1282. https://doi.org/10.1007/s40617-024-00985-3 https://doi.org/10.1007/s40617-024-00985-3
• Halbur, M. E., Kodak, T., Wood, R., & Corrigan, E. (2020). An evaluation of parent preference for prompting procedures. Journal of Applied Behavior Analysis, 53(2), 707-726. https://doi.org/10.1002/jaba.616 https://doi.org/10.1002/jaba.616
• Cowan, L. S., Lerman, D. C., Berdeaux, K. L., Prell, A. H., & Chen, N. (2023). A Decision-Making Tool for Evaluating and Selecting Prompting Strategies. Behavior Analysis in Practice, 16(2), 459-474. https://doi.org/10.1007/s40617-022-00722-8 https://doi.org/10.1007/s40617-022-00722-8
• Cihon, J. H., Ferguson, J. L., Leaf, J. B., Milne, C. M., Leaf, R., & McEachin, J. (2020). A randomized clinical trial of three prompting systems to teach tact relations. Journal of Applied Behavior Analysis, 53(2), 727-743. https://doi.org/10.1002/jaba.617 https://doi.org/10.1002/jaba.617
• Torelli, J. N., Snyder, S. K., Griffin, M. L., Hellemn, M. P., Cagliani, R. R., & Morgan, G. A. (2026). Life skills evaluation in a kindergarten special education classroom. Journal of Applied Behavior Analysis, 59(1), e70041. https://doi.org/10.1002/jaba.70041 https://doi.org/10.1002/jaba.70041
• Alonso-Vega, J., Fellingerová, V., Pereira, G.-L., Estal-Muñoz, V., & Akerø Hylland, A. (2025). Preliminary analysis of rule explicitness on instructional control in immediate and delayed contingencies. The Analysis of Verbal Behavior, 41, 101–116. https://doi.org/10.1007/s40616-025-00216-y https://doi.org/10.1007/s40616-025-00216-y
• Waits, R. A., & Gilroy, S. P. (2025). Development and evaluation of a simultaneous bilingual communication intervention protocol for autistic children. Journal of Applied Behavior Analysis, 58(4), 717–730. https://doi.org/10.1002/jaba.70038 https://doi.org/10.1002/jaba.70038
• Landa, R. K., Hanley, G. P., Gover, H. C., Rajaraman, A., & Ruppel, K. W. (2022). Understanding the effects of prompting immediately after problem behavior occurs during functional communication training. Journal of Applied Behavior Analysis, 55(1), 121-137. https://doi.org/10.1002/jaba.889 https://doi.org/10.1002/jaba.889
• Ritchie, H., Reuter-Yuill, L., Perez, A., & Baker, J. (2021). Assessment-Informed Intervention for Aphasia in an Older Adult: Transfer of Stimulus Control Procedure Considerations. The Analysis of Verbal Behavior, 37(2), 226-236. https://doi.org/10.1007/s40616-021-00153-6 https://doi.org/10.1007/s40616-021-00153-6
• Barry, D., Neufeld, J., & Stewart, I. (2024). Teaching nonarbitrary temporal relational responding in adolescents with autism. The Analysis of Verbal Behavior, 40, 135–152. https://doi.org/10.1007/s40616-024-00210-w https://doi.org/10.1007/s40616-024-00210-w
• Preas, E. J. & Mathews, T. L. (2021). Evaluation of Caregiver Training Procedures to Teach Activities of Daily Living Skills. Behavior Analysis in Practice, 14(4), 958-973. https://doi.org/10.1007/s40617-020-00513-z https://doi.org/10.1007/s40617-020-00513-z
• Leaf, J. B., Cihon, J. H., Ferguson, J. L., McEachin, J., Leaf, R., & Taubman, M. (2018). Evaluating Three Methods of Stimulus Rotation when Teaching Receptive Labels. Behavior Analysis in Practice, 11(4), 334-349. https://doi.org/10.1007/s40617-018-0249-5 https://doi.org/10.1007/s40617-018-0249-5
• Foran‐Conn, D., Hoerger, M., Kelly, E., R. Cross, R., Jones, S., Walley, H., & Firth, L. (2021). A comparison of most to least prompting, no‐no prompting and responsive prompt delay procedures. Behavioral Interventions, 36(4), 1024-1041. https://doi.org/10.1002/bin.1808 https://doi.org/10.1002/bin.1808
• Glodowski, K. R. & Rodriguez, N. M. (2019). The Effects of Scenic Picture Prompts on Variability During the Acquisition of Intraverbal Categorization for Children With Autism. The Analysis of Verbal Behavior, 35(2), 134-148. https://doi.org/10.1007/s40616-019-00120-2 https://doi.org/10.1007/s40616-019-00120-2
• Yarzebski, V., & Dickson, C. (2024). Teaching caregivers to use graduated guidance using video modeling. Behavior Analysis in Practice, 17, 1198–1203. https://doi.org/10.1007/s40617-024-00969-3 https://doi.org/10.1007/s40617-024-00969-3
• Roncati, A. L., Souza, A. C., & Miguel, C. F. (2019). Exposure to a specific prompt topography predicts its relative efficiency when teaching intraverbal behavior to children with autism spectrum disorder. Journal of Applied Behavior Analysis, 52(3), 739-745. https://doi.org/10.1002/jaba.568 https://doi.org/10.1002/jaba.568
• Kahlow, T. A., Sidener, T. M., Kisamore, A. N., & Reeve, K. F. (2019). Teaching the Mand “When?” to Children With Autism Spectrum Disorder. The Analysis of Verbal Behavior, 35(2), 221-234. https://doi.org/10.1007/s40616-019-00115-z https://doi.org/10.1007/s40616-019-00115-z
• Leaf, J. B., Cihon, J. H., Townley-Cochran, D., Miller, K., Leaf, R., McEachin, J., & Taubman, M. (2016). An Evaluation of Positional Prompts for Teaching Receptive Identification to Individuals Diagnosed with Autism Spectrum Disorder. Behavior Analysis in Practice, 9(4), 349-363. https://doi.org/10.1007/s40617-016-0146-8 https://doi.org/10.1007/s40617-016-0146-8
• Mattson, S. L., Higbee, T. S., Nichols, B., Aguilar, J., & Campbell, V. E. (2024). Effects of an activity schedule intervention package on cooperative vocal exchanges during learning centers. Behavior Analysis in Practice, 17, 1175–1190. https://doi.org/10.1007/s40617-024-00939-9 https://doi.org/10.1007/s40617-024-00939-9
• Frampton, S. E., & Axe, J. B. (2025). A preliminary investigation into teaching adolescents with autism to use apps to solve problems. The Analysis of Verbal Behavior, 41, 26–39. https://doi.org/10.1007/s40616-024-00212-8 https://doi.org/10.1007/s40616-024-00212-8
• Frampton, S. E., Axe, J. B., Davis, C. R., Meleshkevich, O., & Li, M.-H. (2024). A tutorial on indicating responses and their importance in mand training. Behavior Analysis in Practice, 17, 1238–1249. https://doi.org/10.1007/s40617-024-00965-7 https://doi.org/10.1007/s40617-024-00965-7
• Olaff, H. S., & Holth, P. (2025). Acquisition of incidental bidirectional naming: Isolating the effects of probing and mixed-operant instruction. The Analysis of Verbal Behavior, 41, 200–234. https://doi.org/10.1007/s40616-025-00221-1 https://doi.org/10.1007/s40616-025-00221-1
• Anderson, B. K., & Wiskow, K. M. (2025). Acquisition of secondary targets during tact and intraverbal instruction with instructive feedback. The Analysis of Verbal Behavior, 41, 40–56. https://doi.org/10.1007/s40616-025-00215-z https://doi.org/10.1007/s40616-025-00215-z
• Garvey, C. C., Ferguson, J. L., Milne, C., Cihon, J. H., Leaf, J. B., Leaf, R., McEachin, J., & Schulze, K. (2022). Comparing In-View to Out-of-View Stimulus Arrangements When Teaching Receptive Labels for Children Diagnosed With Autism Spectrum Disorder. Behavior Analysis in Practice, 15(2), 475-484. https://doi.org/10.1007/s40617-021-00596-2 https://doi.org/10.1007/s40617-021-00596-2
• Thakore, A., Kelly, A., Petursdottir, A. I., & Stockdale, M. (2024). Evaluation of a treatment package for chronic, stereotypic hand mouthing of a child diagnosed with autism. Behavior Analysis in Practice, 17, 915–920. https://doi.org/10.1007/s40617-024-00956-8 https://doi.org/10.1007/s40617-024-00956-8
• Radogna, C., D’Angelo, G., & Lerman, D. C. (2024). Assessing and teaching job‑related social skills to adults with neurodevelopmental disorders in Italy. Behavior Analysis in Practice, 17, 803–814. https://doi.org/10.1007/s40617-023-00873-2 https://doi.org/10.1007/s40617-023-00873-2