Cueing & Prompt Hierarchy in Speech Therapy

Two students end the week at 82% accuracy on initial /s/. On paper they look identical. But one named every picture cold, with nothing on the screen but the image, while the other needed a full spoken model on every single trial to get there. Those are not the same student, and they are nowhere near the same point in therapy — yet raw accuracy flattens them into one number and hides all the support propping up the score.

A prompt hierarchy lets you record the support behind each correct production, so progress reads as reducing scaffolding — growing independence — not just a rising percentage. That single shift is what separates a defensible progress note from a hopeful one.

This guide walks the whole arc: what a cueing hierarchy is and why systematic fading matters, the most-to-least continuum of support, a clean four-level way to operationalize it that you can actually record, when to fade a step away, and — the part almost every other guide skips — how to log prompt level so it flows straight into your notes and goals. It is written for the SLP running the session, so the jargon stays.

One honesty note, stated once: this is a clinical workflow guide, not clinical advice. Sound Safari is a clinical tool, not a medical device — it does not provide diagnoses or treatment recommendations, and clinical decisions always rest with the qualified professional.

What a Cueing Hierarchy Is (and Why Fading Matters)

A cueing or prompt hierarchy is a system for delivering support in a deliberate order. It starts with more supportive cues given more often, then systematically fades to less supportive cues given less often — and it holds at each level until the child succeeds consistently before any support is removed. The support is the variable; the hierarchy is how you move it on purpose instead of by feel.

That structure is not arbitrary. It is scaffolding, grounded in Vygotsky’s zone of proximal development: support is tailored to the learner’s current ability, faded gradually as competence grows, and responsibility is transferred to the learner over time. This is the theoretical reason that reducing support at stable accuracy is progress — the whole point of a scaffold is to eventually come down.

The evidence base is concrete. In a 2020 survey of speech-sound-disorder management published in ASHA’s Language, Speech, and Hearing Services in Schools, instructional cues — prompting, modeling, and feedback — are cited as among the most significant variables affecting motor-skill learning (Kim et al., 2012). Crucially, that support is meant to be faded across therapy: early on, feedback is “structured, explicit, immediate, and frequent”; toward the end it becomes “incidental, natural, delayed, and less frequent.”

A quick terminology note, because it comes up constantly: most SLPs use prompt and cue interchangeably in articulation work. If you want to split hairs, a prompt nudges a response to happen while a cue guides how the sound is produced — but for data purposes what matters is the level of support behind the response, whatever you call it. Don’t over-litigate the words.

Why insist on a systematic ladder rather than ad hoc help? Because “I gave some cues today” is unmeasurable. A fixed ladder turns support into something you can name, record, and reduce on purpose — which is the entire premise of the rest of this guide.

The Continuum of Support: Most to Least

Before narrowing to something recordable, it helps to see the full clinical range. Support in articulation therapy runs along a continuum, from the most hands-on cues to none at all:

Here is the honest problem: published hierarchies do not agree. Different published versions use anywhere from roughly five to eight steps, and they order those steps differently — there is no single universal ladder. What matters clinically is not which published version you adopt, but that yours is consistent from session to session and, above all, recordable.

It is worth knowing which of these techniques SLPs actually reach for. In that survey, phonetic placement cues, iconic gestures, and verbal or pictorial cues are among the most frequently used elicitation techniques — used by roughly two-thirds of SLPs (Brumbaugh & Smit, 2013). The continuum above is not academic; it maps onto everyday practice.

A continuum like this is clinically rich but awkward to log trial by trial — you are not going to tick one of a dozen fuzzy categories on every response in a fast group session. The practical move is to collapse it into a small, fixed set of levels you can record in a tap.

That is exactly what Sound Safari does — and it is worth being precise about what those levels are, and just as importantly what they are not: the app’s levels are defined by what appears on screen, not by tactile or physical placement cues. The continuum above is general clinical background, not a claim about app features.

Sound Safari’s Four Prompting Levels

Sound Safari operationalizes the whole continuum with a single, concrete question: which supports are on the screen? There are exactly three — the Picture (the illustration of the word), the Word (its text), and the Model (auto-played audio of the word) — and the combination in view defines the prompt level. That yields four levels, ordered from most support to least.

The top of that ladder is genuinely counterintuitive, and it is worth slowing down on. Independent (Picture Only) is the least supportive level — even though something is still on screen — because the only thing left is the picture. There is no word text to read and no spoken model to copy, so the child has to retrieve and name the target spontaneously. Minimal (Word Only), by contrast, still hands over the word itself. Read the levels by what has been taken away, not by how much is left on screen, or you will misread the order.

The scaffold is enforced structurally, not just by convention. The spoken model auto-plays only at Maximum — it is intentionally withheld at every level below it, so the hierarchy cannot quietly leak the answer once you have decided to fade past a full model. The star rating is a fast visual read of the same idea, running inverse to the internal ranking: more stars means more support, so a four-star trial tells you at a glance how propped up that production was.

All four levels are enabled by default. You can hide levels that do not fit a particular student — trimming the ladder to the two or three steps you are actually working across — but the app will not let you remove the last one; at least one level always has to stay enabled.

Notice what this buys you: a continuum that was too fuzzy to log has become four named states, each with a fixed cue composition. “Maximum” always means Picture + Word + Model — not “a lot of help, roughly.” That fixed definition is what makes the next two problems — knowing when to fade, and recording what you did — tractable.

How to Fade Prompts as Accuracy Stabilizes

Most guides tell you to fade “when the child is consistent” or “when they’re ready,” which is true and nearly useless — it gives you nothing to write down. A more defensible rule: set an accuracy criterion at the current level, meet it, then drop one step of support. A common, reasonable criterion is about 80% correct across two consecutive sessions before you fade — then you re-establish that same criterion at the lower level and repeat.

In the app’s terms, fading walks the ladder one support at a time:

• Maximum → Moderate: drop the auto-played model; the picture and word remain.
• Moderate → Minimal: drop the picture; only the word remains.
• Minimal → Independent: drop the word; only the picture remains, forcing spontaneous naming.

Each drop removes exactly one support and tests whether accuracy holds without it. That is the whole game — not “less help in general” but a specific, named layer withdrawn and then re-proven.

Which direction you fade depends on where the skill is. Most-to-least — start with high support and fade down — is what you use while a skill is still being established; it keeps success high and errors low, closer to errorless learning.

Least-to-most — start minimal and add support only when the child errs — is what you use later, to probe how much independence has actually generalized (a framing Allison Fors lays out well). The same target can travel from most-to-least in teaching to least-to-most in generalization checks.

Fading is not a one-way ratchet. If accuracy collapses after you drop a support, step back up a level rather than grinding away at a level the student cannot hold. A clean move back to more support is better data — and better therapy — than a string of failed trials at a level that was one step too far.

All of this depends on tracking the criterion at each level, not overall. “80% at Moderate” and “80% at Independent” are wildly different clinical facts, and an average across levels blurs them into a number that means nothing. Keep the accuracy criterion level-specific, and “stable at this level” becomes a real claim you can act on — which raises the question of how you record it in the first place.

Recording Prompt Level in Your Data

Here is where almost every hierarchy guide stops short: none of them connect cue level to the actual data record. Naming a beautiful ladder is worthless if the level never lands on the trial. The fix is simple to state — record prompt level on every scored trial, not once per session — and it changes what your data can tell you. The practice session still carries an overall prompting level as a first-class field, but it is the per-trial record that makes a trend legible.

Do that, and each trial becomes a two-dimensional data point: it captures both the support level and the response quality. Scoring in Sound Safari has three types — Correct, Incorrect, and Approximate — with Approximate an opt-in setting (the default is Correct and Incorrect only). Only Correct counts toward accuracy; approximations are tracked but not scored as correct, so a shaping-in-progress response stays visible without inflating the percentage.

Cross the two dimensions and the picture sharpens. “Correct at Independent” is true, unpropped independence. “Correct at Maximum” is a correct production that needed every available support. Same word, same score type — completely different clinical meaning, and only the recorded prompt level tells them apart.

Which brings back the two students from the top of this article. Both sit at 82%. Once prompt level is on the record, one is Correct-at-Maximum and the other Correct-at-Independent — and the daylight between them is obvious. Accuracy alone erases that difference; prompt-level data restores it. The trend you actually want to read is “the same accuracy at less support over time,” which is growing independence expressed as a number.

One important boundary: not every data track should carry a cue. Probes and the built-in screener deliberately record no prompt at all — they are meant to be a clean, unprompted, independent baseline. Cueing is a treatment variable, not a baseline one, so mixing them would contaminate the very read a baseline exists to give.

The screener itself samples sounds across the initial, medial, and final positions, scored against developmental norms from Crowe & McLeod (2020), the U.S. English companion to the cross-linguistic McLeod & Crowe (2018). For how baselines and probes stay separate from teaching data, see our guide to baseline data in articulation therapy.

The practical payoff is quiet but real: because prompt level is captured as part of scoring each trial, the level is already on the data — there is no separate tally sheet to keep and no end-of-session reconstruction of “how much did I help.” The support behind every correct production is simply part of the record.

From Data to Documentation: Prompt Level in the SOAP Note

Collecting accurate cue data is only half the problem; the other half is getting it into the note without retyping it. This is the second thing competitors leave unsolved, and it is where recording the level during the session pays off.

The recorded level flows into the SOAP note automatically. A merge field, {promptingLevelName}, resolves to the exact level name — Maximum, Moderate, Minimal, or Independent — so the note reflects what actually happened in the session rather than a value you re-key from memory. (A companion numeric field carries the raw 1–4 value for templates that prefer a number.)

Both built-in templates — the app ships two, a Standard SOAP and a Brief SOAP — print the level in the Objective section:

Objective: Student produced initial /s/ in structured word-level tasks. Prompting Level: Moderate. Accuracy: 80% across 20 trials.

The Assessment goes a step further and contextualizes accuracy by support rather than reporting a bare percentage — it reads, in effect, “current 80% at Moderate prompting.” That single phrasing is the whole thesis of this article compressed into one documentation line: the number means little without the level attached.

The auto-generated next-steps then surface documentation prompts that mirror the fading logic. Each is a starting point the SLP reviews and keeps, edits, or overrides before finalizing the note — these are informational prompts, not clinical actions the app takes on its own, and the decision always rests with the SLP.

At 90% accuracy or above, the note surfaces a prompt to consider reducing the prompting level to promote independence. Below 60%, it surfaces a prompt to consider increasing support. At the Independent level with accuracy at 80% or above, it surfaces a prompt to consider evaluating discharge criteria.

The same underlying value is labeled for its reader. In the SOAP draft it appears as “Prompting level”; in a parent-facing update the identical data is labeled “Support level during practice,” because clinical shorthand does not belong in a note sent home. Same fact, right register.

If you want the fuller picture of how session data becomes a finished note, our guide to SOAP notes in speech therapy walks the whole flow. The relevant point here is narrower: because the prompt level was captured during scoring, it arrives in the documentation already correct.

Writing the Fading Target Into the IEP Goal

The loop closes in the goal itself. A well-written articulation goal does not just name an accuracy target — it names a support target, and it moves over time from “with maximum cues” or “with a model” toward “with minimal cues” and, ultimately, “independently.” The cue level is part of the measurable criterion, not a footnote to it.

Put two goals side by side:

Accuracy only: “Student will produce /s/ in the initial position with 80% accuracy.”

Accuracy + support: “Student will produce /s/ in the initial position with 80% accuracy at minimal cueing across two consecutive sessions.”

The first is met by a student who needed a full model on every trial. The second can only be met at a named support level, held across sessions — which is precisely what the recorded prompt level lets you verify. You are progress-monitoring against data, not against a memory of how much you helped.

That verification is only trustworthy because the level is on every trial. When the goal says “at minimal cueing,” meeting it is a data claim you can point to, not a clinical impression you are asking the team to take on faith.

Writing goals like this is easier from a starting bank. Sound Safari’s IEP goal bank spans eight goal categories — Articulation, Phonology, Fluency, Voice, Receptive Language, Expressive Language, Pragmatic Language, and Phonological Awareness — so the cue-level scaffold can be written into a goal in any of them, not just articulation. For copy-paste SMART goal templates that already carry the accuracy-and-cue structure, see our guide to articulation IEP goals.

Record the support, fade it on a criterion, document it automatically, and write it into the goal — and “82% accuracy” stops being a flat number. It becomes a story about a student who needs less help than they did last month, told in data you can defend.

Frequently Asked Questions