The Elicitation Engine: How AI Maps Your Exact Experience to a Personalized Session (according to 2025 research)

Most AI therapy apps follow the same pattern: you type “I feel anxious,” and a chatbot offers you a CBT exercise from a library. The response might be well-written and empathetic, but it’s essentially a lookup — a pre-written protocol matched to a broad category label.

This is useful for many things. It is not personalization.

Personalization means the session itself is built from your input — not just matched to it. That requires a fundamentally different architecture: one that can extract structured data from natural conversation and synthesize it into a personalized guided hypnosis and visualization session.

Here’s how Oriamind’s Elicitation Engine does that.

A 2025 study in Frontiers in Public Health confirmed that generative AI in mental health shows the strongest outcomes when personalization is deep — not just matching topics, but mapping the user’s specific linguistic and sensory patterns.

Step 1: The Elicitation — Extracting Structured Data from Conversation

The Elicitation Engine is not a standard chatbot. It operates as an NLP practitioner — not an interviewer — using specific linguistic techniques to extract the user’s exact internal experience.

Semantic Mapping — The engine identifies and mirrors the user’s exact words and metaphors. If you say “heavy fog,” it works with “heavy fog,” not “it sounds like you’re feeling confused.” The distinction matters because the subconscious encodes experience in sensory-specific language, and the session must use the user’s language to be effective.

Sensory Drilling — Using a structured framework derived from NLP’s submodality model, the engine explores the sensory qualities of the user’s experience:

Visual: What do you see when you feel this way? Is it bright or dim? Close or far?
Auditory: Is there a sound associated with it? Loud or soft? Internal or external?
Kinesthetic: Where in your body do you feel it? What’s the quality — heavy, tight, hollow, electric, numb?
Olfactory/Gustatory: (Less common, but relevant for certain experiences)

Anchor and Depth — The engine identifies the trigger context (when does this happen?) and the desired state (what would you rather feel?) — establishing the before and after that the protocol will bridge.

Each of these data points maps directly to a parameter in the guided hypnosis and visualization session. The problem word becomes the target of the suggestion phase. The submodalities become the sensory pivots. The anchor context becomes the future-pacing scene.

Step 2: Session Synthesis — From Data to Structure

Once the elicitation has collected sufficient data (typically 6 data points: problem word, desired state, submodality-problem, submodality-desired, anchor trigger, depth preference), the session is synthesized.

The synthesis engine constructs a four-phase guided hypnosis and visualization session:

Induction — Progressive relaxation targeting the specific tension pattern identified in the elicitation. If the sensation is “tight band around the chest,” the induction works with chest and breath. If it’s “hollow pit in stomach,” the induction addresses the abdomen.

Deepening — Fractionation techniques calibrated to the user’s depth preference. The counting and imagery are drawn from the user’s own sensory language.

Suggestion — This is where the actual repatterning happens. The suggestion uses Milton Model language patterns — artfully vague constructions that allow the subconscious to fill in the specifics. The problem state is addressed via submodality shift (the “heavy” becoming “light,” the “tight” becoming “warm,” the “close” becoming “distant”). A post-hypnotic anchor is installed for the trigger context identified in the elicitation.

Awakening — Future-pacing that bridges the new state into the user’s daily context. The specific triggers identified in the elicitation are used as the rehearsal scenes.

Step 3: SSML Transformation — Precision in Audio

The raw session text is not ready for audio delivery. It needs to be transformed into Speech Synthesis Markup Language (SSML) — the XML standard that controls how text-to-speech systems render audio.

This transformation is not trivial. A guided hypnosis and visualization session requires:

Prosody control — downward pitch contours for embedded commands, rhythmic patterns for deepening
Tempo variation — slowing during key suggestion passages, pacing during induction
Breath markers — synchronization of breathing suggestions with pause durations
Phase bookmarks — timestamps that enable the app’s breathing orb and grounding overlay to synchronize with the audio

The SSML is rendered through Azure Neural TTS using high-definition voices selected by the user. The output is a high-fidelity audio file that preserves the structure of the session while delivering it in natural, warm speech.

Step 4: The Working Memory — Continuity Across Sessions

Every session is informed by previous sessions. The working memory profile stores:

Language patterns (metaphors, resonant words from past sessions)
Somatic map (body locations and submodalities)
Semantic profile (core issues, desired states)
Behavioral data (depth preference, session ratings)

This means the second session doesn’t start from scratch. The Elicitation Engine already knows your language, your patterns, and what’s worked before. The protocol becomes more precisely targeted with each interaction.

Why This Matters

The difference between a lookup and a generation is the difference between a library and an engine. A lookup matches your input to the closest pre-written option. A generation builds something new from your raw material.

The Elicitation Engine is the component that makes generation possible. Without it, you’re just getting a pre-recorded track with better matching.

With it, the protocol is built from your exact words, your exact sensations, your exact context. That’s not a better library. It’s a different category.

Adam Shaaban is the founder of Oriamind. LinkedIn · X / Twitter

Try It

If you’re curious how your own language would be mapped, try this exercise:

Describe a current challenge in one sentence. Use your natural words.
Notice the metaphors you used — “heavy fog,” “electric current,” “tight knot”
Notice where in your body you experience it
Notice the quality — temperature, texture, movement

This is the same data the Elicitation Engine extracts. Most people discover they have a richer internal language than they realized — and that language is the key to changing the pattern.

This article is part of our How Oriamind works (technical series) series.