AI Training for PR and Communications Teams: What It Should Cover and How to Choose It

Most PR teams already use generative AI. Far fewer have been shown how to use it well. Muck Rack's State of AI in PR 2026 puts usage at 76% of practitioners, yet only 43% have received any AI training at work and just 51% work somewhere with a formal AI use policy. That gap is the reason AI training for PR and communications teams has moved from a nice-to-have to a line in the L&D budget: the tools are in the building whether or not the comms director signed them off, and the difference between a team that saves a day a week and a team that publishes a hallucinated quote comes down to skills, guardrails and habits.

This explainer sets out what good training looks like, what it has to cover, which tools belong where, the governance you cannot skip, and how to judge whether a provider is worth the money. It is written for the person who has to make that call.

Why "let the team experiment" is not a training strategy

The CIPR's State of the Profession 2024 research found that 38% of UK practitioners rated AI as the biggest challenge facing the industry, and 36% named AI as the single largest skills shortage in their organisation, ahead of every other competency. Two years on, the tools have improved faster than most teams' ability to use them, and the failure modes have become visible.

Left to informal adoption, three things tend to happen:

• Uneven capability. One account manager builds a brilliant workflow for coverage analysis; the press officer next to them pastes a whole embargoed release into a free consumer chatbot. Nobody knows which is happening where.
• Quality drift. AI-drafted pitches start to sound identical. Journalists notice. So do clients and boards, who can now spot the tell-tale rhythm of an unedited model output.
• Unmanaged risk. In March 2026, a senior Mediahuis journalist was suspended after publishing dozens of AI-generated quotes that the people named said they never made. He had summarised documents with ChatGPT, Perplexity and NotebookLM and not checked the output. If it can happen to a former editor-in-chief who writes about press standards, it can happen to a press office on deadline.

The PRCA's AI Green Paper, published in January 2026, makes the same point from the profession's side: AI is widely used, but many organisations still lack formal governance, clear policies or confidence in navigating regulation and risk. Training is how you close that gap in weeks rather than years.

What AI training for PR and communications teams should cover

A generic "Introduction to ChatGPT" course teaches people to ask a chatbot for a poem. Training for a comms team should be built around the tasks that fill the team's week. When we scope a programme, we map it against eight blocks; the half-day version keeps the three that matter most (prompting, drafting and governance) and the full day covers everything.

1. How language models actually work, and where they fail

Not a computer science lecture. Twenty minutes on what a model is doing when it predicts text, why it invents citations and quotes with total confidence, why the same prompt gives different answers, and what "context window", "temperature" and "retrieval" mean in practice. Teams that understand this stop treating the tool as a search engine and start treating it as a very fast, very literal junior colleague who needs a brief.

2. Prompting with your own material

The single biggest productivity gain comes from teaching people to brief AI the way they would brief a freelancer: role, audience, objective, format, constraints, examples. Exercises should use the team's real releases, tone-of-voice guide and past coverage, not sample text. By the end, each delegate should have a small library of reusable prompts (and, where the tool allows, saved custom configurations) for their recurring jobs.

3. Media research and journalist lists

AI is excellent at reading a journalist's last 30 articles and summarising their beat, angle and pet hates before you pitch. It is unreliable at producing a list of journalists from scratch, because it will invent names, titles and email addresses. Training should cover the split clearly: research and synthesis, yes; contact-list generation, no. It should also cover the data protection point below, because journalist contact data is personal data.

4. Drafting: releases, statements, briefing notes, Q&As and award entries

The practical skill is not "write me a press release" but structured drafting: producing a first draft from an approved fact sheet, generating 10 headline options against a target publication's style, stress-testing a statement for ambiguity, turning a 40-page report into a two-page briefing note, or building a hostile Q&A ahead of an announcement. Every exercise should end with a human edit and a fact-check against source material.

5. Monitoring, sentiment and coverage analysis

Feeding a month of coverage into a model and asking for themes, message pull-through and sentiment by outlet is one of the highest-value, lowest-risk uses of AI in PR. The training point is how to structure the data so the analysis is repeatable, and how to sanity-check the output against the raw coverage before it goes into a board pack.

6. Crisis preparation

Scenario generation, holding statement variants, stakeholder mapping and red-teaming your own messaging ("what is the worst-faith reading of this line?") are all things a model does quickly. Live crisis response is different: training should be explicit about what may and may not be run through an external tool when the story is moving and legal is involved. This sits naturally alongside conventional media training for spokespeople.

7. Measurement and reporting

AI can turn a spreadsheet of outputs into a narrative report in minutes. The risk is that it produces a polished version of vanity metrics. We build this block around the AMEC integrated evaluation framework, which we use in-house, so the team learns to ask the model to connect outputs to outcomes and business impact rather than just totting up reach.

8. Rules of the road: disclosure, embargoes, confidentiality and sign-off

The governance block is where a workshop turns into an operating model. Which tools are approved, on which accounts. What must never be entered. When AI use is disclosed to clients, journalists or the public. Who signs off AI-assisted content and what "checked" means. Delegates should leave with these written down, not implied.

ChatGPT, Claude or Copilot: which tools belong in a PR team's kit?

Tool choice matters less than most buyers expect, and the landscape shifts every quarter, so training should teach transferable habits rather than button locations. That said, the three most common choices have distinct strengths for comms work.

In practice, most in-house teams end up with one "sanctioned" general assistant plus Copilot if they are a Microsoft house, and specialist tools (media databases with AI features, monitoring platforms with built-in summarisation) layered on top. Training should be delivered in the tools your team will actually be allowed to use, which is why we confirm the approved stack before we design a session.

The safety and governance points you cannot skip

Personal data and UK GDPR

Journalist names, emails, phone numbers and notes about their preferences are personal data. Pasting them into a third-party AI tool is processing, and you need a lawful basis and appropriate contractual safeguards with the provider. The ICO's consultation series on generative AI and data protection sets out its thinking on lawful basis, purpose limitation, accuracy and individual rights. The practical training outcome is a short list of "never enter" categories (personal data, unpublished financials, embargoed material, anything under NDA) and an understanding of which tools, on which licence, are covered by your organisation's data processing agreements. Muck Rack's 2026 data suggests 69% of PR professionals already avoid entering personally identifiable or proprietary information; the aim is to make that 100% in your team, by rule rather than instinct.

Hallucination and verification

Every AI-assisted output that leaves the team needs a named human who has checked facts, quotes, figures and names against source material. Training should make this concrete: a verification checklist, a rule that any statistic or quote generated by a model is treated as unverified until traced to a primary source, and a habit of asking the model to show its working and its sources so the check is quick.

Disclosure and professional standards

The PRCA Green Paper frames transparency, accountability and human judgement as core to responsible use, and the CIPR's guidance points the same way. A sensible policy distinguishes between AI as a drafting aid (generally no disclosure needed, as with a spellchecker) and AI-generated content presented as original human output, such as a ghost-written byline or a synthetic image, where clients and audiences have a reasonable expectation of knowing. Where you draw the line is a judgement call; that you draw it, in writing, is not.

Brand voice and quality

The commercial risk that gets least attention is dilution. When every agency and press office uses the same models with the same default settings, output converges. Training should include a deliberate module on getting the model to write in your voice, using your tone guide and best past work as reference, and on spotting the tics (reflexive triplets, "in today's fast-paced world", balanced-but-empty conclusions) that mark unedited AI text. Our guide to AI ethics for creative teams covers the authenticity and rights questions in more depth.

Writing the AI policy: what it needs to contain

Half of PR professionals now work under a formal AI use policy, up from 21% in 2024, according to Muck Rack. If yours does not yet exist, the most efficient way to write it is in a facilitated session with the people who will live under it, rather than handing it to legal alone. A workable comms-team AI policy covers:

1. Approved tools and account types, with a named owner responsible for reviewing the list quarterly.
2. Prohibited inputs: personal data, confidential client information, embargoed material, financial results, anything subject to legal privilege.
3. Permitted uses by task, with the level of human review required for each (draft only, draft and edit, edit and verify, prohibited).
4. Disclosure rules for clients, journalists, the public and, for agencies, in contracts and proposals.
5. Sign-off: who approves AI-assisted content before publication and what "approved" means.
6. Record-keeping: where prompts, configurations and key outputs are stored so work is reproducible and auditable.
7. Incident handling: what happens when something goes wrong, including who is told and how quickly.
8. Training and review: who gets trained, how often, and when the policy is next reviewed.

We run this as a standalone AI policy workshop: a half-day in which the team drafts its own policy and leaves with a document ready for legal and IT to review, rather than a template nobody owns.

Choosing a format: half-day, full day or a programme

The right format depends on where the team is starting from and how much change you need to land.

Whatever the length, insist on hands-on. Delegates should be at keyboards, in the approved tools, working on their own releases, media lists and coverage reports. A slide deck about AI is not AI training.

How to judge a training provider

The market for AI training is crowded with generalist technology trainers and productivity consultants. For a comms team, ask a provider these questions before you book:

• Do they practise PR? A trainer who has never pitched a national journalist or handled a crisis will teach you to write faster, not to do PR better. Ask what they were working on last week.
• Will they use your material? If the exercises are built on sample content, the learning will not transfer. Good providers ask for your releases, tone guide and coverage before the session.
• Do they cover governance, or just tricks? If the agenda has no session on data, disclosure and sign-off, you will be paying for the risk later.
• Are they tool-agnostic? Be cautious of anyone who insists a single platform is the answer, or who is selling their own software alongside the training.
• How do they measure impact? Time saved per task, quality assessed by editors, and reduction in rework are all measurable. Ask how they help you baseline and follow up.
• Are they accountable to professional standards? CIPR and PRCA membership, chartered status and a record of working within the profession's codes matter when the subject is ethics as much as efficiency.

How we deliver AI training at Prohibition

Our AI training for PR and communications is run by practising consultants from an integrated agency that uses these tools daily across media relations, social, content and video for consumer and B2B clients. Sessions are designed and led by practising consultants, including our founder Chris Norton, a Chartered PR Practitioner, long-standing CIPR member and co-author of the CIPR social media handbook Share This Too and host of the UK's number one marketing podcast in the Apple charts.

Our bookable AI course for PR teams is private to your organisation, delivered at your office, at our Leeds studio or live online, and built around the eight blocks above using your own releases, media lists and coverage reports. No prior AI experience is assumed. Published pricing is £1,200 plus VAT for a half-day or £1,800 plus VAT for a full day, both for up to six people, with additional delegates at £250 each; we can usually deliver within three to four weeks of scoping. It is built for in-house communications teams and press offices across universities, charities, housing associations, NHS trusts, manufacturers, local government and consumer brands, as well as agency teams.

Where a team needs more than a workshop, we scope a programme: a workflow audit to find the bottlenecks worth automating, tool mapping against your approved stack, custom prompts and configurations, the workshop itself and a written AI framework to take away. Specialist add-ons include Copilot training, AI for social media content, the AI policy workshop and AI misinformation and deepfake response training for reputation teams.

Measuring whether the training worked

Treat training like any other communications investment and set the measures before the session, not after. The ones we recommend:

• Time on recurring tasks. Baseline how long a first-draft release, a coverage report or a journalist background brief takes today; re-measure at 30 and 90 days.
• Quality at first review. Ask editors or account directors to score first drafts before and after on a simple rubric. Faster is worthless if rework rises.
• Adoption breadth. How many of the team are using approved tools for at least three task types a fortnight after the session, versus one or two enthusiasts.
• Policy compliance. Zero use of unsanctioned accounts for client work; every AI-assisted output with a named verifier.
• Outcome measures. Over a longer horizon, whether freed capacity has gone into work that moves reputation and commercial metrics, which is where the AMEC framework and our Create, Engage, Convert approach come in. We measure business impact, not vanity metrics, and the same discipline applies to training.

Frequently asked questions

How long does AI training for a PR team take?

A focused half-day (around three hours) is enough to standardise prompting, drafting and governance for a team that is already experimenting. A full day (around six hours) covers media research, monitoring, crisis preparation and measurement as well. Larger departments usually benefit from a short programme with a workflow audit before the workshop and a written framework afterwards.

Should we train on ChatGPT, Claude or Copilot?

Train in the tools your organisation has approved and licensed. Microsoft 365 organisations often start with Copilot because it keeps data inside their tenancy; teams wanting a general assistant with shareable custom configurations tend to choose a ChatGPT business workspace or Claude for an organisation. The prompting and verification skills transfer between all three.

Is it safe to put journalist contact details into an AI tool?

Journalist contact data is personal data under UK GDPR, so you need a lawful basis and a provider covered by an appropriate data processing agreement. Most teams treat contact lists as a prohibited input for general-purpose chatbots and use AI only to research published work, which involves no personal data beyond the byline.

Do we need an AI policy before we train the team?

No, and it often works better the other way round. A team that has been trained understands the risks well enough to write a policy it will actually follow. Many organisations combine a training day with a facilitated policy session so both land together.

Will journalists know if our pitches are written with AI?

If they are unedited, yes: journalists receive hundreds of pitches and recognise generic model output quickly. Used well, AI does the research and structuring and a human writes the final pitch in a specific, personal voice. The goal of training is to make AI invisible in the output because the human judgement is still there.

Where to start

If your team is already using AI without a shared method or a written policy, you have the problem this training solves, and every month of informal adoption makes habits harder to change. Start by auditing which tools are actually in use and on which accounts, decide your approved stack with IT and legal, then book training built on your own material rather than a generic course. Well-designed AI training for PR and communications teams pays for itself in the first few weeks of recovered time; the larger return is a team that can use these tools with confidence, in public, without putting your reputation on the line. If you would like to talk through what a session for your team would look like, get in touch and we will scope it with you.