The phrase "AI doctor" describes several different products: symptom checkers, patient-facing preparation tools, clinician evidence services and systems participating in care delivery. Nolla Health's October 2026 initial-acne-prescription announcement adds another important distinction: recommending that a patient discuss treatment is not the same as initiating treatment within a permitted service.
Originally published on 21 February 2026, this guide has been updated on 5 October 2026 to retain the broader comparison and add Nolla. Product descriptions are attributed to the companies' current public material, and regulatory scope is checked against Utah's own account where relevant. This is a desk-researched comparison, not a hands-on trial or a clinical accuracy ranking.
The article is published by iatroX and includes iatroX among the reference and learning tools discussed. Its purpose is to identify what each service is trying to do, where it might fit and which questions remain unresolved, rather than place unlike products in a single league table.
First: what kind of "AI doctor" are we talking about?
1) Patient-facing "organise and orient" tools
These tools help people organise records and results, construct a health timeline, prepare appointment questions and identify possible issues for discussion. The useful output may be a clearer account of what happened, rather than a diagnosis. Medome's positioning, checked on 5 October 2026, includes this combination of health-information organisation and appointment preparation. Medome's product description.
A patient with reports from several providers may benefit from arranging the chronology before an appointment. Whether AI improves that process needs to be evaluated, but the job itself is distinct from independently deciding what treatment the patient should receive.
2) Symptom checkers and care navigation
This category gathers information about symptoms and suggests possible explanations or a next step in obtaining care. Ada and Buoy describe structured or conversational assessment experiences; K Health's broader proposition also connects AI-supported assessment with care delivery. Ada, Buoy and K Health, checked on 5 October 2026.
Self-care information, a routine appointment suggestion and an urgent-care recommendation are different outputs. Evaluation must examine what users understand and do next, not simply whether the suggested condition appears somewhere in a list.
3) AI-first care workflows with human clinicians in the loop
Care workflows may include intake, follow-up questions, note generation, draft plans and limited or supervised prescribing. Doctronic, Akido and Nolla belong in this discussion because their public descriptions connect AI with a care process, although the permitted action and supervision differ. Utah's pilot register and Akido's Scope AI account, checked on 5 October 2026.
The phrase "human in the loop" is only a starting point. A doctor approving treatment before it reaches the patient performs a different safety function from a doctor auditing a sample afterwards. A service's actual operating phase matters as much as its long-term design.
4) Clinician-facing evidence synthesis ("second brain" tools)
Here the user is a clinician seeking evidence to inform an assessment or management decision. OpenEvidence's clinician-facing experience provides citation-backed answers, while iatroX combines source-linked reference with professional-learning tools. These functions do not by themselves constitute patient treatment. OpenEvidence's educational material and iatroX's methodology, checked on 5 October 2026.
This is a description of the experiences being compared, not a claim that any company's entire product range is permanently confined to one category.
5) Diagnostic AI in constrained tasks (imaging, pathology, etc.)
This guide does not attempt to review radiology or pathology AI. Their importance here is methodological: performance on a specified image-recognition task is not equivalent to performance across a complete consultation. A result from a constrained modality cannot simply be transferred to a general medical chatbot or prescribing service.
NollaMD's visual-medicine research offers a current example of that boundary. Its developer report concerns a defined specialist-model experiment, not validation of every step in Nolla's patient-care pathway. TrainLoop's NollaMD report, 9 April 2026.
Why this category is exploding now
The original February article identified three changes: patients using AI before appointments, clinicians using it for notes and reasoning support, and people receiving direct AI summaries instead of navigating conventional search results. Those remain useful questions about adoption, but this guide does not assign them a prevalence or assume that all patient groups behave alike.
The current product descriptions demonstrate that companies are addressing those different entry points. Medome emphasises the patient's preparation, Akido the assembly of clinical context, and OpenEvidence the clinician's evidence question. Nolla now adds a condition-specific route from assessment towards treatment. The resulting market is broader than a contest between conversational interfaces.
Our interpretation is that reducing friction is valuable only when the process preserves the information and escalation needed for the task. A rapid answer can save effort, but it can also make it easier to act before understanding the answer's scope. Both possibilities deserve evaluation.
Tool-by-tool review
The following sections preserve the distinction between a vendor's stated design, a plausible use case and demonstrated outcomes. Product claims are presented as checked on 5 October 2026 unless another publication date is specified.
Medome
What it appears to be optimising
Medome presents a consumer-facing service that brings together health records and supports understanding before, during and after care. Its current website describes organising records and results, preparing for appointments and summarising consultations. It also identifies access to an affiliated but independent medical group. These are published service descriptions, not evidence that every user's history is complete or every concern will be detected. Medome.
What is genuinely useful here
The potential value is reducing informational disorder. A fictional patient might have a scan report in one portal, a medication list elsewhere and a personal account of symptoms that changed between appointments. Organising those sources could help them explain the chronology and ask a focused question.
The original examples remain relevant: timeline building, arrangement of laboratory results and scans, medication and symptom organisation, appointment preparation and revisiting assumptions when concerns persist. A fragmented account should not automatically be blamed on a patient being a "poor historian"; the information may itself be scattered.
My view (balanced)
For someone primarily trying to prepare a coherent account for a consultation, Medome's organising role is more directly relevant than a claim that AI can replace a doctor. The key checks are whether the summary preserves uncertainty, distinguishes original records from generated interpretation and can be corrected.
Privacy and overreliance still need attention. A reassuring interface or a company privacy statement does not remove the need to understand what will be stored and shared. This is a general evaluation requirement, not an allegation that Medome mishandles data.
Doctronic
Why Doctronic matters more than a typical symptom checker
Doctronic's AI-doctor positioning combines conversational assessment with access to licensed clinicians. Its Utah prescription-renewal pilot makes the operational boundary particularly visible: AI participates in deciding whether an existing treatment can continue under a specific agreement. Utah's Doctronic description.
The current qualification is essential. As checked on 5 October 2026, the state's register still says Phase 1, with every request authorised by a licensed medical practitioner. It does not permit new treatment initiation or treatment-plan changes, and a later phase requires state approval. Current pilot register.
Why this is a big deal
Once an output enters a prescription workflow, the assessment extends beyond whether the conversation is informative. It includes the permitted action, liability arrangements, audit trail and failure modes being monitored. A patient should be able to establish who is providing care and how a concern reaches a clinician.
For example, a request to continue an established medicine may disclose a new symptom. That does not automatically mean the medicine should be changed, but it does mean the original prescription alone cannot answer every question about continuation.
My view (balanced)
The potential access argument concerns continuity, including people facing geographical or practical difficulty obtaining a routine renewal. The competing concern is whether an apparently routine request contains information requiring reassessment. The state's own early update distinguishes these intended benefits from robust outcome evidence not yet established in that account.
iatroX's dedicated Utah-pilot article provides the fuller UK comparison. The relevant question here is the actual renewal pathway, not whether the "AI doctor" label sounds reassuring or alarming.
Ada Health
What Ada is best known for
Ada describes symptom assessment and care navigation, with both a patient-facing presence and organisational applications. Its public material places emphasis on structured assessment and medical expertise. That establishes its stated approach, not a comparative claim that every Ada output is safer than every newer entrant's. Ada's current product information.
What I think Ada gets right conceptually
The useful design questions concern structured questioning, consistent assessment logic, navigation at the point of entry to care and handover into an appropriate pathway. None requires the sweeping claim that a symptom checker diagnoses better than a doctor in general.
A hypothetical health-system implementation should be assessed beyond the screen showing possible causes. Does the information reach the receiving service? Does the suggested destination fit the person's actual access options? Can the user correct an answer that was misunderstood? Those tests concern the pathway, not only the model.
My view (balanced)
Ada belongs in a serious comparison of symptom-assessment approaches, including its enterprise and clinical-handover positioning. The earlier article's description of a comparatively strong quality narrative should not be mistaken for an independently measured safety ranking.
Atypical presentations, incomplete inputs and user interpretation remain important test categories. Organisational deployment and published design controls are relevant evidence about a product's development, but the intended setting still needs its own assessment.
Buoy Health
What Buoy represents
Buoy illustrates conversational symptom checking: asking follow-up questions, narrowing possible explanations and suggesting a type of care. Its current site presents symptom assessment and navigation rather than an unrestricted autonomous prescribing service. Buoy's product description.
Why Buoy is worth taking seriously in comparisons
Buoy has been studied in relation to how people seek health information. A 2022 cross-sectional questionnaire study in the Journal of Medical Internet Research examined use of an intelligent web-based symptom checker. That is relevant evidence about behaviour and the setting in which the tool is used, not an outcome trial proving that every navigation recommendation improves health. Published study.
It creates useful evaluation questions: what prompted the person to use the service, what information did they already have, and did the output change the next step? A successful conversation cannot be defined only as reaching its final screen.
My view (balanced)
Human factors deserve direct assessment alongside model performance. A plausible explanation could be misunderstood as reassurance, and a user might delay obtaining necessary help. That is a potential failure mechanism to test, not a claim that Buoy caused such an event in the study cited here.
For readers comparing symptom checkers, the important distinction is between a possible cause, an urgency assessment and a completed connection to care. Those outputs should not be treated as interchangeable.
K Health
What makes K Health a different beast
K Health is more usefully understood as a hybrid care proposition than as a standalone symptom checker. Its public positioning connects AI-supported assessment and information gathering with clinician-delivered care. The current website also describes infrastructure for health-system use, so the product's role should be assessed in the actual service being considered. K Health, checked on 5 October 2026.
Where the real value sits
The potential operational gains identified in the original article remain distinct: faster intake, a better structured history, more appropriate routing, less administrative duplication and continuity into human care. They are hypotheses to measure, not benefits that follow automatically from adding AI.
In a fictional encounter, a patient completes a detailed intake but still needs the clinician to resolve a contradictory answer. An effective hybrid process should make that uncertainty easy to see. Requiring the clinician to reconstruct everything from an overconfident summary would be a different result, even if the intake itself was fast.
My view (balanced)
The strategic appeal of a hybrid model is that assessment support, escalation and follow-up can sit within one care arrangement. That may be valuable where a patient's need cannot be completed by an informational answer alone.
Quality at scale remains a question for evaluation. Measures should examine appropriate treatment and resolved concerns alongside speed and conversion, without assuming that commercial growth either guarantees or necessarily damages care quality.
Akido Scope AI
Why Akido Scope AI is strategically important
Akido places Scope AI within care delivery rather than describing it only as a public chatbot. Its current product page discusses assembling patient information, provider feedback used in reinforcement learning, review layers, guidance grounding and continuing output evaluation. These are the company's design claims, not a certification that each resulting plan is appropriate. Akido's Scope AI description.
The distinction matters because assembling the case can influence the clinician before a final recommendation is made. Which detail appears first, which contradiction is retained and which missing information becomes a follow-up question can all affect the work that remains.
What this category could unlock
The potential gains concern turning an unstructured patient account into a focused summary, useful follow-up questions, clearer documentation and a draft plan for review. In a capacity-constrained service, including an underserved setting, reducing unnecessary reconstruction might release time for clinical judgement.
That remains a conditional argument. A summary that is shorter but less faithful would not establish the same benefit. Nor would faster completion be sufficient if unresolved concerns are pushed into a later encounter.
My view (balanced)
For a care organisation evaluating workflow support, Akido raises questions that a generic chatbot comparison would miss. What is automated, what is suggested and what requires professional sign-off? How are errors reviewed, and how are unusual cases identified?
The current Akido website, checked on 5 October 2026, distinguishes availability through its medical network from wider availability described as forthcoming. That distinction should remain visible rather than treating the whole future service model as already universally accessible.
OpenEvidence
Important distinction: this is not a patient AI doctor
This heading refers to the clinician-facing evidence experience reviewed here, not a claim that OpenEvidence has no other product surfaces. Its educational and product material describes natural-language clinical questions answered with supporting citations for point-of-care use. OpenEvidence's educational material and product information, checked on 5 October 2026.
An answer that helps a clinician orient themselves in the literature does not itself perform a patient assessment, issue a prescription or establish a continuing care relationship. That is the category distinction the original article intended to make.
Why this matters clinically
The practical questions are familiar: has guidance changed, does a particular study apply, and what evidence addresses the patient's specific concern? Finding and appraising that material takes work, and the consultation does not pause indefinitely while the clinician searches.
A proposed evaluation should therefore examine whether the tool helps the clinician locate relevant evidence and recognise its limitations. The existence of a citation is only the beginning: the source must support the claim and be applicable to the question.
My view (balanced)
"Second brain" is a metaphor for assistance with information, not an independent professional second opinion. Clinician-facing evidence systems still need scrutiny for selective retrieval, unsupported inference, overconfident summaries and application to the wrong patient context.
Those requirements also apply to iatroX. This comparison does not establish that either service is more accurate. It distinguishes the task of supporting evidence use from the separate work of providing care.
Nolla Health
What its October announcement adds
Nolla announced a Utah pathway for initial acne prescriptions on 5 October 2026. Utah's register describes a pilot for topical treatment of mild to moderate acne in adults, excluding oral medication and severe acne. That is different from Doctronic's renewal of an already established treatment. Nolla's announcement and Utah's independent description.
The service's advertised starting price was $4.99 per month in the 5 October 2026 launch material. This is a patient-service price, not a directly comparable unit to a professional-learning subscription, an individual consultation or medication costs. The headline should not be assumed to include every associated cost. Company launch release.
Why the product names and oversight need care
Nolla's broader website includes Nolla Derm and a Health Assistant, while NollaMD also appears in specialist-model research. These labels should not be collapsed into one identical clinical workflow. Its website, checked on 5 October, distinguishes live dermatology and virtual urgent care from developing services and roadmap areas. Nolla's current service descriptions.
For the Utah pilot, published descriptions move from physician approval before treatment towards possible retrospective review and sampling. The state retains control over progression. Nolla said access was open on 5 October, while the state's register still described the demonstration period as not started; no later phase should be inferred from the launch announcement.
My view (balanced)
For an eligible patient seeking a defined acne-treatment pathway, Nolla is more relevant than a clinician literature-search tool. Its assessment must nevertheless extend beyond image recognition to suitability, adverse effects, unresolved uncertainty and completed follow-up.
Utah explicitly says its sandbox is not endorsement or product approval. The opportunity is to evaluate a bounded service, not to declare autonomous care validated on the basis of permission to test it. iatroX's dedicated Nolla guide, safety article and renewal-versus-initiation comparison explore those distinct questions without turning the company into another generic AI-doctor profile.
Comparison: what each tool is actually optimised for
This table summarises the documented roles above as checked on 5 October 2026. Potential strengths are task matches, not results from a comparative trial. Risks are evaluation questions, not allegations of observed failures.
| Tool | Primary user in this comparison | Core job | Potential strength for that job | Main question or limitation |
|---|---|---|---|---|
| Medome | Patient | Organise information and prepare for care | A longitudinal account and focused appointment questions | Fidelity of summaries, overreliance and privacy expectations |
| Doctronic | Patient, with clinical oversight | AI-supported assessment and established-prescription renewal within a specified pilot | More convenient continuity for eligible requests | Actual phase, suitability, monitoring, liability and escalation |
| Ada Health | Patient or care organisation | Symptom assessment and navigation | Structured questions and a defined handover route | Atypical cases, input quality and whether users reach appropriate care |
| Buoy Health | Patient | Conversational symptom assessment and navigation | Dynamic questioning and next-step information | Interpretation, reassurance and potential delay in obtaining care |
| K Health | Patient or care organisation | AI-supported intake connected with clinician care | Less duplication and continuity within a hybrid service | Quality, scope control and unresolved work as use expands |
| Akido Scope AI | Clinicians and care organisations | Assemble context and support care workflows | More focused information and draft work for review | Meaningful oversight, auditability and handling of exceptions |
| OpenEvidence | Clinician using its evidence experience | Citation-backed clinical information | Orientation in relevant evidence | Source selection, unsupported inference and applicability |
| Nolla Health | Patient seeking care | Assessment, treatment and follow-up; a bounded Utah acne pilot | A more complete route than an informational answer alone | Initial-treatment suitability, operating phase and outcomes |
| iatroX | Clinician or medical learner | Source-linked reference and structured professional learning | Connecting a clinical question with explanation and practice | Verification remains necessary; not a patient-prescribing service |
What clinicians and founders should notice (the strategic bit)
Our strategic interpretation is that model performance alone will not determine the usefulness of these businesses. Workflow fit, provenance, escalation, operational responsibility and an honest interface can all affect whether a capable model becomes a dependable service.
The questions are concrete. Does the product save work where work is actually being duplicated? Can the user trace a consequential claim? Does an escalation reach someone who can help? Are uncertain or incomplete tasks visible, or merely hidden behind a completed status?
These are hypotheses about durable product value, not predictions of which company will win. A specialist can be useful without owning the most general model, and a broad platform still needs evidence for the particular task it offers. Nolla's treatment pathway and OpenEvidence's evidence experience should therefore be assessed against different operational endpoints.
The real safety issue: confidence without provenance
A plausible failure scenario combines an authoritative tone, fast answers, weak source visibility, difficult-to-find qualifications and an unclear human route. The resulting output may be easy to trust without being easy to inspect. This can matter both to a worried patient and to a clinician working under pressure.
The remedy is not simply to add more disclaimers. The interface should make clear what is known, what was inferred and what remains unresolved. It should also distinguish a recommendation from an action already taken. A prescription prepared for review and one already sent to a pharmacy should never be represented as the same state.
For UK clinical use, professional responsibilities concerning sufficient information, appropriate assessment and continuity remain relevant regardless of the interface. The GMC's remote-consultation principles, checked on 5 October 2026, provide that professional context rather than a product-specific verdict on these US services.
A practical framework for evaluating any "AI doctor" / second-opinion tool
1) What is the claimed job?
Identify whether the function is education, symptom assessment, navigation, documentation, evidence synthesis, prescription workflow or diagnosis. A product may span several, but evidence for one should not silently validate the others. Ask what outcome the service is actually promising to complete.
2) Who is the user, and what is their risk level?
A clinician has different training and responsibilities from a patient, but professional status does not make every AI answer safe to accept. Assessment should reflect the intended users, patient population, consequences of error and realistic ability to recognise a problem.
3) What happens when uncertainty is high?
Look for explicit uncertainty handling, appropriate escalation of concerning information, urgent-care instructions where relevant and a usable route to a human. Then test the route. Displaying an escalation message is not equivalent to ensuring that the person obtains the help they need.
4) Can the user inspect provenance?
For an evidence answer, inspect the source and whether it supports the statement. For a record summary, inspect the original information and any change in meaning. For a treatment proposal, establish what assessment and patient-specific information support it. Provenance means more than a familiar publication name.
5) What is the operational governance?
For tools participating in care, request a clear account of logging, audit, incident review, monitoring and version changes. Establish who can stop a workflow, contact an affected patient and correct a downstream record. A demonstration of successful completion should be accompanied by a demonstration of appropriate refusal, interruption and recovery.
My bottom line
A patient trying to organise a complicated health history should assess Medome's preparation role. Someone seeking symptom assessment or help choosing a care route should examine Ada or Buoy against that specific need. Someone needing a clinician-connected service should inspect the actual K Health, Akido or Doctronic pathway, including availability and oversight.
For an eligible patient seeking initial acne care, Nolla adds a more specific treatment proposition, but its pilot should not be confused with unrestricted prescribing. For a clinician seeking evidence, OpenEvidence and Ask-iatroX belong in a reference comparison. For structured revision or reasoning practice, iatroX's learning functions answer another professional need.
None of these scenario matches establishes universal superiority. The central question remains where the tool creates useful support in the care or learning journey, and whether the safeguards fit the action that follows.
Choose the right tool for the task
iatroX's September 2026 product information describes free Ask-iatroX and free question access without a trial expiry or verification gate. Its reference methodology links answers to NICE, CKS, SIGN and SmPC information from emc, with retrieval, ranking, citation grounding, output checks and uncertainty handling. Those are design features to inspect, not a promise that every output is correct. iatroX's methodology.
A clinician can use a general clinical question as the starting point for source checking and further learning. Patient-specific decisions still require the necessary history, examination or other information and the appropriate professional care arrangement. Choosing a tool by that distinction is more useful than asking whether every product carrying an AI label is the same kind of doctor.
Frequently asked questions
Are all AI doctor apps treatment services?
No, some organise information or assess symptoms, while others provide clinician-facing evidence or participate in a defined care pathway. The service's actual role, jurisdiction and human involvement need to be checked separately.
How is Nolla's Utah pilot different from Doctronic's?
Nolla's announced pilot concerns initial topical acne treatment within a limited adult pathway, whereas Doctronic's pilot concerns renewal of an existing prescription. Both have staged oversight, and a launch announcement does not establish progression to a later stage.
Does a cited answer count as an independent medical second opinion?
No, citations can support an informational answer without establishing a complete patient assessment or a professional care relationship. The source must still support the claim and fit the patient's circumstances.