Sohail Kasmani. Brand and marketing strategist in London: pharma brand management at Cipla and Alembic, now marketing strategy and AI adoption at ACTAsia. MBA, Brunel University London.
AI in pharma marketing: what it does and what UK rules allow
AI in pharma marketing mostly means three jobs at the moment: drafting and pre-checking promotional content before medical, legal and regulatory (MLR) review, helping field teams prepare for and write up their calls, and listening to what doctors and patients say online. In the UK all of it sits under the same ABPI Code as everything else, and the regulator has said in writing that AI changes nothing about who is responsible.
The rule that matters most is Clause 8.1. The Prescription Medicines Code of Practice Authority (PMCPA), which runs the ABPI Code, published guidance in June 2025 saying the requirements of Clause 8.1 cannot be met by AI alone, because promotional material must not be issued until its final form has been certified by a nominated signatory. A model can read your leavepiece, but it can’t sign it.
What is AI actually doing in pharma marketing?
The most detailed public example is still Pfizer’s. In February 2024 Digiday reported that Pfizer had built a generative AI platform called Charlie (after the company’s co-founder, Charles Pfizer) with Publicis Groupe, on top of Publicis’s own Marcel platform. It drafts digital media, emails and presentations for sales teams, helps with medical articles, and is used by hundreds of people in Pfizer’s central marketing team and thousands across its brands, plus agency partners. The detail I find most useful is how it handles review: Bill Worple, Pfizer’s VP of customer engagement platforms and technology, described a red, yellow and green labelling system, where a headline the company has used many times needs less scrutiny than a new medical claim. That is the right instinct, because most of the cost in pharma content is review, and the drafting was never the slow part.
The software vendors have followed. In December 2025 Veeva released AI agents for its PromoMats content management and Vault CRM products. Its Quick Check Agent scans content against editorial, brand, market, channel and compliance guidelines to catch issues before MLR review, and its Free Text Agent flags potential problems in reps’ call notes. There’s also a Pre-call Agent that suggests actions from data, content and activity before a rep visits a doctor. None of these replace the signatory; they try to make sure the signatory spends their afternoon on real questions rather than on a missing reference and a logo in the wrong colour.
Field teams are the third area, and the UK already has a PMCPA case on it. In AUTH/3820/9/23, an anonymous complainant objected to Daiichi Sankyo piloting an AI tool that recorded and analysed calls between its representatives and health professionals. The case, completed in October 2024, found no breach: the tool analysed the representative’s voice for training, the audio was deleted after extraction, health professionals gave verbal consent, and five doctors who refused suffered no consequences. Consent and a clear, non-promotional purpose carried the case.
How many pharma companies are really using generative AI?
Fewer than the conference agendas suggest. ZS, the life sciences consultancy, surveyed 127 technology executives at pharma, biotech and life sciences companies through Harris Poll in July and August 2024. Around a third of employees in each department had company-approved access to generative AI tools, but only around 10% of those with access were using them consistently every week. Meanwhile 93% expected to increase investment in data, digital and AI in 2025. So we’ve bought the licences, held the town hall, and most people with access are still writing their emails the old way, which is not a technology problem.
The reason this matters for marketers is that low use usually means the tool doesn’t fit the workflow, and in pharma the workflow is shaped by review. A brand team that can draft a campaign in a day and then waits three weeks for certification hasn’t saved much time. I’d look at where your content actually gets stuck before choosing anything.
What do the UK rules say about AI in pharma marketing?
The ABPI Code and the PMCPA
The PMCPA’s June 2025 guidance is short. The Code’s requirements are unchanged whether or not AI is used, and the company keeps full responsibility. Companies should have policies setting out standards, train staff on them, and carry out a risk assessment tailored to their own use, because laws beyond the Code apply. The Panel may look at a company’s AI governance when it considers a complaint, which is the sentence I’d show any team that has been using a public chatbot on draft materials without telling anyone. The PMCPA also points companies to the AI ethics toolkit published by IFPMA, the international pharma trade body, for internal assessment. For the wider rules on promotion, I’ve written about the ABPI Code and MHRA requirements separately.
Patient-facing chatbots and Clause 26
Under Clause 26 of the 2024 Code, prescription only medicines must not be advertised to the public, and information about them must be factual, balanced, must not raise unfounded hopes, and must not encourage people to ask their doctor for a specific medicine. The Code also says requests from individual members of the public for advice on personal medical matters must be refused and the person told to consult their own doctor. A large language model’s whole personality is wanting to be helpful, so a patient chatbot for a UK pharma company has to be built to do the thing it least wants to do, which is decline politely and send people to their GP.
When a chatbot becomes a medical device
The Medicines and Healthcare products Regulatory Agency (MHRA) set out its view in a 2023 blog by Johan Ordish: large language models developed for, or adapted or directed towards, medical purposes can qualify as medical devices, and a developer’s claims about medical use are what bring them into scope. General-purpose tools without those claims are unlikely to qualify. So a disease awareness chatbot that starts suggesting what a patient’s symptoms might mean has drifted from marketing into territory that needs evidence it is safe and performs as intended.
Data protection and inferred health data
The Information Commissioner’s Office (ICO) guidance on special category data, updated in April 2024, says that if you process data to infer something about someone’s health, or treat them differently because of an inferred health condition, you are processing special category data regardless of how confident you are in the inference. AI targeting is very good at inference, so a model that decides someone browsing late at night is probably a migraine sufferer has created health data, and you need an Article 9 condition for it.
Adverse events from AI social listening
The ABPI’s guidance on adverse events from digital media (version 2.0, April 2013) says companies must regularly screen the digital media they manage for suspected adverse reactions, and suggests considering daily monitoring for the duration of a project. There’s no legal requirement to monitor sites the company doesn’t sponsor, but if you choose to listen there and find adverse events, you must report them, and employees should pass them to pharmacovigilance within one business day. An AI listening tool that surfaces 400 posts about side effects is also a tool that creates 400 pieces of work for the safety team, so I’d agree the process with pharmacovigilance before the tool goes live, not after the first weekly report lands. There’s more on this in my piece on pharma social media in the UK.
Content seen in the EU
From 2 August 2026, Article 50 of the EU AI Act requires deployers to label deepfakes and, in some cases, AI-generated text. Law firm Lewis Silkin’s summary of the rules notes that they reach UK companies that foresee, direct or authorise distribution of AI content into the EU, and that text published to inform the public on matters of public interest without human review is caught. Their reading is that routine AI-drafted ad copy will generally sit outside that, provided it doesn’t include claims about matters such as health. Fines go up to 15 million euros or 3% of global annual turnover. For a UK affiliate with a European parent, the practical answer is that health content gets a human reviewer, which pharma does anyway, and that anything synthetic showing a person gets a label.
Healthcare marketing outside prescription medicines
For over-the-counter products, private clinics, health apps and anyone else under the CAP Code, the Advertising Standards Authority’s sister body CAP published four points on AI and deepfakes in June 2026. The Code is media-neutral, so AI-made ads are judged like any other ad. In the Stripe and Stare ruling, the advertiser was still held primarily responsible for ads that Google’s systems had generated automatically. If your ad platform writes the headline, you still own the claim.
Where does AI earn its place in a UK pharma marketing team?
| Use case | What AI does well | Where it breaks | UK rule to check first |
|---|---|---|---|
| First drafts of HCP emails and modular content | Variants from approved claims, fast | Invents claims or references if not restricted to approved content | Clause 8.1 certification |
| Pre-MLR quality check | Missing references, prescribing information, brand rules | Can’t judge whether a claim is balanced in context | Clause 8.1, company SOPs |
| Pre-call planning and call notes | Summarising history, flagging risky wording in notes | Suggests off-label topics if trained on loose data | Consent, transparency (see AUTH/3820/9/23) |
| Social listening | Volume, themes, early signals | Every adverse event it finds becomes a reporting task | ABPI pharmacovigilance guidance |
| Patient-facing chatbot | Answering approved, non-personal questions | Gives personal medical advice because it wants to help | Clause 26, MHRA device rules |
| Audience targeting | Finding lookalike audiences | Inferring health conditions | ICO special category data |
The pattern is that AI is safest where it works on material a signatory has already approved, and riskiest where it generates something new in front of a patient. I worked in pharma brand management at Cipla and Alembic, and the first row is where I’d start, because approved claims are the most valuable data a brand team owns and they are usually sitting in a shared drive with names like FINAL_v7_approved_USE_THIS.
How should pharma marketers prepare for doctors asking AI instead of reps?
This is the change I’d watch most closely. An IQVIA analysis from March 2026 cites a survey by EPG Health (an IQVIA company) from March and April 2025 in which 54% of health professionals were already using generative AI tools to access scientific information, rising to 75% among those born from 1990 onwards, against 48% of those born earlier. 38% rated generative AI as a critical or very important source of scientific information.
IQVIA’s argument is that influence will increasingly depend on the credibility of the evidence these systems draw on, which means publications, guideline inclusion and real-world data, and that field teams will shift towards fewer, more consultative conversations. I agree with the direction. What worries me is that our certified detail aid won’t be what an AI tool reads, while an old conference poster with a subgroup analysis might be. So the brand’s job widens: make sure the public evidence about your medicine is accurate, current and findable, because the doctor may meet it through a chatbot before they ever meet your rep. My piece on how pharma engages HCPs digitally in the UK covers the channels this sits alongside.
A step-by-step plan for using AI in a UK pharma brand team
- List every place AI is already being used, including the unofficial ones. Someone in your team has pasted a draft into a public chatbot, and the PMCPA has said the Panel may look at your governance.
- Write a short policy and train people on it, as the PMCPA guidance expects, covering which tools are approved and what data can go into them.
- Do a risk assessment per use case, including UK GDPR. Any targeting or listening that infers health status needs an Article 9 condition.
- Start with drafting from approved claims and a pre-MLR check. Keep the nominated signatory as the final step and measure review cycle time before and after.
- Agree the adverse event process with pharmacovigilance before any listening tool goes live, including who reads the alerts at the weekend.
- For anything patient-facing, test that it refuses personal medical advice and check with regulatory whether its claims make it a medical device.
- Audit what public AI tools say about your medicine and your disease area every quarter, and fix the evidence they draw on.
The open question for me is the last step: if doctors increasingly meet your brand through an AI answer you didn’t write and can’t certify, who in the company owns making sure that answer is right, medical or marketing?
FAQs
Can AI review pharma promotional material under the ABPI Code?
AI can help, but it can’t do the job on its own. The PMCPA’s June 2025 guidance says Clause 8.1 cannot be satisfied by AI alone, because promotional material must be certified in its final form by a nominated signatory. Companies keep full responsibility, should have policies and training for AI use, and the Panel may consider their AI governance when reviewing a complaint.
Is AI used in healthcare marketing regulated in the UK?
There is no separate AI rulebook for healthcare marketing. Existing rules apply in full: the ABPI Code for prescription medicines, the CAP Code for other health advertising, UK GDPR for data, and MHRA medical device rules if a tool makes medical claims. CAP’s June 2026 guidance confirms the advertiser stays responsible even when an ad platform generates the ad automatically.
Can a pharma company use a chatbot to answer patient questions?
Yes, within limits. Under Clause 26 of the ABPI Code, information about prescription medicines given to the public must be factual and balanced, and requests for advice on personal medical matters must be refused with the person told to see their own doctor. A chatbot that starts interpreting symptoms may also become a medical device in the MHRA’s view.
Does AI social listening create pharmacovigilance obligations?
It can. ABPI guidance says companies must regularly screen digital media they manage for suspected adverse reactions. There is no legal duty to monitor sites the company doesn’t sponsor, but if a company listens there and finds adverse events, it must report them, with employees passing them to pharmacovigilance within one business day of receipt.
How many health professionals use generative AI for medical information?
According to an EPG Health survey from March and April 2025, cited by IQVIA in March 2026, 54% of health professionals were already using generative AI tools to access scientific information. Use was 75% among those born from 1990 onwards and 48% among those born before 1990, and 38% rated generative AI as a critical or very important source.
Sources
- PMCPA, Can the PMCPA give companies any guidance about using artificial intelligence (AI) to assist in the review of promotional materials and activities covered by the ABPI Code? (last revised June 2025)
- PMCPA, 2024 ABPI Code of Practice, Clause 26: Relations with the Public, Including Patients and Journalists (accessed October 2026)
- PMCPA, AUTH/3820/9/23: Complainant v Daiichi Sankyo (completed 11 October 2024)
- Digiday, With ‘Charlie,’ Pfizer is building a new generative AI platform for pharma marketing (22 February 2024)
- Veeva Systems, Veeva AI Agents Now Available to Increase Productivity and Customer Centricity (3 December 2025)
- ZS, 2025 AI trends: Life sciences leaders on data, digital and AI (21 November 2024)
- IQVIA, The evolution of pharma engagement as AI becomes a front door to medical information (March 2026)
- ABPI, Guidance notes on the management of adverse events and product complaints from digital media, version 2.0 (8 April 2013)
- ICO, What is special category data? (updated 9 April 2024)
- MHRA MedRegs blog, Johan Ordish, Large Language Models and software as a medical device (3 March 2023)
- Lewis Silkin, The new AI labelling rules for “deployers” in the advertising supply chain (31 July 2026)
- ASA and CAP, AI and Deepfakes: Four Things Advertisers Need to Know Before They Hit “Run” (11 June 2026)