📋 Executive Summary
The Best AI for Therapists in 2026 is not the product that writes the longest note or promises the most automation. It is the system that reduces documentation time without blurring clinical responsibility, and that distinction is becoming legally consequential as Colorado now requires professional review and written informed consent for AI-assisted recording or transcription. I found that Upheal offers the strongest all-in-one fit for many independent practices, Mentalyc provides the deepest therapy-specific note controls, and SimplePractice Note Taker makes the most sense for clinicians already committed to its EHR.
I approached this comparison as a clinical workflow decision rather than a software popularity contest. The seven tools covered here serve different jobs: drafting progress notes, maintaining treatment-plan continuity, operating an EHR, supporting clinical evidence searches, standardising documentation quality, and measuring fidelity to evidence-based practice. A solo counsellor recording six sessions a day does not have the same risk profile as a community behavioural health organisation reviewing thousands of notes, so a single universal winner would be misleading.
The most important lesson from this evaluation is that transcription accuracy is only the first gate. Therapists also need a signed business associate agreement where applicable, precise retention terms, reliable client consent, control over what reaches the designated medical record, and a review process that catches invented interventions or compressed clinical nuance. This guide compares features, current US pricing, integrations, hidden caps, deployment workflows, research evidence, and practical failure modes. It also explains where general chatbots and AI companions should remain outside the clinical pathway, even when they sound empathetic or persuasive.
Best AI for Therapists: The 2026 Verdict
How We Ranked the Best AI for Therapists
Our 2026 verdict is use-case based. Upheal is the best overall choice for a solo or small practice that wants an AI-native EHR rather than another application sitting beside an existing record system. Mentalyc is the best therapy-specific documentation specialist because its higher plans support couples, family, child, group, EMDR, play therapy, psychiatry, treatment planning, progress tracking, supervision, and more than 100 templates. SimplePractice Note Taker is the least disruptive option for an established SimplePractice user because the draft remains inside the platform where scheduling, telehealth, billing, and charting already happen.
AutoNotes is the strongest budget alternative for clinicians who value unlimited note generation and flexible capture modes, although buyers should review its application-level privacy documents rather than relying on the public website policy alone. Heidi is the most capable general clinical scribe in the comparison, particularly for free documentation and evidence support, but its workflow is designed across healthcare rather than specifically around psychotherapy modalities. Eleos and Lyssn occupy a different tier: both are enterprise systems, with Eleos focused on behavioural health documentation, compliance, and revenue-cycle integrity, and Lyssn focused on training, fidelity measurement, and quality improvement.
The ranking gives more weight to clinical fit, governance, and downstream editing than to vendor claims about minutes saved. Evidence from multisite medical settings suggests ambient scribes can reduce documentation time, but the 2026 JAMA study found a modest average effect of 16 fewer documentation minutes and no significant change in outside-hours EHR time. That is valuable, yet it is not proof that a therapist can stop reviewing notes. In psychotherapy, a fluent error can be more dangerous than an obvious transcription failure because it can quietly alter risk assessment, client meaning, or the record of an intervention (Rotenstein et al., 2026).
| Tool | Best Fit | Price Signal | Primary Limitation |
| Upheal | Solo or small practice replacing its EHR | $1/session, capped at $69/month | Migration and feature-release dependency |
| Mentalyc | Therapy-specific notes and treatment continuity | $14.99-$99.99/month annually | Note and session caps by tier |
| SimplePractice | Existing SimplePractice users | $35/clinician/month standard add-on | No external telehealth capture |
| AutoNotes | Budget, unlimited note drafting | $24-$58/month annually | Application privacy terms require review |
| Heidi | General clinical scribing and evidence | Free; $110 Clinician annually billed | Not therapy-specific |
| Eleos | Enterprise documentation integrity | Custom | Procurement and implementation complexity |
| Lyssn | Fidelity, training, quality improvement | Custom | Not a solo note-writing product |
What Therapists Actually Need From Clinical AI
A therapist-facing AI system must do more than convert speech into prose. The minimum viable workflow includes secure capture, speaker separation, clinically relevant summarisation, configurable note formats, treatment-plan continuity, draft review, and controlled export. The stronger products add consent records, template governance, compliance checks, supervision workflows, client-type support, and an audit trail showing who reviewed or changed the generated note. These requirements are related to the broader market for clinical AI tools for doctors.
The first hidden requirement is separation of records. HIPAA gives psychotherapy notes special protection when they are the mental health professional’s separate analysis of a counselling conversation, while ordinary progress notes used for treatment, billing, or operations belong to the designated record set under different access rules. An AI transcript, a generated progress note, a private process note, and a client-facing summary should therefore not be treated as interchangeable objects. A practice needs a written map showing which artefact is created, where it is stored, who can access it, and when it is deleted (HHS, 2026).
The second requirement is controllable specificity. Payer-ready notes often demand diagnosis linkage, functional impact, interventions, response, progress, and medical necessity. Clinical integrity, however, may require leaving out intimate details that are not necessary for the treatment record. The best tools let the clinician constrain verbosity and map the note to an approved template. They do not reward maximal capture. This is where therapy-specific products generally outperform generic meeting transcription, even though practices can still learn useful operational lessons from AI meeting-notes implementation patterns.
The third requirement is graceful failure. A session may include silence, crying, overlapping speakers, a child moving around the room, poor internet, code-switching, or a phone call on speaker. A robust system should flag low-confidence segments or missing audio instead of smoothing the gap into a plausible story. During evaluation, ask vendors how they represent uncertainty, whether failed recordings remain recoverable, and whether drafts can be regenerated without retaining unnecessary source audio.
Seven Tools Compared by Features and Fit
The products fall into three architectural groups. Upheal and SimplePractice combine AI with practice-management or EHR functions. Mentalyc, AutoNotes, and Heidi act primarily as documentation layers that must exchange information with another system. Eleos and Lyssn are organisational platforms sold through custom deployment. That architecture affects cost, identity management, auditability, support, and the amount of copy-and-paste work left for clinicians.
A feature tick is not automatically evidence of a safe workflow. For example, an integration can mean a direct API, a browser extension, a structured export, or simply a guided copy process. Similarly, a promise to delete audio does not reveal how long transcripts, generated notes, telemetry, backups, model inputs, or de-identified datasets persist. The feature matrix below records only functions documented on current vendor pages. Where an API, named EHR connector, or retention period is not publicly itemised, the table says so rather than filling the gap with an assumption.
The distinction also appears in adjacent medical-record and imaging workflows.
For therapists, the practical question is not whether a tool can generate a SOAP or DAP note. Most can. The question is whether it understands the client type, modality, treatment objective, payer requirement, and consent boundary associated with that note, then preserves enough friction for a licensed human to remain accountable.
| Tool | Capture and Clinical Features | Documented Integrations | Governance Notes |
| Upheal | Audio, video, text, uploads; notes, forms, telehealth, scheduling, messaging, billing support, compliance | SimplePractice and TherapyNotes migration; calendar sync; no public general API | BAA; default audio deletion; opt-in improvement data has separate retention |
| Mentalyc | In-person, browser telehealth, uploads, dictation; therapy templates, alliance, treatment, progress, supervision | SimplePractice, TherapyNotes, Jane App, ICANotes, TheraNest | BAA; SOC 2 Type II; vendor states audio deletion and no model training on clinical data |
| SimplePractice | In-person mobile and native telehealth; couples and groups; prior-note style adaptation | Native SimplePractice EHR only for capture workflow | HIPAA and HITRUST claims; clinician review required; client opt-out supported |
| AutoNotes | Type, dictate, upload, live capture; broad note formats; continuity and custom templates on First Class | Other-platform capture; named direct EHR/API list not publicly itemised | Signed BAA; application privacy documents separate from website policy |
| Heidi | Unlimited documentation, templates, evidence, patient context and team functions | EHR integration options, details vary by plan and region | General clinical product; public therapy-specific validation not found |
| Eleos | Documentation, insights, compliance, revenue cycle | EHR-agnostic browser extension; SmartCare deployment documented | Enterprise controls; public self-serve pricing and full API catalogue unavailable |
| Lyssn | Fidelity measurement, training, quality improvement | Custom organisational deployment; public API list unavailable | Research-led enterprise platform; no solo plan or public price |
Upheal: Best Overall for an AI-Native Practice
Upheal earns the overall recommendation because it addresses the entire solo-practice workflow instead of treating documentation as an isolated task. Its current product combines AI-generated therapy notes, scheduling, forms, telehealth, messaging, client billing, compliance checking, client portals, and migration support. Sessions can be captured through video, audio, text, dictation, or uploaded recordings. The pricing model is unusual: Premium is charged at $1 per completed session and capped at $69 per month, while a free plan provides AI note creation. For a variable caseload, that makes costs track actual clinical activity without exceeding the monthly ceiling (Upheal, 2026a).
The strongest operational advantage is continuity across the record. A therapist can move from session capture to a structured note, treatment-plan context, client messaging, and scheduling without repeatedly transferring protected information between vendors. Upheal also documents migration from SimplePractice and TherapyNotes, calendar visibility, mobile recording, and a business associate agreement. Its security material says recordings are deleted by default, while opt-in data contributed for product improvement can have longer retention. That distinction is critical: deletion claims should always be read together with consent settings and derived-data terms, not as a single blanket promise (Upheal, 2025).
Upheal’s limitation is that becoming an all-in-one system creates migration and lock-in risk. A practice must validate every imported demographic field, note, consent, balance, and appointment before making the old EHR read-only. Insurance claims submission was still described by the vendor as coming in summer 2026, so a billing-heavy practice should confirm that feature is live in its account before replacing an established claims workflow. Phone capture is also documented as a workaround rather than the preferred method. These are not minor details when the system is positioned as the practice’s operating layer.
Juraj Chrappa, Upheal’s Co-Founder and CEO, framed the product philosophy plainly in April 2026: “Therapists should be doing the clinical work. Software should be doing the rest.” The line is useful because it sets the correct boundary, but the clinician still owns every clinical statement. Upheal is the best fit when a practice is willing to redesign operations around one platform. It is less compelling when the existing EHR is contractually entrenched, heavily customised, or linked to complex payer and reporting infrastructure.
Mentalyc: Best for Therapy-Specific Clinical Depth
Mentalyc is the most specialised documentation product in this comparison. All plans support in-person capture, browser-based telehealth capture, uploaded audio, dictated recaps, text descriptions, and cross-device access. The plan boundaries, however, matter. Mini and Basic cap sessions at 75 minutes and focus on individual adult therapy. Pro extends the limit to 130 minutes and adds couples, children and teenagers, family therapy, EMDR, play therapy, psychiatry, a full therapeutic-alliance assistant, treatment planning, progress tracking, supervision, automatic CPT support, and more than 100 templates. Group notes appear only on Super and Team (Mentalyc, 2026).
That structure makes Mentalyc attractive to clinicians whose documentation must reflect modality and relationship context rather than simply summarise a conversation. Its published integration list includes SimplePractice, TherapyNotes, Jane App, ICANotes, and TheraNest. The company states that it signs BAAs, is HIPAA compliant and SOC 2 Type II audited, deletes audio after processing, and does not use clinical data to train its models. Those are stronger public commitments than a generic statement that an application is secure, although every group practice should still obtain the executed contract and confirm regional data handling.
The hidden cost is note volume. Monthly plans provide 40 notes on Mini, 100 on Basic, 160 on Pro, and 330 on Super. A therapist completing 25 sessions each week can outgrow Basic quickly, and a high-volume group can exhaust Super even before accounting for intakes, group-member notes, or regenerated drafts. The 14-day trial includes the full Pro feature set but is capped at 15 notes. This is enough to test several session types, but not enough to establish long-term error rates across a large caseload.
Mentalyc is therefore the best fit for a therapist who wants detailed clinical structure while retaining a separate EHR. It is not automatically the best economic choice for every clinician. Before subscribing, calculate notes per month rather than sessions per week, include extra notes generated for group members, and test whether the EHR transfer is direct enough for the practice’s risk tolerance. A tool can be excellent at drafting and still create an unacceptable final-mile workflow if staff must manually reconcile client names, dates, diagnosis codes, or signatures.
SimplePractice: Best Integrated EHR Choice
SimplePractice Note Taker is the logical choice for clinicians already using SimplePractice because it avoids a separate clinical-data repository and keeps the draft close to the chart. The add-on supports in-person sessions through the mobile app and SimplePractice Telehealth sessions through the web workflow. It can prepare notes for individual, couples, and group appointments, reference a clinician’s previous finalised note to adapt style, and generate drafts in a broad set of supported languages. The product is available across Starter, Essential, and Plus plans after a 30-day trial.
The standard add-on price is $35 per clinician each month. The public pricing page displayed a 50% promotion at $17.50 when checked on 27 July 2026, but the support documentation still identifies $35 as the regular price. Base subscription pricing and transaction fees remain separate. For a ten-clinician practice, regular Note Taker pricing adds $350 monthly before the underlying EHR plans, claims charges, payment processing, or premium telehealth options. That is why embedded software should still be budgeted as a separate product, even when procurement is simpler (SimplePractice, 2026a).
There are important capture constraints. SimplePractice documents a maximum of four hours for its Telehealth workflow and two and a half hours for in-person recording. It does not support sessions held on external telehealth platforms, so a clinician using Zoom, Doxy.me, or another service cannot assume the same capture path. The vendor also warns that generated notes may contain errors, bias, or missing details and requires clinician review. That warning should be treated as an operational control, not boilerplate.
Care Aide, announced in June 2026 after input from more than 1,000 mental health practitioners, expands the direction of travel with session preparation, treatment planning, intake review, client summaries, and documentation support. At the time of review, several Care Aide functions were described as rolling out through an extended trial for selected Note Taker users. Practices should distinguish generally available Note Taker capabilities from preview or phased-release features. SimplePractice is the best low-disruption option, but it does not eliminate the need for consent, quality sampling, or an exit plan for AI-generated text stored in the legal record.
AutoNotes and Heidi: Budget and General Scribe Options
AutoNotes and Heidi serve different buyers, yet both can be rational alternatives to therapy-specific premium plans. AutoNotes Economy costs $29 monthly or $24 monthly when billed annually and includes unlimited AI notes, typed or dictated inputs, standard formats, uploaded recordings, treatment plans, summaries, and secure storage. First Class costs $69 monthly or $58 monthly annually and adds live in-person and virtual recording, capture from other platforms, note-to-note continuity, and custom templates. Enterprise pricing is custom and adds team controls, analytics, workflow configuration, and onboarding (AutoNotes, 2026).
AutoNotes supports a wide range of formats, including SOAP, DAP, BIRP, GIRP, PIE, progress notes, intakes, discharges, treatment plans, couples, family, group, and EMDR documentation. Its trust centre states that the secured application operates under a HIPAA-aligned programme, supports signed BAAs, and aligns with PIPEDA and PHIPA principles. A useful due-diligence detail is that the public website privacy policy does not govern PHI entered into the secured application. Buyers should request and review the Application Privacy Policy, application terms, EULA, and BAA as a matched set. Scheduling and native AI telehealth were labelled coming soon, so they should not be counted as current features.
Heidi’s Free plan offers unlimited AI documentation and clinical evidence functions, but advanced actions are limited to 10 per month. Clinician is shown at $110 per user per month when billed yearly, and Practice at $180 per user per month billed annually. Paid tiers add advanced templates, personalisation, live evidence suggestions, patient-context answers, team templates, sharing, onboarding, and EHR integration options. The product is unusually capable for a free general medical scribe, but therapists may need more manual template work to prevent medical-style notes from becoming over-detailed or clinically tone-deaf (Heidi Health, 2026a).
Dr Thomas Kelly, Heidi’s Co-Founder and CEO, summarised the company’s model strategy in June 2026: “You don’t need frontier scale to reach frontier quality.” Heidi reported parity of 49.9% against Sonnet 4.6 in a blinded side-by-side evaluation, supported by HealthBench Pro and an internal medical safety set. That is a meaningful vendor benchmark, not an independent psychotherapy trial. It measures answer preference and medical safety rather than the fidelity of a therapy note, the handling of therapeutic alliance, or the effect of generated documentation on client outcomes.
Eleos and Lyssn: Enterprise Quality and Fidelity
Eleos is built for community behavioural health, substance-use treatment, home-based care, and other organisations where documentation quality directly affects compliance and revenue. Its platform covers documentation, a Clinical Insights Agent, compliance, and revenue-cycle workflows. The company describes an EHR-agnostic browser extension that can work without a custom API build, while its 2026 California partnership provides access through Streamline’s SmartCare environment. Public pricing is not available, which means total cost depends on scope, implementation, user volume, integration, and contract terms.
The enterprise value is not simply faster notes. Eleos is designed to scan documentation at scale, identify missing or inconsistent elements before billing, and surface clinical interventions. In March 2026, Tracie Del Torto, LCSW and Chief Clinical Officer at Lighthouse Behavioral Wellness Centers, explained a responsible implementation principle: “We were not bringing this tool on to expect that clinicians were seeing more people and billing more.” That governance stance matters. Reinvesting saved time in supervision, recovery, access, or better client preparation is different from converting every saved minute into a larger caseload.
Lyssn is even less like a conventional AI note taker. It applies research-backed models to training, quality improvement, fidelity monitoring, and evidence-based practice across health and human services. The company reports more than 4.5 million analysed sessions, over 65 peer-reviewed publications, and more than 100 organisations, but its public site does not provide a self-serve price, complete API catalogue, or solo-clinician plan. Those omissions are appropriate to flag because implementation will usually involve programme design, data governance, workforce training, and contract-level integration.
The choice between Eleos and Lyssn depends on the organisational problem. Eleos is stronger when the immediate target is documentation integrity, compliance, and operational follow-through. Lyssn is stronger when leaders need to measure whether evidence-based skills are actually being demonstrated, then use those measurements for coaching and quality improvement. Neither should be purchased as a simple substitute for a $29 monthly note generator. Their value and risk are collective, which means a formal governance group, workforce consultation, validation dataset, and phased deployment are prerequisites rather than optional extras.
Pricing Matrix, Hidden Caps, and Total Cost
Headline prices conceal three different billing models: per-session, per-user, and usage-capped subscriptions. Upheal’s $1 per completed session with a $69 ceiling is attractive for fluctuating caseloads. Mentalyc’s annual prices look lower, but note limits and client-type restrictions determine the real tier. SimplePractice’s embedded add-on reduces integration work but multiplies by clinician count. AutoNotes offers unlimited notes, while Heidi keeps basic documentation free and charges substantially more for advanced individual and practice functions. Eleos and Lyssn require custom quotes.
The full cost calculation should include subscription fees, EHR fees, telehealth, claims, payment processing, migration, consent administration, staff training, security review, support time, and human editing. Review labour is the most frequently ignored line. If a tool saves eight minutes in drafting but creates five minutes of checking, formatting, and transfer, its net benefit is three minutes. If a flawed note triggers a payer query or creates a legal inconsistency, the apparent saving can become negative. The correct metric is reviewed, signed, and correctly filed notes per clinician hour, not generated drafts per minute.
A second trap is plan mismatch. A couples therapist on Mentalyc cannot use Mini or Basic for couples workflows. A SimplePractice clinician using an external telehealth platform cannot use the standard Note Taker capture path for that session. An AutoNotes Economy user who needs live capture and continuity requires First Class. A Heidi Free user who relies on advanced actions will hit a 10-action monthly ceiling. A group selecting custom enterprise software may also face implementation, minimum volume, or integration charges that are not publicly itemised. Those terms must be obtained in writing.
Pricing was verified against official vendor pages on 27 July 2026. Promotional pricing can change without notice, and dynamic pages may display region-specific tax, currency, or billing-cycle differences. The matrix therefore separates standard published pricing from temporary offers and labels unavailable information as custom or not publicly confirmed.
| Tool and Plan | Current Price | Included Volume or Cap | Hidden Limit or Cost Driver |
| Upheal Free | $0 | Unlimited AI notes as published | Confirm which EHR and operational features require Premium |
| Upheal Premium | $1/completed session; $69/month cap | No price increase beyond cap | Migration, claims availability, and payment fees need validation |
| Mentalyc Mini | $19.99 monthly; $14.99 annually billed | 40 notes; 75-minute sessions | Individual adults only |
| Mentalyc Basic | $39.99 monthly; $29.99 annually billed | 100 notes; 75-minute sessions | Limited alliance function; no couples, family, or children |
| Mentalyc Pro | $69.99 monthly; $59.99 annually billed | 160 notes; 130-minute sessions | No group notes |
| Mentalyc Super | $119.99 monthly; $99.99 annually billed | 330 notes; 130-minute sessions | Volume can still be exceeded by groups or regenerated notes |
| SimplePractice Note Taker | $35/clinician/month standard | 30-day trial; promotion displayed at $17.50 | Requires base EHR plan; multiplied by clinician count |
| AutoNotes Economy | $29 monthly; $24 annually billed | Unlimited AI notes | No live recording or note continuity |
| AutoNotes First Class | $69 monthly; $58 annually billed | Unlimited AI notes | Still requires separate EHR workflow |
| Heidi Free | $0 | Unlimited core documentation; 10 advanced actions/month | Advanced templates, context, and actions are capped |
| Heidi Clinician | $110/user/month billed yearly | Advanced individual features | Annual commitment shown; therapy-specific fit may require configuration |
| Heidi Practice | $180/user/month billed annually | Team and practice functions | Integration add-ons and enterprise terms may apply |
| Eleos and Lyssn | Custom | Contract-defined | Implementation, integration, minimum volume, and support not public |
Privacy, Consent, and the New Regulatory Boundary
Clinical AI is moving from broad privacy guidance to explicit psychotherapy rules. Colorado HB26-1195 permits administrative and supplementary AI support when the regulated professional remains responsible, but it restricts direct therapeutic communication, unreviewed treatment recommendations, and emotion or mental-state detection. It also requires advance written disclosure and informed consent when an AI system records or transcribes a session. This is a useful operational model even for practices outside Colorado: define the purpose, obtain documented consent, preserve a non-AI option, and require human approval before any output influences care (Colorado General Assembly, 2026).
The legislation fits a wider pattern documented in state AI bills shaping care.
HIPAA compliance is necessary but not sufficient. A signed BAA does not answer whether a vendor stores audio, retains transcripts, creates embeddings, uses de-identified data, trains models, keeps backups, logs prompts, permits subcontractor access, or supports deletion. Upheal’s public material, for example, distinguishes default deletion from longer retention when clients opt into product improvement. Mentalyc states that audio is deleted after processing and clinical data is not used for model training. AutoNotes separates website terms from application-level PHI documents. Each structure can be responsible, but the practice must know which contract controls which data.
Psychotherapy information also deserves more granular treatment than ordinary medical dictation. HHS says psychotherapy notes maintained separately from the medical record receive special protection, while progress notes and other chart information may fall under the designated record set. An AI implementation should therefore prohibit automatic merging of full transcripts or private process reflections into the clinical record. The same concern applies to evidence and search products. Our coverage of Perplexity Health privacy commitments.
Public policy is also drawing a bright line around simulated relationships. Governor Kathy Hochul’s January 2026 proposal sought to restrict integrated AI chatbot features for children and strengthen youth mental-health safeguards. Her direct warning was blunt: “Let’s disable A-I chatbots that can cause serious mental health problems.” For a therapist, the practical implication is to keep client-facing companions outside the care pathway unless a product has an approved clinical purpose, clear crisis handling, consent, supervision, and evidence appropriate to the population.
A Safe Implementation Workflow
A safe rollout begins with a written problem statement. Decide whether the objective is faster progress notes, treatment-plan continuity, supervision, compliance review, or quality measurement. Do not begin with a vendor demo and allow its feature list to define the problem. The practice should identify the smallest data set needed, the sessions that are in scope, the clinicians authorised to participate, and the metrics that will determine whether the pilot succeeds.
Next, complete contractual and privacy review. Obtain the BAA where applicable, data-processing terms, subprocessors, security reports, incident-notification commitments, retention schedule, deletion process, training-data policy, regional hosting details, and export procedure. Record which system holds audio, transcript, draft, final note, and audit log. Confirm whether the clinician can delete an errored draft without deleting the legal record, and whether termination returns or destroys protected information. A trust badge is not a substitute for this document set.
Then configure the clinical workflow. Use one approved template for each initial test, map required fields to payer and organisational standards, turn off unnecessary narrative detail, and create a client-consent script. The therapist should explain what the AI does, what it does not do, what information is captured, and how the client can refuse or revoke consent without losing care. Test in-person, telehealth, accented speech, overlapping speakers, long pauses, risk language, couples sessions, and low-bandwidth conditions. These tests expose different failure modes.
Finally, run a controlled pilot and measure the signed-note outcome. Compare pre-pilot and pilot time from session end to final signature, number of edits, missing required elements, unsupported additions, client opt-out rate, failed captures, help-desk incidents, and clinician satisfaction. Review a random sample every week. Do not allow the generated draft to auto-file or auto-send. The control matrix below translates the workflow into explicit gates, owners, and evidence.
| Gate | Required Evidence | Owner | Stop Condition |
| Purpose | One documented use case and excluded uses | Clinical lead | Vendor scope expands beyond approved purpose |
| Contract | BAA, privacy terms, subprocessors, retention, deletion, export | Privacy and legal | Data lifecycle remains unclear |
| Consent | Written script, refusal path, revocation process | Clinical operations | Client cannot receive equivalent care without capture |
| Configuration | Approved templates, minimum necessary fields, role controls | Documentation committee | Output includes unnecessary sensitive detail |
| Pilot | Test sessions across modalities and difficult capture conditions | Pilot clinicians | High failure or unsupported-statement rate |
| Review | Human sign-off and weekly sample audit | Licensed clinician and QA | Auto-filing or auto-sending bypasses review |
| Scale | Measured net time, edit rate, opt-out rate, incidents, export test | Executive sponsor | Benefits do not exceed operational and privacy burden |
Bottlenecks, Failure Modes, and What AI Cannot Do
The most consequential failure in therapy documentation is clinical compression. A model may turn a client’s ambivalence, humour, silence, or contradictory statements into a clean narrative of progress. The note reads professionally, but the uncertainty that guided the therapist’s judgement disappears. This can make treatment look more linear than it was and may distort risk, readiness, or alliance. Reviewers should compare drafts against a concise clinician memory or structured checklist, not only ask whether the grammar sounds correct.
A second failure is continuity contamination. Tools that learn from the previous finalised note can improve style and save time, but an old mistake may be carried forward as context. A wrong pronoun, copied intervention, outdated risk statement, or inaccurate diagnosis linkage can become a recurring pattern. Practices should periodically start from a clean template, audit repeated phrases, and require explicit revalidation of high-risk fields such as suicidality, medication changes, safety plans, and mandated-reporting facts.
A third bottleneck is capture inequality. Ambient systems work best in quiet, well-connected environments with clear speakers. Behavioural health often occurs in homes, schools, community sites, or rooms with children and family members. Accents, code-switching, non-English sessions, whispered disclosures, and emotional speech can reduce reliability. Research on ambient scribes has also identified concerns about note length, editing, and non-English performance, even when clinicians report lower workload and better engagement (Shah et al., 2025). A product’s supported-language list does not prove equivalent clinical accuracy across languages.
AI also cannot assume licensure, establish therapeutic accountability, or carry the relationship. Consumer companions may sound supportive, but Replika’s therapy limitations.
Neither is Pi’s reflection-versus-therapy boundary.
The boundary is straightforward: use AI to support administrative cognition, not to outsource the clinician’s duty to notice, interpret, decide, and remain present.
Evidence Workflows Beyond Note Drafting
Therapists increasingly use AI for literature discovery, psychoeducation preparation, policy scanning, and treatment-plan research. These uses can produce value without recording a session, but they require a different verification discipline. A citation-generating system should be treated as a search and synthesis layer, not as the source itself. The clinician or editorial reviewer must open the underlying paper, confirm the population and intervention, distinguish preprint from peer review, and check whether a finding applies to the client in question.
Heidi’s Evidence function illustrates the convergence of documentation and clinical retrieval. Its 2026 model post reported more than 3.5 million questions answered since launch and more than 300,000 weekly, alongside a blinded preference benchmark. Those figures show scale, but they do not remove the need to inspect the cited evidence or identify contradictory guidance. For psychotherapy, outcome evidence may depend heavily on diagnosis, age, setting, cultural context, therapist training, fidelity, and therapeutic alliance. A compact answer can hide those moderators.
A practical evidence workflow begins with a structured question, records the query date, saves the source citations, and grades the evidence. Our guide to Consensus AI evidence review.
The practice should also separate clinical decision support from client communication. A therapist may use a verified evidence summary to prepare for supervision, then translate the finding into language suited to the client. Sending an unreviewed AI answer directly to a client can overstate certainty, introduce unfamiliar terminology, or imply that a general study is a personalised recommendation. Evidence retrieval is most useful when it strengthens the therapist’s reasoning without pretending to replace it.
Buying Framework for Solo, Group, and Enterprise Practices
A solo therapist should prioritise low setup burden, a predictable ceiling, a signed BAA, clear consent, exportability, and a workflow that does not double-enter data. Upheal is the best all-in-one candidate when the clinician is prepared to move core operations. Mentalyc is the better choice when therapy-specific documentation depth matters more than replacing the EHR. AutoNotes is compelling when unlimited draft volume and budget are primary, while Heidi Free is suitable for testing general documentation and evidence support with no subscription cost.
A group practice should model cost by clinician, note volume, session type, and review labour. SimplePractice Note Taker is attractive when the group already uses the EHR and standardised telehealth. Mentalyc Super or Team better supports group, child, family, and supervision workflows, but note caps and contract terms need close inspection. Groups should also require role-based access, central template governance, audit logs, offboarding, incident response, and a mechanism for clients to refuse capture without disrupting scheduling or care.
An enterprise behavioural health organisation should begin with governance and measurable operational pain, not with a generic AI mandate. Eleos is the leading fit for documentation integrity and compliance across community behavioural health. Lyssn is the specialist choice for fidelity, training, and quality improvement. Both require custom procurement, integration design, workforce engagement, and validation against the organisation’s actual programmes. A small pilot should include clinicians who are sceptical, multilingual workflows, complex cases, and the most difficult service settings rather than only enthusiastic users in quiet offices.
The final decision should be conditional. Choose Upheal for an integrated solo-practice operating system, Mentalyc for specialised therapy documentation, SimplePractice for embedded EHR convenience, AutoNotes for low-cost unlimited drafting, Heidi for broad clinical scribing and evidence, Eleos for enterprise documentation integrity, and Lyssn for programme fidelity. Reject any vendor that will not document retention, deletion, model-training use, subcontractors, consent support, and data export. A sophisticated interface cannot compensate for an unverifiable data lifecycle.
Our Research Methodology
We treated this as a product comparison and governance review rather than a transcription-speed contest. Pricing, plan caps, supported session types, capture limits, integrations, security claims, and release status were checked against official vendor pricing, support, trust-centre, press, and product pages available on 27 July 2026. Where a vendor did not publish pricing, API details, named integrations, or retention terms, we labelled the information custom or not publicly confirmed instead of estimating it.
The evidence review used current US government guidance, Colorado HB26-1195, official 2026 product announcements, and peer-reviewed studies in JAMA and JAMA Network Open. We separated vendor benchmarks from independent research. The JAMA evidence covers ambient clinical scribes across medical settings and is directionally useful for documentation burden, but it is not a direct randomised comparison of the seven therapy products in this article. The Heidi benchmark was treated as a vendor evaluation of a clinical model, not as proof of psychotherapy-note quality.
The editorial scoring considered therapy-specific coverage, capture options, note formats, treatment continuity, EHR fit, consent support, public security documentation, pricing transparency, usage limits, workflow friction, and documented failure conditions. We did not input real patient information or claim a live clinical trial. Hands-on observations in this article refer to reproducible workflow inspection of documented interfaces and processes, not treatment outcomes. The live Perplexity AI Magazine sitemap was blocked by an anti-bot verification page during research, so internal links were selected from live, indexed site articles and checked for topical relevance rather than fabricated.
This article was researched and drafted with AI assistance and reviewed by the Sami Ullah Khan editorial desk at Perplexity AI Magazine. All data, citations, pricing figures, and named quotes have been independently verified against primary sources before publication.
Conclusion
The best AI for therapists is now a question of clinical architecture, not novelty. Upheal leads for practices ready to adopt an AI-native EHR, Mentalyc offers the richest therapy-specific documentation controls, and SimplePractice provides the cleanest route for clinicians already inside its ecosystem. AutoNotes and Heidi broaden the price range, while Eleos and Lyssn address enterprise problems that a solo note generator cannot solve.
The evidence supports cautious optimism. Ambient scribes can reduce documentation time and perceived burden, but average gains are modest, outside-hours work does not always fall, and fluent drafts still require disciplined review. Therapy adds further stakes because meaning is relational, sensitive details may not belong in the billing record, and the difference between a progress note and separate psychotherapy notes carries legal significance.
Regulation is also becoming more specific. Colorado’s 2026 law gives practices a practical boundary: administrative support may be appropriate, while direct therapeutic communication, unreviewed treatment recommendations, and undisclosed recording are not. Other jurisdictions will continue to develop their own rules, and vendor capabilities will change faster than published research. The safest decision is therefore reversible and measurable. Choose the tool that fits the narrowest defined need, document the data lifecycle, preserve client choice, audit signed notes, and keep the therapist visibly responsible for every clinical conclusion.
Frequently Asked Questions
What Is the Best AI for Therapists in Private Practice?
Upheal is the strongest overall option for a solo or small private practice that wants AI notes, scheduling, telehealth, messaging, billing support, forms, and an EHR in one system. Mentalyc is better when therapy-specific documentation depth matters more than replacing the existing EHR. The right choice depends on caseload, session types, consent requirements, and the cost of final review and transfer.
Is There a Free HIPAA-Compliant AI Note Taker for Therapists?
Upheal offers a free AI-note plan, and Heidi offers free unlimited AI documentation with a 10-action monthly cap on advanced functions. HIPAA claims should still be verified through the applicable BAA, privacy terms, security documentation, and data-retention schedule. A free price does not remove the therapist’s obligations to obtain consent, minimise protected information, review every draft, and store the final record correctly.
Can Therapists Use ChatGPT for Clinical Notes?
A therapist should not place identifiable client information into a general consumer chatbot without an appropriate healthcare agreement, security configuration, authorised workflow, and legal review. Even with an enterprise arrangement, a general model may lack therapy-specific templates, consent records, clinical safeguards, and EHR controls. Purpose-built products are usually easier to govern, but they still require a BAA where applicable and human review.
Does HIPAA Allow AI to Record Therapy Sessions?
HIPAA does not create a universal permission to record sessions. The practice must have a lawful purpose, appropriate vendor agreements, safeguards, and policies, while state recording and psychotherapy laws may impose additional requirements. Colorado’s 2026 law specifically requires advance written disclosure and informed consent for AI recording or transcription. Practices should provide a non-recorded option and document consent or revocation.
Which AI Tool Is Best for Couples or Family Therapy Notes?
Mentalyc Pro supports couples, family, children and teenagers, while group workflows require Super or Team. SimplePractice Note Taker supports couples and group appointments inside its EHR. AutoNotes also lists couples, family, group, and EMDR formats. The evaluation should include speaker attribution, consent from all relevant participants, session-length caps, and whether the generated note separates member-specific information appropriately.
How Much Do AI Note Takers for Therapists Cost?
Current self-serve pricing ranges from free plans to more than $100 per user monthly. Upheal Premium is $1 per completed session capped at $69 monthly. Mentalyc annual plans range from $14.99 to $99.99 monthly, with note caps. AutoNotes ranges from $24 to $58 monthly when billed annually. SimplePractice Note Taker is normally $35 per clinician monthly, excluding the EHR subscription.
Can AI Replace a Therapist?
No. The products compared here are documentation, evidence, quality, or operational tools. They do not hold licensure, maintain therapeutic accountability, or replace clinical judgement and relationship. Colorado now restricts AI from directly engaging in therapeutic communication or generating unreviewed treatment recommendations in covered psychotherapy settings. Consumer companions may provide reflection or wellness content, but they should not be represented as equivalent to professional psychotherapy.
What Should a Therapist Check Before Buying an AI Tool?
Check the BAA, retention and deletion rules, model-training policy, subprocessors, security reports, consent workflow, supported session types, language performance, EHR transfer method, note caps, capture limits, export procedure, and incident response. Pilot the product with difficult real-world conditions using non-identifiable test material, then measure edits, failed captures, signed-note time, and unsupported clinical statements before wider deployment.
References
AutoNotes. (2026). Pricing.
Colorado General Assembly. (2026). HB26-1195: Psychotherapy artificial intelligence restrictions.
Heidi Health. (2026a). Pricing.
Kelly, T. (2026, June 25). Fine-tuning a clinical AI model to frontier parity. Heidi Health.
Mentalyc. (2026). Pricing.
Olson, K. D., et al. (2025). Use of ambient AI scribes to reduce administrative burden and professional burnout. JAMA Network Open.
Rotenstein, L. S., et al. (2026). Changes in clinician time expenditure and visit quantity following adoption of ambient AI scribes. JAMA.
SimplePractice. (2026, June 17). SimplePractice launches AI-powered Care Aide.
U.S. Department of Health and Human Services. (2026). Information related to mental and behavioral health.
