What AI Can and Cannot Do in an Indian Clinic: A Plain-English Guide for Doctors

Most Indian doctors currently receive information about artificial intelligence from two sources: vendor demonstrations and anxious news headlines. Neither is useful for a practitioner trying to make a grounded decision about their clinic. The vendor demo shows AI at its best - curated datasets, controlled conditions, a polished interface. The headline shows AI at its most alarming - misdiagnosis, data breaches, job displacement. Neither version resembles the actual choice in front of a practitioner managing a thirty-patient day in Pune or Coimbatore.

The practical reality in 2026 is that the AI tools most immediately available and relevant to an independent Indian clinic fall into three distinct categories - administrative, communication, and clinical decision support - and these categories differ enormously in their risk profile, governance requirements, and near-term return on investment. Getting the category wrong is the root cause of most failed AI adoptions and most legitimate professional concerns. An administrative scheduling tool and a diagnostic imaging AI are both "AI," but deploying them without distinguishing between them is like treating paracetamol and a new biological the same because both are "medicine."

After reading this article, you will understand the three-category framework, know which specific tools belong to each category, understand what Indian regulations (DPDPA, NMC guidelines, ABDM) actually require of you, and have a concrete checklist for this month. You will also understand, plainly and directly, where AI genuinely cannot yet help - and where the hype is ahead of the evidence.

---

Dr. Kavitha Subramanian runs a mid-sized gynaecology practice in Bengaluru's Jayanagar neighbourhood. She has three staff, sees forty patients a week, and manages a WhatsApp number that receives an unbroken stream of messages from 7 a.m. to 11 p.m.: appointment requests, prescription queries, post-procedure questions, forwarded family medical reports, and the occasional anxious midnight message about symptoms that may or may not be relevant to her care. Last year, a medical equipment rep demonstrated an "AI diagnostic assistant" during her lunch break. It was impressive in the demo. She never bought it, partly because the price was steep, partly because she was not sure whether using it would create legal exposure, and partly because - though she would not say this to the rep - she did not understand clearly what it was actually doing.

Her uncertainty is shared by the large majority of Indian practitioners who have heard about AI but have not formed a clear view of it. The information environment around clinical AI in India in 2026 is not information-poor - it is information-chaotic. There are genuine, published, peer-reviewed deployments of AI in Indian healthcare: Qure.ai's chest X-ray analysis has been deployed at the national TB elimination programme level; Tricog's ECG platform operates across cardiac care networks; Niramai's thermal imaging tool is used in breast cancer screening contexts. These are real products solving real diagnostic problems. They are also hospital-grade, regulatory-grade deployments, not tools that belong in a private gynaecology clinic in Jayanagar without formal clinical governance infrastructure.

The conflation of these categories - "AI for enterprise hospital diagnostics" with "AI for a small independent practice" - produces two dysfunctional responses. The first is paralysis: a practitioner concludes that AI is too complex, too risky, and too expensive, and therefore ignores the administrative tools that could meaningfully reduce her staff's burden this week. The second is reckless adoption: a practitioner, excited by a demo, deploys an AI tool in a clinical decision-support context without understanding its validation data, its failure modes, or what the NMC's professional conduct guidelines require when an AI-assisted recommendation turns out to be wrong.

Both responses harm patients - the first by leaving the practice under-resourced and the practitioner burnt out; the second by introducing unvalidated clinical risk. The problem is not that AI is good or bad. The problem is that "AI" is a category that contains tools as different as a Gmail autocomplete and a chest CT reading algorithm, and practitioners need a reliable way to sort them.

---

India's healthcare context makes this sorting problem more consequential than it would be in, say, the NHS. Research by Sharma et al. (2025), published in Global Health Action (PMC11998304), surveyed 5,061 PM-JAY eligible individuals across seven Indian states and found that 48.0% sought outpatient care from private providers, against just 18.3% using public facilities - with a striking 23.1% accessing no regular outpatient care at all. This means independent private practitioners carry a disproportionate share of India's primary care burden. When a solo practitioner gets an AI adoption decision wrong - in either direction - the downstream effects reach a larger share of the population than the same decision by a single consultant in a Western hospital system.

The platform infrastructure compounds this. India crossed 800 million smartphone users in 2024, with a median population age of 29. A significant and growing share of patient-doctor interaction in India now happens via WhatsApp, Google search, Practo, and JustDial - not just in the consultation room. The practical implication is that AI tools that assist with digital communication channels are not peripheral for Indian practitioners; they are central. An AI tool that helps a practitioner respond accurately and quickly to WhatsApp queries, or that helps manage Practo reviews, is operating at the actual patient interface - which raises both the benefit potential and the communication-error risk.

The regulatory environment is evolving fast, and practitioners should not wait for it to stabilise before acting. The Digital Personal Data Protection Act 2023 (DPDPA) creates enforceable obligations around how personal data - including health data - is collected, stored, processed, and shared. Using a foreign AI tool to process patient details without appropriate data processing agreements is not a grey area; it is a compliance exposure. The Ayushman Bharat Digital Mission (ABDM) and the ABHA health ID framework are building a national health data infrastructure that AI tools will increasingly need to interoperate with. And the National Medical Commission's professional conduct regulations do not yet have AI-specific guidance, but the general principle is clear: the doctor remains professionally responsible for every clinical decision, whether or not an AI tool contributed to it.

Cultural factors shape AI adoption in ways that technology benchmarks from the US or UK do not capture. A FICCI-EY Parthenon survey of over 1,000 patients and 100 clinicians in India (October 2025, an industry report - not peer-reviewed) found that patients rely heavily on informal proxies such as brand reputation and word-of-mouth when choosing healthcare providers, while 83% expressed an aspiration to access clearer health information. The aspiration gap is real: patients want to understand more, and practitioners who use AI to communicate more clearly and accessibly can convert that aspiration into trust. But the same word-of-mouth dynamic means that an AI communication error - a chatbot that gives wrong medication information, a draft that contains a factual mistake - travels fast and damages reputation disproportionately.

---

The evidence base for AI in Indian clinical settings is real but uneven. Qure.ai's qXR platform has been evaluated in peer-reviewed settings for chest X-ray analysis, including tuberculosis detection. A 2021 study published in PLOS ONE (Qure.ai investigator-led, with the limitation that it was conducted by the company developing the tool) demonstrated sensitivity and specificity competitive with radiologist reads for TB findings on chest X-ray. The platform has since been integrated into India's national TB elimination programme. This represents genuine validation at scale - but it is a hospital-network and public-health deployment, not a tool for a general practitioner to deploy without institutional oversight. The evidence supports the tool; the governance context matters as much as the evidence.

Tricog's STAT12 ECG analysis platform has published outcome data from its deployments across cardiac hospitals in India, showing reductions in door-to-balloon time in AMI cases where the platform flagged critical ECGs for immediate specialist review. Again, the directional evidence is positive, and the deployment context is instructive: the tool is positioned as a triage and routing aid within a defined cardiac care protocol, not as a standalone diagnostic. The clinician remains in the loop; the AI flags and prioritises. This "augmentation without replacement" architecture is the operational pattern that the evidence consistently supports in clinical settings - and consistently departs from in marketing materials.

For AI in administrative healthcare contexts, the relevant evidence base is largely from US and other high-income country settings and must be labelled accordingly: directionally applicable to India, not generalisable without qualification. A systematic review by Sinsky et al. published in JAMA Network Open (2021) found that administrative tasks - documentation, coding, prior-authorisation workflows - represent the largest single source of physician time burden in ambulatory care settings. AI tools that reduce this burden showed statistically significant improvements in clinician-reported burnout scores in the studies reviewed. These are US findings; the specific administrative structures differ from Indian private practice. But the directional principle - that administrative AI has a favourable benefit-to-risk ratio compared to clinical AI - is consistent across contexts and systems.

For AI in patient communication specifically, the evidence is more mixed and warrants honest characterisation. A 2023 study in JMIR Medical Informatics (directionally applicable to India) evaluated chatbot-assisted patient triage in a primary care setting and found high patient satisfaction scores but also a 12% rate of "hallucinated" information - factually incorrect statements generated with apparent confidence by the language model. The 12% figure should give every practitioner pause. It does not mean communication AI is unusable; it means the "review required" label on that category is not bureaucratic caution but an evidence-based minimum requirement. Every AI-drafted patient communication must pass through a trained human reviewer before it reaches a patient.

The research on AI for documentation - specifically, AI scribing tools that transcribe and summarise clinical encounters - is among the most promising and the most immediately applicable. Multiple US studies (directionally applicable to India) show that AI scribing reduces post-consultation documentation time by 40-60%, with high accuracy for common clinical note structures. In India, where many practitioners still type their own notes or maintain paper records, the efficiency gain could be substantial. The main risk is over-reliance: if a practitioner stops reading the generated note and just signs it, errors propagate into the medical record without detection. The evidence supports use with review; it does not support unsupervised automation of clinical documentation.

---

The following section is Influx Health's interpretation and opinion, not research.

Working with over sixty Indian healthcare organisations - clinics, hospitals, diagnostic centres, dental chains - over the past three years, we have observed a consistent pattern in how AI adoption goes wrong: organisations skip the administrative tier and attempt communication or clinical AI first, because administrative automation feels less exciting. The result is predictable. A chatbot deployed on a hospital's WhatsApp line without a review loop produces three or four factual errors within the first two weeks. Staff spend more time managing fallout from chatbot errors than they would have spent answering messages manually. The organisation concludes "AI doesn't work for us" and retreats entirely - missing genuine administrative gains that would have been low-risk and impactful.

The second observation is that DPDPA compliance is being treated as an afterthought in the majority of the smaller clinic deployments we encounter. Practitioners are using international AI tools - often free-tier products with unclear data processing terms - to process patient names, contact numbers, and clinical details. The question of where that data is stored, who has access, and what the tool provider's data retention policy is, is often not asked. This is not a minor legal technicality. Health data falls within the sensitive personal data category under DPDPA, with heightened obligations. This will matter more, not less, as enforcement develops. It is better to build compliant workflows now than to retrofit them later.

The third observation is perhaps the most counterintuitive: the highest-impact AI adoption we have seen in Indian clinical settings has not been diagnostic AI or even communication AI - it has been AI for content creation and patient education. Clinics that use AI to produce clear, medically accurate WhatsApp explainers, FAQ documents, and post-procedure instruction sheets - reviewed and approved by the treating doctor - consistently see improvements in patient follow-through, reduced re-call rates, and, anecdotally, a lower volume of after-hours queries. The AI does not "do" medicine. It translates medicine into language patients can act on. That is a meaningful clinical contribution, and it sits entirely within the administrative-communication tier where risk is manageable.

The fourth observation is a caution for the opposite direction: we have also seen organisations deploy clinical AI tools - imaging AI, risk-stratification algorithms - without the institutional infrastructure to govern them. No documentation of validation data reviewed. No protocol for what happens when the AI and the clinician disagree. No staff training on the tool's known limitations. No patient disclosure. This is not an AI problem; it is a governance problem. Clinical AI can only be safely deployed inside a governance framework that the organisation has deliberately built. Most small-to-medium Indian private healthcare organisations have not built that framework yet. This is not a permanent barrier - it is a sequencing point. Build the administrative tier first. Build the governance infrastructure. Then evaluate clinical AI.

---

A structured method for categorising and deploying AI tools in your practice

C - Categorise before you adopt. Before evaluating any AI tool, assign it to one of three tiers: administrative (scheduling, documentation, billing), communication (patient messaging, content creation, review responses), or clinical (diagnostic support, risk stratification, treatment recommendations). The tier determines your governance requirement, not the price or the vendor's marketing. A "smart scheduling assistant" is Tier 1 even if it uses sophisticated machine learning. An algorithm that flags high-risk patients is Tier 3 even if it ships in a friendly interface.

L - Lead with administrative applications. Administrative AI - appointment reminders, prescription draft generation, call-log transcription, billing code suggestions - has the highest benefit-to-risk ratio of any AI tier in a clinic context. Errors in this tier are usually visible, correctable before they reach patients, and low-consequence when they do occur. If your clinic has not yet deployed any AI, this is the correct starting point. Begin with one tool, one workflow, and measure the time saved over four weeks before expanding.

I - Inspect every patient-facing output. No AI-generated message, document, or response should reach a patient without a review step by a qualified staff member or clinician. This is not optional caution - it is an evidence-based requirement given published error rates in language model outputs. Build the review step into your workflow explicitly: assign the reviewer, document the sign-off, and never assume a piece of communication is accurate because it sounds authoritative. Automated does not mean verified.

N - Navigate clinical AI only with governance infrastructure. Clinical decision support AI should not be deployed in your practice until you have: (a) reviewed the tool's published validation data in a population comparable to your patient base, (b) established a written protocol for disagreements between the AI and the clinician's judgment, (c) documented staff training on the tool's known failure modes, and (d) determined your disclosure obligations under NMC professional conduct guidelines. If any of these four are missing, defer the deployment - not indefinitely, but until the infrastructure exists. The tool itself may be excellent. The gap is in readiness.

I - Integrate DPDPA compliance from day one. Before deploying any AI tool that processes patient data - names, contact numbers, symptoms, diagnoses - confirm where the data is stored, whether the provider has a valid Data Processing Agreement available, whether data is retained after processing and for how long, and whether the tool complies with DPDPA's requirements for sensitive personal data. Do not use free-tier versions of international AI tools that route patient data to foreign servers under consumer terms of service. This is not bureaucratic excess; it is the legal floor.

C - Calibrate expectations: AI augments, it does not replace. The single most useful mental model for AI in a clinic is this: AI is a capable junior assistant who needs supervision. It can draft, organise, flag, summarise, and remind. It cannot exercise clinical judgment, bear professional responsibility, understand patient context the way you do, or navigate the subtleties of a consultation. Practitioners who approach AI as a force multiplier for their own capabilities - rather than as a replacement for them - make better decisions, catch more errors, and get more sustainable value from the tools they deploy.

---

• This week - audit your current workflows: List every task your staff or you perform more than five times per day that does not require clinical judgment (appointment booking, reminder messages, repeat prescription requests, documentation formatting). These are your Tier 1 AI candidates. Prioritise the two highest-volume ones.

• This week - check your current tool exposure: If you are using any AI tool (including WhatsApp Business automations, chatbot plugins on your website, or general-purpose language models for writing) to process patient names, phone numbers, or health details, check the provider's data processing terms. Confirm whether a DPDPA-compatible Data Processing Agreement is available.

• This month - pilot one administrative AI tool with a defined review protocol: Choose one tool (appointment reminder automation, AI-assisted note drafting, or patient FAQ document generation), deploy it for one workflow only, assign a named staff member to review every output for the first four weeks, and keep a log of errors. Decide whether to expand based on the error rate and time saved.

• This month - create a patient communication template library: Use an AI tool (with clinical review) to draft ten to fifteen standard patient education pieces: post-procedure instructions, common medication FAQs, preparation instructions for standard investigations. Once reviewed and approved by you, these become your communication asset base - accurate, consistent, and available to staff without requiring per-message AI generation.

• Before considering any clinical AI - build your governance checklist: Write a one-page protocol covering: (1) how clinical AI recommendations will be reviewed, (2) what happens when AI and clinician disagree, (3) how the tool's validation data has been assessed, and (4) how patients will be informed. This document does not need to be complex - it needs to exist before the tool is deployed.

• Ongoing - subscribe to NMC and DPDPA update channels: The regulatory environment for AI in Indian healthcare is actively evolving. NMC has not yet issued AI-specific guidelines, but guidance is expected. DPDPA enforcement is expanding. Follow updates at nmc.org.in and meity.gov.in and review your AI tool inventory against new guidance as it is issued.

---

Q: Can I use ChatGPT or a similar tool to help write patient education content?

Yes, with the right workflow in place. General-purpose language models like ChatGPT, Claude, or Gemini are capable of producing clear, readable first drafts of patient education material - medication instructions, post-procedure guidance, explanation of common conditions. The critical requirement is clinical review before publication or distribution. These tools generate fluent-sounding text that is occasionally factually wrong, and they have no way of knowing the specific context of your patient population, your clinical protocols, or regional treatment preferences. Use them to draft, use your clinical judgment to verify and edit, and never send AI-generated medical content to a patient without reading every sentence yourself. The DPDPA compliance question also matters: do not input patient-specific details (names, diagnoses, identifiers) into free-tier consumer products that may retain your inputs for model training.

Q: My practice already uses Practo or a similar platform. Does that count as "using AI"?

Partially. Platforms like Practo embed varying degrees of automation and data analysis, some of which use machine learning in the back end. But the distinction that matters is whether you are using AI to assist your clinical decisions or to assist administrative and communication workflows. Practo's appointment scheduling, review aggregation, and patient communication features are primarily administrative-tier tools - which means a lower governance burden and a more favourable risk profile. If Practo or any platform you use introduces a feature that generates clinical recommendations (risk flags, diagnosis suggestions), treat that feature as Tier 3 clinical AI and apply the CLINIC model's governance requirements before relying on it.

Q: What are the NMC's current rules about AI-assisted clinical decisions?

As of mid-2026, the NMC has not issued AI-specific guidance for clinical practice. The relevant framework is the NMC's general professional conduct regulations, under which the treating doctor bears full professional responsibility for clinical decisions - regardless of whether an AI tool contributed to those decisions. This means that if an AI-assisted recommendation leads to patient harm, the treating clinician's professional accountability is not reduced by the AI's involvement. The practical implication is that "the AI suggested it" is not a defence before the NMC's disciplinary committees. This will likely sharpen over time as AI-specific guidance is developed. In the interim, operate on the principle that every AI-assisted clinical decision must be one you can defend on its clinical merits, independently of the AI output.

Q: Are the Indian diagnostic AI tools - Qure.ai, Tricog, Niramai - appropriate for a private outpatient clinic?

These are validated, established tools with genuine evidence behind them. Whether they are appropriate for your clinic is a governance question, not a product quality question. Qure.ai's chest X-ray analysis is validated for TB screening at scale; it is most appropriately deployed within a structured TB screening protocol with radiologist oversight and a defined referral pathway. Tricog's ECG tool is designed for integration with a cardiac care network where the triage decision leads to a defined clinical response. Niramai operates in a structured screening programme context. If your clinic can provide the governance infrastructure - defined clinical protocols, oversight from appropriate specialists, patient consent processes, and a clear action pathway for positive findings - these tools can add genuine value. If the governance infrastructure is not in place, deploying them anyway increases clinical risk without capturing the clinical benefit.

Q: I've been approached by a vendor selling an "AI that reads reports and flags abnormals for my patients." Should I look at it?

Approach it carefully. "AI that reads reports" is a clinical decision support tool, which places it in Tier 3 of the CLINIC model. Before agreeing to a demo, ask the vendor four questions: (1) What published validation data exists for this tool in an Indian patient population? (2) What is the tool's known false-positive and false-negative rate? (3) What is the recommended clinical protocol for handling a positive flag from this tool? (4) Who bears professional liability for a missed finding or a false positive that leads to unnecessary intervention? If the vendor cannot answer all four questions clearly and in writing, that is a significant signal about the product's maturity. If they can, the answers will tell you whether your clinic is ready to deploy it or whether you need governance infrastructure first.

---

Internal - Influx Health Institute - Patient Data Privacy for Indian Clinics: A DPDPA Compliance Primer - What the Digital Personal Data Protection Act 2023 actually requires of healthcare providers, with a step-by-step audit checklist. - AI Diagnostic Tools in India 2026: What Clinicians Need to Know About Qure.ai, Tricog, and Niramai - A clinical governance assessment of the three most widely deployed AI diagnostic platforms in the Indian market. - ChatGPT for Clinic Communications: What Is Safe, What Requires Review, and What to Avoid - A practical workflow guide for practices using general-purpose language models in patient-facing communications.

External - Authoritative Sources - Sharma et al. (2025), "Healthcare utilization among PM-JAY eligible individuals," Global Health Action. Full text: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC11998304/ - National Medical Commission - professional conduct regulations and guidance: https://www.nmc.org.in - Ministry of Electronics and IT - Digital Personal Data Protection Act 2023 resources: https://www.meity.gov.in/data-protection-framework

---

Read Next: AI Diagnostic Tools in India 2026: What Clinicians Need to Know About Qure.ai, Tricog, and Niramai - the second article in this series moves from the framework above to a detailed clinical governance assessment of the three leading Indian diagnostic AI platforms.

Assess Your Practice: Run a free Digital Presence Meter scan at /dpm - a 90-second scan of your practice's current digital footprint across Google, Practo, JustDial, and social channels. Understand where you stand before deciding where to invest.

Chat with Influx Health: Speak with our team at /contact - if you are considering AI adoption for your practice and want a structured conversation about sequencing, governance, and realistic timelines, our team works with healthcare organisations across India and can give you a direct, non-sales assessment of where to start.

--- ---

# Content Derivatives: Center 7, Article 1

---

Subject line: The three types of AI in your clinic - and why the category matters more than the tool

---

Dear Dr. [Name],

If you have attended a medical conference in the past eighteen months, you have almost certainly sat through at least one session on AI in healthcare - and come away with a mixed feeling of possibility and uncertainty. The presentations tend to be impressive and the claims large. The gap between "this is what AI can do" and "this is what I should actually do in my clinic on Monday" tends to be wide.

This month, the Influx Health Institute published a plain-English guide to exactly that question. The central argument: the most important thing you can do before evaluating any AI tool is identify which of three categories it belongs to.

The first category - administrative AI - covers appointment management, reminder automation, documentation drafting, and billing assistance. This tier has a high benefit-to-low risk ratio, and most private clinics can begin deploying it this month with minimal governance overhead. The second category - communication AI - covers patient-facing messaging, content creation, and chatbot-assisted FAQ responses. This tier offers real benefits but requires a review step before anything reaches a patient; published error rates in language model outputs make unsupervised deployment genuinely risky. The third category - clinical decision support AI - covers diagnostic tools, risk-stratification algorithms, and report analysis. This tier requires institutional governance infrastructure - validation data review, written protocols, staff training - before it is safe to deploy, regardless of the tool's quality.

Three additional points are worth your attention. First, DPDPA compliance is not optional: if you are using any AI tool that processes patient names, contact numbers, or health details, check whether you have a compliant data processing agreement in place. Free-tier consumer products almost certainly do not meet the standard. Second, the highest-impact AI adoption we have seen in Indian clinics is in patient education content - AI-drafted, clinician-reviewed explainers that improve patient follow-through and reduce after-hours queries. Low risk, high practical value, and deployable this month. Third, "the AI suggested it" is not a defence before an NMC disciplinary committee. Professional responsibility for clinical decisions remains with the treating doctor, regardless of AI involvement.

The full article - including a six-element CLINIC framework, a five-item action checklist, and detailed answers to the questions practitioners most commonly ask - is available at the link below. If you would also like to understand where your practice stands digitally before making technology decisions, our free Digital Presence Meter scan takes ninety seconds.

[Read the full article] | [Run your Digital Presence Meter scan at influx-health.com/dpm]

Warm regards, The Influx Health Institute Research Team

---

AI in your clinic: what actually matters in 2026 (3-minute read)

Most Indian doctors receive AI information from two sources: vendor demos (too optimistic) and news headlines (too alarming). Neither helps you make a practical decision.

Here is what actually matters:

The three categories: 1. Administrative AI - scheduling, reminders, documentation drafts. Low risk. Start here. 2. Communication AI - patient messaging, content creation. Medium risk. Always review before sending. 3. Clinical AI - diagnostic support, report analysis. High governance requirement. Build the protocol first.

What is working in Indian clinics right now: - Appointment reminder automation (reducing no-shows without adding staff burden) - AI-drafted patient education content, reviewed by the doctor before distribution - AI-assisted note transcription (with clinician review of every note)

What requires caution: - Any AI tool that processes patient data without a DPDPA-compliant data processing agreement - Chatbots deployed on patient-facing channels without a human review step - Clinical decision support tools deployed without validation data reviewed and protocols written

The NMC position (current): No AI-specific guidelines yet. The existing rule applies - you are professionally responsible for every clinical decision, whether AI contributed to it or not.

Full article: influx-health.com/institute/center-7-ai-healthcare/what-ai-can-and-cannot-do Free practice scan: influx-health.com/dpm

---

Every few weeks I speak with an Indian doctor who has made one of two mistakes with AI: deploying it too cautiously in the wrong place, or deploying it too enthusiastically in the wrong place.

The cautious version goes like this: a practitioner reads about AI misdiagnosis risks, decides AI is dangerous, and declines to automate appointment reminders or patient education content - tasks where the risk is low and the time savings are real. They continue answering routine WhatsApp messages manually at 10 pm.

The enthusiastic version goes like this: a practitioner sees a compelling demo of a diagnostic AI tool, deploys it without reviewing its validation data or writing a protocol for when it disagrees with the clinician, and encounters an error that creates both clinical and reputational damage. They conclude AI does not work.

Both mistakes share the same root cause: treating "AI" as one category.

The Influx Health Institute's new guide for Indian practitioners separates AI tools into three distinct tiers - administrative (low risk, start here), communication (medium risk, review required), and clinical decision support (high governance requirement, build the protocol first). The category a tool belongs to determines the governance overhead, the risk profile, and the realistic return on investment for your practice. Getting the category right before evaluating any specific tool is the most valuable thing you can do.

The article covers: - Specific tool examples for each tier (including Qure.ai, Tricog, and Niramai for clinical AI; WhatsApp automation and AI scribing for administrative AI) - What DPDPA actually requires when you process patient data through AI tools - What the NMC's current position means for your professional liability - A six-element CLINIC framework for structured AI adoption - Honest answers to the questions practitioners most commonly ask

This is the first of three articles in Center 7 of the Influx Health Institute - practical, evidence-based, written for the Indian clinical context.

Link in the comments. The free Digital Presence Meter scan is also there for practices that want to understand their current digital footprint before making technology decisions.

---

1/ Most Indian doctors are getting AI information from two places: vendor demos and alarming headlines. Neither is useful. Here is a plain-English framework for what AI can and cannot actually do in an Indian clinic. (Thread)

2/ "AI" is not one thing. A scheduling reminder tool and a diagnostic imaging algorithm are both "AI" - in the same way paracetamol and a biologic are both "medicine." The category matters more than the label.

3/ Three categories, in order of risk: - Administrative AI: scheduling, reminders, documentation drafts → low risk, start here - Communication AI: patient messaging, content → medium risk, review every output - Clinical AI: diagnostics, risk scoring → high governance requirement, protocol first

4/ Indian context matters here. Sharma et al. 2025 (PMC11998304): 48% of PM-JAY eligible patients use private outpatient care. Independent practitioners carry a disproportionate share of India's primary care burden. Getting AI adoption right - or wrong - has population-level consequences.

5/ DPDPA is not optional. If you are using any AI tool that processes patient names, contact numbers, or health details, you need a DPDPA-compliant Data Processing Agreement with that provider. Free-tier consumer AI products almost certainly do not meet the standard. Check now.

6/ The highest-impact AI use case in Indian clinics right now is NOT diagnostic AI. It is AI-drafted, clinician-reviewed patient education content. Better follow-through. Fewer after-hours queries. Entirely within the low-risk administrative-communication tier.

7/ On clinical AI (Qure.ai, Tricog, Niramai - all real, validated products): before deploying any of them, you need four things in writing: (a) validation data reviewed, (b) protocol for AI/clinician disagreements, (c) staff training on failure modes, (d) patient disclosure approach.

8/ The NMC's current position: no AI-specific guidelines yet. The existing rule applies. Professional responsibility for clinical decisions remains with the treating doctor, regardless of AI involvement. "The AI suggested it" is not a defence.

9/ The CLINIC framework for structured AI adoption: Categorise the tool → Lead with administrative → Inspect every patient-facing output → Navigate clinical AI with governance → Integrate DPDPA compliance → Calibrate expectations (AI augments, it does not replace).

10/ Full plain-English guide - with a six-element framework, five-item action checklist, and detailed FAQs - at the Influx Health Institute: influx-health.com/institute/center-7-ai-healthcare/what-ai-can-and-cannot-do

Free 90-second Digital Presence Meter scan for your practice: influx-health.com/dpm

--- Article published by the Influx Health Institute. Influx Health is a patient acquisition agency for healthcare organisations in India.