NotuDocs vs Upheal: Template-First Notes vs AI Session Recording for Therapists

If you are researching AI documentation tools for therapy practice, you have probably come across both NotuDocs and Upheal. Both set out to solve the same problem: the post-session paperwork that follows therapists home. But they solve it from opposite ends of the problem, and those different starting points matter more than any feature comparison.

This article covers how each tool actually works, what the documented hallucination risk looks like in practice, the privacy implications of session recording, how pricing compares across different use cases, and which practitioners are genuinely better served by each approach.

How Each Tool Works

The architectural difference between these two tools explains almost every downstream tradeoff. Understanding it first makes the rest of the comparison easier to interpret.

Upheal: Record the Session, Generate the Note

Upheal is built around session recording. You conduct the therapy session while the recording runs, Upheal transcribes the audio, and the AI generates a structured progress note from that transcript. The basic workflow is:

1. Obtain client consent to record the session
2. Start Upheal at the beginning of the appointment (works in-person and for telehealth)
3. Conduct the session normally
4. Review the AI-generated note after the session ends
5. Edit, approve, and sign

Beyond documentation, Upheal includes analytics for monitoring client progress over time, tracking emotional patterns across sessions, and generating treatment plan summaries. For therapists who want a platform that encompasses more than note generation, those features are part of what the pricing reflects.

The core appeal is real: you press record, conduct the session, and a structured note is waiting for you afterward. For therapists who find post-session writing genuinely taxing, or who conduct long, complex sessions where writing feels impossible immediately after, this automation level has clear value.

NotuDocs: Write Your Observations, AI Fills Your Template

NotuDocs works from the opposite direction. No recordings, no audio, no session capture. You write your own observations after the session (or in brief moments between sessions), select a documentation template you have already built or customized, and the AI fills the template placeholders using only the content you provided.

The workflow looks like this:

1. During or right after the session, write your notes in your own words (bullet points, brief sentences, clinical observations, key quotes)
2. Select your template (SOAP, DAP, BIRP, GIRP, custom format, or any combination)
3. The AI fills the template structure using only what you wrote
4. Review, adjust, export, and sign

The constraint here is intentional. If a template section has no corresponding input from you, it flags the gap rather than inventing content to fill it. The AI's role is structural organization, not authorship.

NotuDocs is $25 per month for Pro, with a free tier that includes 3 templates and 3 notes per month.

The Hallucination Question: Not Hypothetical

This is the most consequential part of any comparison between recording-based AI documentation and template-first documentation. It deserves direct treatment.

Generative AI tools that produce clinical notes from session audio have a structural problem: the model must fill a structured template from a transcript that is often incomplete. Quiet passages, overlapping speech, client affect that doesn't translate to words, pauses, and ambiguous phrasing all leave gaps. When the model needs to fill those gaps, it produces text that sounds clinically coherent. That is what language models are designed to do.

In clinical documentation, that behavior has serious consequences.

Upheal has been specifically cited in therapist communities for a documented incident in which the tool fabricated a childhood abuse history in a session note. The generated note attributed a childhood abuse history to a client who had never disclosed one, in a session where the topic was never raised. The therapist caught the error before the note was finalized. But the question the incident raised was not whether the tool had a bug. It was whether the architecture itself makes this kind of error predictable.

The answer, based on how generative AI documentation works, is yes. When a tool generates clinical progress notes from audio transcription, any ambiguous or absent content in the transcript creates an opportunity for the model to fill in something plausible. A model trained on clinical documentation will produce clinically plausible content, including trauma histories, expressions of suicidal ideation, descriptions of abuse, and other high-stakes clinical details. These are exactly the kinds of fabrications that therapists in professional communities have reported across multiple AI note tools, not only Upheal.

The Upheal incident stands out because childhood abuse history is not a minor formatting error. It is the kind of clinical detail that shows up in legal proceedings, insurance audits, safety planning documentation, and records that follow a client across providers. A fabricated history in that field is a professional liability, not a transcription glitch.

Upheal is aware of this problem and has iterated on reducing hallucination rates. The risk is lower than it was in earlier AI documentation tools. But the architecture remains: the model is generating clinical content from an audio source, and wherever that source is incomplete, the model fills in.

Template-first documentation eliminates this category of risk by design. When you write your own observations and the AI organizes them into a template you defined, the AI's job is structural. It cannot fabricate a client history you didn't write. The content of the note comes from you and from nothing else.

A practical test for any tool you are evaluating: submit a note with one required section intentionally left blank. Does the tool flag the empty field, or does it fill it with something? The answer tells you how the tool handles clinical uncertainty.

Privacy and the Recording Question

Therapy works because clients trust that what they say in the room stays in the room. Recording-based documentation requires you to change that assumption explicitly and proactively with every client.

To use Upheal compliantly, you need to:

• Explain the recording process to each client and obtain explicit, documented consent before the first recorded session
• Disclose how the audio is processed, where it is stored, and for how long
• Have a process for clients who consent initially but withdraw consent later
• Understand your obligations if audio recordings are ever subject to a legal request
• Think through what happens with session content that was transmitted to Upheal's servers during transcription and note generation

Some clients handle this without any issue. Others do not. The populations for whom recording is most likely to be clinically complicated include clients with trauma histories involving surveillance or coercive control, domestic abuse survivors, undocumented individuals, clients in active custody proceedings, and anyone whose past experience with documentation has been harmful. For these clients, the recording itself (not just what happens to it afterward) can shift what is possible in the therapeutic relationship.

This is not a reason to say recording-based tools are wrong. Millions of telehealth sessions are conducted on video platforms that are inherently recorded. Clients give consent and therapy proceeds effectively. But for therapists whose caseloads include privacy-sensitive populations, the recording requirement is a clinical consideration before it becomes a technical one.

NotuDocs removes this question entirely. The input is text you write. No audio is captured, no session content is transmitted as a recording, and the in-session dynamic stays unchanged. The privacy surface is smaller, and the consent conversation does not need to happen.

It is worth noting: Upheal does retain session audio and transcripts. The specific terms around storage duration and access should be reviewed in their current privacy documentation before adoption. The relevant questions for any recording-based tool are: where is the audio stored, for how long, under what conditions can it be accessed, and what does any BAA actually cover.

NotuDocs does not currently carry HIPAA compliance or offer BAAs. For practices where a signed BAA is a hard requirement before adopting any third-party documentation tool, that is a consideration that may settle this comparison for some readers before other factors come into play.

Template Control and Clinical Voice

A common complaint about AI-generated clinical notes is that they do not sound like the clinician who signed them.

Notes generated from session transcripts tend toward a generic clinical register. Phrases like "client verbalized distress regarding interpersonal difficulties" and "therapeutic rapport was established" appear regardless of what actually happened. The model has learned what therapy notes sound like and reproduces that pattern. The note passes a surface-level check but doesn't reflect your clinical thinking, your vocabulary, or your professional judgment about what was significant.

This matters for multiple reasons. Supervision, insurance audits, and peer consultations all involve someone reading your notes and drawing conclusions about your clinical reasoning. Notes that sound like they were generated from a template rather than written by a clinician raise questions you would rather not have to answer.

Template-first documentation inverts this problem. The structure is yours from the beginning. If your SOAP notes have a specific way of framing the Assessment section that reflects eight years of training and practice, you build that into the template. If you document differently for individual adult sessions versus family sessions, you build separate templates. If you have a particular format required by a supervising clinician or a specific payer, you define it once.

In NotuDocs, templates are fully editable. You set the sections, the field names, the clinical language, and the prompt language the AI uses to fill each field. The result sounds like you because the structure was authored by you.

Upheal allows some note customization, but the generation model is still making authorial decisions based on the transcript. You edit what the AI wrote. Template-first means the AI is filling what you wrote.

Pricing: The Real Comparison

Upheal's pricing runs from $19 to $99 per month depending on the plan.

The $19 entry tier covers basic note generation with a limited number of sessions per month. The higher tiers (ranging to $99/month) include the analytics features, treatment plan tools, expanded session capacity, and access to Upheal's platform integrations. For therapists who want the full Upheal platform including progress tracking and outcome measurement, the pricing reflects that scope.

NotuDocs is $25 per month for Pro, with a permanent free tier that includes 3 templates and 3 notes per month.

The pricing comparison depends on what you are actually using. If you primarily want to reduce time spent on session documentation and have no use for progress analytics or outcome measurement dashboards, the relevant Upheal tier is the lower end of their range, and the difference from NotuDocs' $25 is modest.

For a solo practitioner in private practice seeing 15 to 20 clients per week, the more meaningful question is: what do you actually need the tool to do, and does the workflow match how you already practice? A $19/month tool that requires session recording is not automatically better than a $25/month tool that doesn't, if recording is something you cannot do with your client population.

The free tier structure also differs. Upheal offers a free trial period before billing begins. NotuDocs' free tier is permanent at reduced volume, which means therapists who see fewer clients (or who want to test the tool over an extended period with real session data) can evaluate it without a billing clock running.

Language Support

Upheal is built primarily for English-language clinical documentation. Spanish is not a native design feature, and the depth of Spanish-language clinical terminology support is limited accordingly.

NotuDocs was built bilingual from the beginning. English and Spanish are both fully supported across the template editor, note generation, and exports.

This matters practically. Spanish-language clinical documentation has its own conventions. Terms like "alianza terapéutica" (therapeutic alliance), "ideación suicida" (suicidal ideation), "regulación emocional" (emotional regulation), or "rumiación cognitiva" (cognitive rumination) appear in clinical Spanish as established vocabulary, not as translations of English terms. A tool built for bilingual clinical practice handles these terms differently from one that processes Spanish text as a secondary capability.

For therapists in the US working with Spanish-speaking clients, or for practitioners in Latin America documenting in Spanish, this distinction has direct functional implications. A generated note that uses awkward or imprecise Spanish clinical terminology creates problems that are invisible until a colleague, supervisor, or auditor reads it.

A Side-by-Side Look at the Two Workflows

Take a therapist seeing nine clients on a Thursday. The last session ends at 5:45 PM.

With Upheal, nine AI-generated notes are waiting in the queue. She reviews each one: checking whether the clinical content accurately reflects what happened (not what the AI inferred from audio), correcting transcription errors, adjusting language that sounds generic rather than clinical, and verifying that no section contains fabricated content. She has consent forms on file for each client who agreed to recording. One client declined recording; those notes she wrote manually.

With NotuDocs, she wrote brief notes during the transitions between sessions: four to six bullet points per client covering what the client reported, her clinical observations, the interventions she used, and the client's response. After the last session, she runs each set of notes through her template, reviews the formatted output, makes small adjustments, and signs.

Neither workflow is better for everyone in absolute terms. A therapist who finds post-session writing genuinely difficult, who conducts long complex sessions, or who is already conducting all sessions via telehealth may find the automation of recording-based notes worth the review burden. A therapist who values control over clinical record content, sees clients for whom recording is not workable, or needs native Spanish support will find template-first documentation a better fit.

What Upheal Does Well

This comparison should be fair. Upheal is a thoughtfully built product, and its differentiators are real.

The progress tracking and outcome measurement features are substantive. If you need to demonstrate treatment effectiveness to insurers or supervisors, or if you want a longitudinal view of how your clients are doing across sessions, those analytics are genuinely useful and go beyond what a documentation-only tool provides.

Upheal's telehealth integration makes the recording workflow smoother for practices already conducting sessions online. If your entire practice is telehealth and your clients are comfortable with recording, the friction of capturing sessions is lower than it would be for in-person work.

The platform has also invested in addressing the hallucination problem specifically. They are aware of it and have iterated on reducing fabricated content in generated notes. The rate of error is lower than it was in earlier versions of the tool and earlier AI documentation tools generally.

Comparison Summary

Who Each Tool Is For

Upheal works well if you:

• Conduct primarily telehealth sessions with a consent workflow already in place for recording
• Value progress tracking and outcome analytics as active parts of your clinical practice
• Want maximum note automation and are disciplined about reviewing AI-generated content before signing
• Have clients who are consistently comfortable with session recording
• Need HIPAA compliance infrastructure and a signed BAA

NotuDocs works well if you:

• Already write some form of post-session notes and want the AI to structure them, not replace them
• Want complete control over note structure, section names, and clinical language
• Prefer not to record sessions, whether for client comfort, client population sensitivity, or personal clinical judgment
• Work in English and Spanish, or document primarily in Spanish
• Want to pay a flat $25/month without navigating tier pricing
• Want a permanent free tier to test with real session data before committing

The Honest Assessment

The core choice here is not about features. It is about what role you want AI to play in your clinical record.

Recording-based tools ask the AI to author notes from session content. That gives you more convenience upfront and requires more scrutiny at review. The documented incident of fabricated clinical content in Upheal's outputs (including childhood abuse history) reflects a structural property of this approach, not a specific software defect. Any tool that generates clinical text from audio transcription faces this risk because the model fills gaps with plausible content.

Template-first tools ask the AI to organize content you wrote. That requires more from you at the note-taking stage. The tradeoff is full authorship: the content of the final note comes from you, and the review step is checking your own words, not auditing someone else's.

For documentation that may face external review (insurance audits, legal proceedings, supervision, licensing inquiries), the provenance of clinical content matters. "The AI wrote it" is not an answer that transfers professional responsibility. Template-first documentation gives you a clearer answer to where the content came from.

Both tools can reduce the documentation burden that follows therapists home. Which one does it in a way that fits how you practice is a judgment only you can make.

---

Related reading:

• AI Hallucination in Clinical Documentation: What Professionals Need to Know
• How to Choose a Clinical Documentation Tool
• SOAP vs DAP vs BIRP: Which Note Format Is Right for Your Practice
• NotuDocs vs Mentalyc: Template-First vs AI-Generated Therapy Notes
• NotuDocs vs Berries: Template-First Notes vs AI Scribe for Therapists