Author: Editorial Team airassvet.ru. This guide was prepared on September 29, 2026. Requirements for remote medicine sales are quoted from the Rules approved by Decree of the Government of the Russian Federation No. 697 dated May 16, 2020 (as amended on April 4, 2024, ConsultantPlus), and requirements for pharmacy staff are taken from the Rules of Good Pharmacy Practice, approved by Order of the Russian Ministry of Health No. 259n dated April 29, 2025 (effective September 1, 2025). This article is for informational purposes only: it is not medical advice and does not replace consultation with a doctor, pharmacist, or lawyer. The editorial team has no commercial relationship with the companies mentioned, and there are no affiliate links.
Short answer. An AI pharmacist is an AI agent on a pharmacy chain’s website, app, or messenger that answers customers about availability, price, analogs with the same INN, how to take and store over-the-counter medicines based on product instructions and the pharmacy catalog, collects orders, and passes complex cases to a pharmacist. It is useful as a first filter: it takes routine questions off pharmacists’ desks. It should not diagnose, choose treatment, advise on prescription drugs, or make decisions in urgent situations — it must stop and hand those cases to a human, and if there are signs of an emergency, direct the customer to a doctor or emergency services.
Below: what pharmacy chains are already doing, what rules set the boundaries, where the AI pharmacist’s limits are, and how to test it in a pilot.
Contents
- What pharmacy chains are already doing
- What rules set the boundaries
- What an AI pharmacist is and what it does
- Where AI must stop
- Data: without a catalog and instructions, AI is risky
- Other AI tasks in a pharmacy chain
- How to launch an AI pharmacist pilot
- Pre-launch checklist
- Frequently asked questions
What pharmacy chains are already doing
A recent public example is the 36.6 Pharmacy Chain. According to a CNews report dated September 21, 2026, in December 2025 the chain launched an AI agent that wrote short product descriptions, and over nine months turned it into an online pharmacist. According to the company, the agent understands symptoms and suggests what to choose, warns about risks, explains how to take medicines in plain language, and if needed transfers the customer to a live pharmacist.
The key point in this example is not the feature list, but the order of priorities. Marina Nastaeva, head of the chain’s own sales division, told CNews that the company put the rules for when the AI pharmacist should stop, transfer to a human, or add a warning ahead of new features. According to her, the online assistant does not replace a pharmacist — it works as a first filter.
For the pharmacy business, that is the practical model: AI responds quickly where the answer clearly follows from the instructions and catalog, and everything else goes to a human. The company did not disclose performance metrics for this rollout, so below we rely on the rules and process logic, not someone else’s numbers.
What rules set the boundaries
Two documents define what a pharmacy must tell customers and how AI can help with that.
Rules for remote sales (Government Decree No. 697). Paragraph 2 allows remote retail sales of medicines, except for prescription drugs, narcotic and psychotropic drugs, as well as alcohol-containing products with an ethyl alcohol content above 25 percent by volume. Paragraph 16 requires the authorized employee of the pharmacy organization or the marketplace owner, when taking an order, to inform the customer about indications, retail price, expiration date, dispensing conditions, storage rules, and interactions with other medicines. The same paragraph explicitly allows this information to be provided through a website or mobile app. Paragraph 15 requires protection of the customer’s personal data confidentiality.
Rules of Good Pharmacy Practice (Order No. 259n). They replaced Order No. 647n starting September 1, 2025. Paragraph 15 describes what information employees provide to customers: the availability of medicines with the same International Nonproprietary Name (INN), including lower-priced options, how to take them, dosing regimens, therapeutic effect, contraindications, interactions with other medicines and with food, and rules for storage at home.
| What to tell the customer | Source of the requirement | Can AI prepare the answer? |
|---|---|---|
| Availability and price, analogs with the same INN, including cheaper ones | Para. 15 of Rules No. 259n; Para. 16 of Rules No. 697 (price) | Yes, based on the pharmacy catalog and inventory |
| Indications, how to take it, dosage per instructions | Para. 15 of Rules No. 259n; Para. 16 of Rules No. 697 | Yes, word for word from the instructions, with a link to them |
| Contraindications and drug interactions | Para. 15 of Rules No. 259n; Para. 16 of Rules No. 697 | Only list them from the instructions and suggest consulting a pharmacist |
| Expiration date, dispensing conditions, storage | Para. 16 of Rules No. 697; Para. 15 of Rules No. 259n | Yes, based on the product listing data |
| Prescription drugs remotely | Para. 2 of Rules No. 697 — excluded from remote sales | No, only state that the medicine is dispensed by prescription |
The right-hand column is an editorial assessment, not a legal rule. Separate experimental legal regimes for remote sale of prescription drugs are not covered in this article: before building scenarios around them, confirm the current status with a lawyer.
What an AI pharmacist is and what it does
An AI pharmacist is an AI agent that answers a pharmacy customer in natural language, relying on the catalog, inventory, and official medicine instructions, and follows strict rules to hand the question to a pharmacist when the answer goes beyond those data. Typical feature set:
- Search and availability. The customer types a product name or INN — the agent shows availability, price, and pickup pharmacies.
- INN analogs. The agent lists medicines with the same INN and dosage, including cheaper options, as required by Para. 15 of Rules No. 259n. The decision to substitute remains with the customer and the doctor.
- How to take it, in plain language. A plain-language paraphrase of the instruction section on use and storage, with a link to the source document.
- Warnings. If the question mentions pregnancy, a child, a chronic condition, or other medicines, the agent displays the relevant section of the instructions and suggests consulting a pharmacist.
- Order placement. Cart assembly, pharmacy or delivery selection, order status.
- Transfer to a pharmacist. Along with the chat history, so the customer does not have to repeat the question.
For products that are not medicines — cosmetics, medical devices, mother and baby products — AI can operate more freely, but the rule “answer only from the product listing” still applies.
Where AI must stop
The breakdown below is an editorial risk guideline; the responsible chain pharmacist sets the actual rules.
| Situation | AI action | Who decides |
|---|---|---|
| Question about availability, price, delivery, order status | Answers | AI |
| How to take and store an over-the-counter medicine | Paraphrases the instructions with a link | AI, pharmacist spot-check |
| “What should I take for…,” selecting a remedy based on symptoms | Provides only general over-the-counter options from the assortment, reminds the customer to see a doctor, and escalates if there is any doubt | Pharmacist |
| Pregnancy, breastfeeding, children, older age, chronic conditions | Shows the relevant section of the instructions and hands it over to the pharmacist | Pharmacist or doctor |
| Combination of several medications | Escalates to the pharmacist | Pharmacist |
| Prescription medication | States that it is dispensed by prescription and that remote sale is not permitted under Clause 2 of Rule No. 697 | In-person pharmacy visit |
| Chest pain, difficulty breathing, signs of poisoning, overdose, suicidal statements | Stops the consultation and recommends calling 112 or 103 immediately | Emergency services |
The main rule: the AI pharmacist does not diagnose or prescribe treatment. Stop scenarios are designed before new features and tested just as carefully as answer quality. That is exactly the priority order described by 36.6.
A stop scenario is conveniently built from three parts: a list of triggers (words and topics), mandatory response text, and an action — escalation to a pharmacist or a recommendation to seek emergency help. All of this is stored in configuration, not “in the model’s head,” and can be changed only with the approval of the responsible pharmacist.
Data: without a catalog and instructions, AI is risky
A large language model without access to the pharmacy’s data will answer from general knowledge and may make a dosing error or invent an equivalent. That is why the AI pharmacist is built on the RAG principle: the agent searches for the answer in approved sources and cites what it finds. The minimum set of sources includes:
- a catalog with the INN, dosage, dosage form, dispensing conditions, and prescription status flag;
- inventory and pricing by pharmacy, updated automatically;
- official instructions for medical use in the current version;
- product cards for non-medicinal products;
- pharmacist-approved stop rules and warning templates;
- delivery, pickup, and return rules.
Also pay special attention to personal data. Health-related questions are sensitive information, and Clause 15 of Rule No. 697 requires customer data confidentiality. Practical options include anonymizing data before sending it to the model, using a Russian cloud environment, or deploying the model on your own servers. We covered how to choose between them in the article Can customer data be sent to AI?and on why RAG makes mistakes and how to measure that, in the article RAG accuracy and hallucinations.
Other AI use cases in a pharmacy chain
The AI pharmacist is the most visible use case, but not the only one. Lower-risk tasks that you can also start with:
- Product descriptions. This is where 36.6 started as well: AI drafts a short description based on the instructions and product card, and the content manager approves the text.
- Responses to reviews and common inquiries. Order status, returns, pharmacy hours — without medical advice.
- Search across internal policies for frontline pharmacy staff. Dispensing rules, storage rules, handling complaints — through a corporate knowledge base; we described how it is structured in the article corporate knowledge base with an AI agent.
- Demand and inventory forecasting. Seasonal medications and stockouts are a machine learning forecasting task; see more in the article demand forecasting with AI.
How to launch an AI pharmacist pilot
A process you can repeat in a pharmacy chain of any size:
- Choose a channel and assortment. For example, only the website chat and only over-the-counter medicines and cosmetics.
- Record baseline metrics for 2–4 weeks. How many questions come in through the online channel, how many are routine (availability, price, delivery), how much time the pharmacist spends responding, and how many inquiries go unanswered.
- Build a test set. 200–300 real customer questions with no personal data, including 50–100 “risky” ones: pregnancy, children, drug combinations, prescription medications, and signs of an emergency condition.
- Test the agent on the set before launch. For each question: the answer is based on the instructions or catalog; there are no invented equivalents or dosages; all risky questions are escalated to a pharmacist or ended with an emergency-help recommendation. The release criterion is approved by the responsible pharmacist; a reasonable threshold for risky questions is zero misses.
- Launch with oversight. During the first few weeks, the pharmacist reviews a sample of dialogues daily and reviews all escalated cases in full. Expand the assortment and channels once errors are isolated and understandable.
Pilot size and thresholds are our recommendation. For metrics for projects like this, see the article AI agent performance metrics.
Pre-launch checklist
- The catalog includes the INN, dosage, dispensing conditions, and prescription status for each item.
- Instructions are uploaded in current versions, and the agent responds with a link to them.
- Prescription, narcotic, psychotropic, and alcohol-containing products above 25% are excluded from remote scenarios in accordance with Clause 2 of Rule No. 697.
- Stop scenarios are approved by the responsible pharmacist and covered by test questions.
- For signs of an emergency condition, a standard response is configured recommending calling 112 or 103.
- Escalation to a pharmacist preserves the conversation history; the pharmacist response time is defined.
- It is decided where customer data is processed and which model is used.
- Baseline metrics and the test set have been recorded.
- A person responsible for daily spot checks of conversations has been assigned.
Frequently asked questions
What can an AI pharmacist do?
It answers questions about availability and price, shows alternatives with the same INN, explains how to take and store a product based on the official instructions, places orders, and escalates questions about contraindications, drug combinations, and special customer groups to a pharmacist.
Can AI advise on prescription medications?
No. Clause 2 of the Rules approved by Russian Government Decree No. 697 excludes prescription medications from remote retail sales. AI can only say that the medication is dispensed by prescription and suggest how to buy it in person at a pharmacy.
What must a pharmacy tell a customer for a remote order?
Under Clause 16 of Rule No. 697 — indications for use, retail price, shelf life, dispensing conditions, storage rules, and interactions with other medications. Information may be provided through the pharmacy’s website or mobile app, or through an aggregator.
Will AI replace a pharmacist?
No. AI handles routine questions and speeds up order placement, but diagnosis, treatment selection, complex drug combinations, and emergency situations remain the responsibility of the pharmacist, doctor, and emergency services.
Where should a pharmacy start with AI implementation?
Start with one channel and an over-the-counter assortment, baseline online consultation metrics, a test set with risky questions, and stop scenarios approved by the responsible pharmacist. An easier first step is AI-generated product descriptions checked by a content manager.
How the airassvet.ru team helps deploy an AI pharmacist across a pharmacy chain
If pharmacists in your online channel are drowning in routine questions, but handing consultations over to a “raw” neural network feels risky, the team airassvet.ru can:
- audit online consultations: what questions come in, which ones are routine, where stop scenarios are needed, and when to hand off to a pharmacist;
- build a RAG system using the product catalog, inventory, and official package inserts so responses are grounded in source materials;
- develop an AI agent for your website, app, and messengers, integrate it with the catalog and orders, and test stop scenarios against a question set;
- deploy the model on the pharmacy chain’s own servers if customer data cannot be sent to an external service.
The first step is to choose one process, for example answering questions about availability and how to take over-the-counter medicines in the website chat, document how it works today, what data and constraints it has, and what metric will define a successful pilot. If you have selected the process, discuss the project.
Conclusion
AI for pharmacies is most useful where the answer can be determined directly from the catalog and the official package insert: availability, price, generic alternatives by INN, directions for use, and storage. Rules No. 697 and No. 259n define what information a pharmacy must provide and exclude prescription drugs from remote sales. The 36.6 example shows the right sequence: start with the rules for when the AI pharmacist must stop and hand the customer off to a pharmacist, then add new features. Start with the OTC assortment, a test set with risky questions, and a pilot on one channel.