Coding taxonomies
Record a role-play, paste the transcript, and code it the way a MITI coder would—reflections, questions, affirmations—with your ratios updating as you go. Then train the ear with two drills: name the skill, and find the reflection that goes deeper.
Free · No sign-up · Runs entirely in your browser
Paste a practice session with speaker labels. Click a clinician turn and press a key for each skill you hear; the reflection-to-question ratio, complex reflections and the rest update as you go. The example is already coded so you can see the whole picture first.
↑/↓ move · Q S C A K … toggle codes · O open/closed
1Thanks for coming in. What would you like to talk about today?
Client: My doctor said I should cut down on drinking. My liver numbers were off.
2So your doctor is worried, and you're here partly because of that.
Client: Yeah. I mean, I don't think it's that bad. Everybody at work has a few beers after shifts.
3Drinking is part of how you unwind with your coworkers, and it doesn't feel unusual to you.
Client: Exactly. It's not like I'm drinking in the morning.
4Do you drink every day?
Client: Most days. Four or five beers. More on weekends.
5How do you feel about those liver results?
Client: Honestly? It scared me a bit. My dad had liver problems.
6Seeing those numbers brought your dad to mind, and that's not a road you want to go down.
Client: No. He was sick for a long time before he died.
7That must have been really hard to watch.
Client: It was. I was the one driving him to appointments.
8You took care of him when he needed it. That says a lot about you.
Client: I guess. I just don't want my kids to go through that.
9Part of you feels drinking is normal and harmless, and part of you is worried you could end up where your dad did.
Client: Yeah, that's pretty much it.
10You really need to stop drinking completely, you know. Four or five a day is way over the limit.
Client: I knew you'd say that.
11I'm sorry, I jumped ahead. What you do about this is entirely your decision.
Client: I appreciate that. I could maybe try not drinking during the week.
12Would it help if I shared what other people in your situation have tried?
Client: Sure, go ahead.
13Some people keep weekdays alcohol-free and plan something else for after work, like the gym or a call home. Others just switch to smaller cans. Which of those sounds more like you?
Client: The after-work thing. I could go straight home and play with my kids.
14Going home to your kids instead of the bar would be doing exactly what you said matters most to you.
1.50 : 1
reflections : questions
5 of 6 reflections · fair 40%, good 50%
3 of 4 · MITI 3.1 benchmark · fair 50%, good 70%
3
MI-adherent
Affirm + seek + autonomy
1
Non-adherent
Persuade + confront
62%
Clinician talk
Share of all words
Fair/good bands are the MITI 4.2.1 manual's suggestions, which it describes as expert opinion without normative data—use them for feedback, not as a pass mark.
Session skyline
Each column is a clinician turn; colours stack its codes.
Reference · for the curious
Motivational interviewing is learned the way any conversational skill is learned: by doing it, hearing it back, and getting specific feedback. Workshops teach the ideas; what changes practice is counting. Trainees who code their own role-plays discover quickly that they ask far more questions than they reflect, and that most of their reflections repeat rather than deepen. This page is built around that feedback loop.
Miller and Rollnick's core skills are summarised as OARS: open questions, affirmations, reflections and summaries. The Motivational Interviewing Treatment Integrity code (MITI), the standard instrument for measuring whether a conversation was MI, counts these skills with its own behaviour codes. Questions are one code in MITI 4—open and closed are no longer distinguished. Reflections split into simple and complex. Affirmations sit with seeking collaboration and emphasizing autonomy as the three MI-adherent behaviours. Summaries have no code of their own and are coded as reflections, usually complex ones. On the other side, persuading without permission and confronting are the MI non-adherent behaviours.
Two coding rules matter for anyone counting by hand. Each clinician turn earns at most one reflection code—complex if any part of it was complex—and at most one question code, however many questions were asked. Without those rules, one long turn full of questions would swamp the ratios, which is why the tool enforces them.
Moyers and colleagues (2016) define the summary scores the tool computes. The reflection-to-question ratio is total reflections divided by total questions. Percent complex reflections is complex reflections divided by all reflections. The two MI-adherent and non-adherent totals are plain counts. The Technical global averages the ratings for cultivating change talk and softening sustain talk; the Relational global averages partnership and empathy.
The MITI 4.2.1 manual suggests thresholds for "fair" and "good" practice: a ratio of 1:1 and 2:1, 40% and 50% complex reflections, and global averages of 3 and 4 (technical) and 3.5 and 4 (relational). The manual is candid that these are expert opinion without normative data, and they are shown here the same way—as feedback targets, not a pass mark. For comparison, the recorded sessions used to validate MITI 4 averaged about 0.9 reflections per question. The earlier MITI 3.1 also set targets for open questions—50% fair, 70% good—which MITI 4 dropped but many trainers still use, so the tool tracks open versus closed as an optional extra.
A simple reflection tells the client they were heard. A complex reflection tells them they were understood, and it steers: by choosing which part of what the client said to reflect, and how, the practitioner shapes where the conversation goes next. Reflecting the value under a complaint, the feeling under a refusal, or both sides of an ambivalence at once invites the client to keep talking about change. That is why the MITI rewards the proportion of complex reflections, and why the second drill on this page trains nothing else. Our explainer on change talk and sustain talk covers the client side of the same conversation.
A workable routine: record a 15–20 minute role-play, transcribe it (automatic transcription is fine for this), paste it in, and code it before the supervision meeting. Bring the feedback card and two or three turns you were unsure how to code—those disagreements are where the learning happens. Coding the same session again after a few weeks of practice is the simplest progress measure there is.
If two people code the same session—a supervisor and a trainee, or two research coders—their agreement can be measured rather than argued about. For MITI counts the ICC calculator is the right tool for summary scores across sessions, and the inter-rater reliability calculator handles code-by-code agreement on the same utterances. The differences between the MITI and its predecessor, the MISC, are covered in MISC vs MITI, and a full walkthrough of MITI coding is in our MITI coding guide.
It does not decide the codes for you: apart from pre-marking turns that end in a question mark, every code is yours. It does not reproduce the MITI manual, whose definitions and examples are needed for formal coding; the one-line definitions here are our own summaries. It treats each clinician turn as one volley, so turns that should be split are coded as one. And it cannot judge the global ratings from the transcript—those need a human listening to the whole session. For research-grade fidelity monitoring, the MI research workflow covers coding full sessions with calibrated coders and AI assistance.
OARS stands for Open questions, Affirmations, Reflections and Summaries—the four core communication skills of motivational interviewing described by Miller and Rollnick. Open questions invite the client to elaborate; affirmations recognise genuine strengths and efforts; reflections restate or deepen what the client said; summaries pull several reflections together. In MITI coding, summaries are coded as reflections (usually complex ones), and affirmations are one of three MI-adherent behaviours.
The MITI 4.2.1 manual suggests at least one reflection per question (1:1) as fair and two per question (2:1) as good, and describes both as expert opinion rather than normative data. For context, the untrained and mixed-skill sessions used to validate MITI 4 averaged about 0.9 reflections per question (Moyers et al., 2016). A trainee who asks question after question usually lands well below 1:1, which is why the ratio is the most useful single number to track in practice.
A simple reflection repeats or rephrases what the client said, adding little or no meaning: “You've tried quitting three times.” A complex reflection adds something the client implied but did not say—a feeling, the value underneath, the other side of their ambivalence, or where they seem to be heading: “You keep coming back to it, even though it's been frustrating—this matters to you.” The MITI's percent complex reflections is CR ÷ (SR + CR); 40% is suggested as fair and 50% as good.
A MITI coder listens to a 20-minute segment and assigns behaviour codes to each clinician utterance: question, simple or complex reflection, affirm, seeking collaboration, emphasizing autonomy, giving information, persuade, persuade with permission, and confront. Only one reflection code and one question code are given per volley. The coder also rates four global dimensions from 1 to 5. Summary scores are then computed from the counts. This tool reproduces the counting and the summary scores; formal MITI coding for research also requires trained, calibrated coders.
The righting reflex is the helper's urge to fix the client's problem—advising, warning or arguing for change before the client is ready. In motivational interviewing it tends to provoke sustain talk, because a client who hears one side argued for them tends to voice the other. In MITI terms it usually shows up as persuade without permission or confront, both of which count as MI non-adherent. The reflection drill includes a righting-reflex option in every item so you learn to recognise it.
No. It implements the published MITI 4 summary score formulas and uses our own paraphrased definitions of each code; it does not reproduce the MITI manual, and it does not replace training with calibrated coders. Use it for practice and supervision feedback. For research-grade fidelity coding, the MITI manual is available from the Center on Alcohol, Substance use, And Addictions (CASAA) at the University of New Mexico.
Written by Enrique Gutiérrez, PhD (Computer Science)—founder of Tagaroo and Associate Professor of Computer Science, working on inter-rater reliability, measurement and annotation methodology (ORCID).
Last verified: 24 September 2026. Formulas, thresholds and cited figures on this page were checked against their original sources on that date. Every calculation runs in your browser; nothing you enter is transmitted or stored.
Hand-coding a 20-minute session takes an experienced coder about an hour. Tagaroo applies MITI behaviour codes to full transcripts with a rationale for every code, then measures its agreement with your human coders—so supervision feedback and fidelity checks stop waiting on coder time.
Motivational interviewing skills practice · tagaroo.ai/materials/motivational-interviewing-skills-practice · figures verified against primary sources 2026-09-24. Educational scoring aid, not a diagnosis, and not reviewed by a licensed clinician.