AI for comms: a starter guide to what you need to know in October 2026

Q4 2026 edition. Dated facts were checked on 21, 23 and 27 October 2026.

The Brief

• Most of your peers already use it. Four in five UK PR practitioners (82%) use AI tools at least occasionally, and 76% of the PR professionals in Muck Rack’s survey use generative AI; adoption has levelled off, so what separates teams now is whether they use it safely and well.
• Start with the assistant your organisation licenses, under its business terms. Get your footing there, then try other leading models, keeping free consumer versions for material that is already public.
• Pick one task this week, run it in that tool, and check every line before it leaves you. Then ask the tool for a better prompt for next time. That habit is worth more than any tool choice.

The Story

First up, let me know if this sounds familiar: a team meeting, a conversation about workload, and someone admits, a little sheepishly, that they have been using ChatGPT for months. Nobody has ever told them whether that is allowed. A couple of colleagues nod. The leader in the room makes a note to find out what the policy says, and discovers there isn’t one – or if there is, that it’s already out of date.

That room is the profession in 2026. The CIPR’s State of the Profession 2026, from 738 responses gathered in April and May, found that 82% of practitioners use AI tools at least occasionally, 45% regularly. A quarter of those users have no formal policy or guidance, a third have used tools their organisation never approved, and 40% of practitioners have had no AI training. Ben Verinder’s reading of the report is worth checking out: the quiet majority is using AI; the question is how.

This guide is for anyone in that room: where things stand, the main assistants and what each is good for in comms, five things to try this week, the words you keep hearing, and where to go next. I don’t have all the answers. Nobody does. But the ground rules are clearer than the noise suggests.

1. Where things stand

Beyond the UK, Muck Rack’s State of AI in PR 2026 (564 PR professionals, surveyed in December 2025) puts generative AI use at 76%, largely unchanged from the year before.

The people using it say it helps: 82% say it improves the quality of their work. Where they use it is telling: editing and refining (86%), research (76%), writing (74%) and strategy (68%). The worry has moved on from whether to use it: 77% worry that heavy use will stop the next generation learning the fundamentals. Only 12% of AI users have adopted agents (expect that to climb now that Meta’s Muse and OpenAI’s dots are out, though UK access to each is still limited), and 89% of practitioners would be more comfortable with an agent if a person had to approve what it does.

Charities are no slower. The Charity Digital Skills Report 2026 found 79% of charities routinely using AI (76% a year earlier), while only 36% of small charities have an AI policy, 35% cite a lack of trust in the tools (15% last year) and 44% take no action to manage the risk. Zoe Amar, the report’s co-author, is our guest on 18 November.

Two regulatory lines matter for a comms team. In the EU, the AI Act’s transparency duties have applied since 2 August 2026: people must be told when they are dealing with an AI system, AI-generated content must carry machine-readable marking (systems already on the market have until 2 December 2026), and anyone publishing a deepfake must say so. The Digital Omnibus pushed most high-risk deadlines to December 2027 (August 2028 for AI built into regulated products) and left those transparency duties alone, and the Act reaches organisations outside the EU when their AI output is used inside it. In the UK there is no AI-specific act. UK GDPR applies to any personal data you put into a tool, and the ICO’s AI and data protection guidance is the reference your data protection lead will reach for.

2. The main assistants, and what each is good for in comms

Four assistants cover most comms teams. The table says what each is good for and what to watch, with the date each row was checked, because these facts move weekly. One disclosure first: I’ve drafted this article with assistance from Claude, and Claude also finished first in our Comms AI Pentathlon, a blind test in which the assistants judged each other, with the self-preference caveat that implies. Comms With AI has no commercial or affiliate relationship with any vendor.

Assistant	Where you’ll meet it	Good for in comms	Watch for	Checked on
ChatGPT (OpenAI)	The one your team likely already uses on their phones. GPT-6 is now the default: Sol for Plus, Pro, Business and Enterprise from 7 October; the lighter Luna for Free and Go from 8 October. Answers can now include charts, diagrams and small interactive tools.	Deep online research, first drafts, turning numbers into charts. Third in the Pentathlon (39.4 of 50).	Custom GPTs retire on 11 December 2026; no new ones since 26 October. If your team relies on one, migrate it to a plugin now. Which model your plan gets.	27 Oct 2026
Claude (Anthropic)	Less likely to be the default at work; often the one people actively choose. Opus 5.5 and Sonnet 5.5 arrived in late September, Haiku 5.5 on 7 October; Docs, Slides and Design are now on every plan, including Free.	Long documents, writing to a house voice, Projects for a standing brief, presentations and design work. First in the Pentathlon (42.8).	It can eat up usage plans quickly – especially the flagship Fable 5.1 model.	27 Oct 2026
Gemini (Google)	Inside Google Workspace: Gmail, Docs, Meet notes. Since 9 October, personal accounts without a subscription get “Auto” model selection, which sends most prompts to Flash-Lite. Google’s notice covers personal accounts; on a managed Workspace account, ask your admin what you get. Gems are giving way to skills in Workspace, with Gems usable until at least March 2027.	Teams on Google Workspace: Meet notes that draft the follow-up, Gmail drafting. Seventh in the Pentathlon (35.0).	The free tier now gets the lightest model most of the time. The Gemini you tried in September is not the one you get today.	27 Oct 2026
Microsoft Copilot	Inside Microsoft 365: Outlook, Teams, and the Office apps if your organisation pays. Since 15 April, free Copilot Chat has left Word, Excel and PowerPoint for large organisations (Microsoft’s own notice puts the line at 2,000 users; smaller ones get a restricted version); it stays in Outlook. The paid Microsoft 365 Copilot add-on lists at £23.10 a user a month on an annual commitment, or £16.10 as Copilot Business for organisations up to 300 users, ex VAT, as one UK reseller recorded them on 25 September; check Microsoft’s UK page before you budget.	Teams on Microsoft 365, working inside Outlook and Teams with their own files. Sixth in the Pentathlon (35.3).	What your Microsoft licence includes. Ask IT before assuming you have the paid version.	27 Oct 2026

Others worth knowing. Perplexity, for search with cited sources (handy for checking facts). Siri AI on iOS 27, in beta since 14 September, last in the Pentathlon (24.1), and it refused a crisis brief outright: consumer-only for now, worth watching for AI visibility once it gets a stronger model. GLM, DeepSeek and Kimi scored near the top of the Pentathlon and are Chinese-developed, with DeepSeek and Kimi storing data in China: strong results, though the wrong home for staff or client material.

Which kind of tool suits which task (any tool for public material, your enterprise tool for internal work, an approved tool or no AI for the high-risk stuff) is set out on Using AI safely, so I won’t repeat it. Hands-on reviews of individual tools are in the tools directory.

3. What you can do with them this week

Five ordinary days, one task each. These are the same five as the comms week on /learn/. Pick one.

Why these five: editing, research, writing and strategy are where practitioners already use AI, in that order (Muck Rack 2026). One habit to add from the first day: log what went in and what came out. (Mine arrives by scheduled task each evening, using the Daily AI Use Wrap-Up prompt; more below.) The AI Use Review and Risk Log is a page, and it is the thing an auditor, a client or a sceptical director will ask for.

What I actually do

My own week primarily runs on Claude: most of the Faur and Comms With AI work, and the drafting process for every Applied piece, in dedicated Projects that hold the house style. The Projects now live on my own computer, with Claude given gated access to the files. A year in I trust it, but start with very limited access and widen it.

ChatGPT gets used at multiple points every day, often to double-check Claude’s work rather than to start anything. Gemini now and then, usually for email and working with files in Drive for sharing with others. Copilot when a client has it, because the useful test is their tool, not mine! Siri AI on my own phone since the beta, for personal use. I don’t generally use DeepSeek, GLM or Kimi – even though all three finished in the Pentathlon’s top five – for the data reason above.

What never goes in. Anything confidential, especially anything to do with clients, unless it’s a trusted tool where I’ve verified how my data is used.

How the use log works. It arrives by email at the end of each day, so the next morning I can see what I did and achieved, which answers the eternal question of what the hell I actually did all day. It’s a concise bullet list, in a format I designed, and it closes with the one standout highlight and the one nagging thing to address the next day. Here’s the prompt, if you’d like to borrow it. Paste it in at the end of the day with your notes, or set it up as a scheduled task if your assistant can run one. The Daily AI Use Wrap-Up in the library has the scheduled-task version, a Friday roll-up and a worked example:

At the end of my working day, put together my AI use log from today’s work [or: from the notes below].

1. A concise bullet list of what I did and achieved today, grouped by client or project, one line each.
2. For each task where I used an AI tool: which tool, what went in (in general terms, with no confidential detail), what came out, and whether I checked it before it was used.
3. The one standout highlight of the day.
4. The one nagging thing to address tomorrow.

Keep it to one screen, in plain English. Send it to me with the subject line “Use log: [today’s date]”.

How I get my thoughts down. I often use Wispr Flow for voice dictation, so I can free-flow with my thoughts. I’m on the free plan, so sometimes I run out of words for the week, and then I switch to Granola, or use a combination of the two.

One thing that goes wrong. Voice dictation still isn’t perfect, especially for a Scottish accent, so I find myself correcting some weird bungles.

4. A short glossary

The words you keep hearing, in plain English. Dated facts in the notes were checked on 21 October 2026.

Missing one? Tell me and it goes in the next edition.

5. Where to go next on the site

• Using AI safely: the ground rules, what never goes in, and the paste check.
• Start here: the library, by the problem you are trying to solve.
• AI Readiness: a scorecard to run with your team.
• The Briefing: what changed this week, landing Fridays.
• Events: Billy Partridge on 28 October, Zoe Amar on 18 November, both free.
• Help for your team, when you want someone in the room.

The Practice: what you can put into practice today

1. Find out which assistant your organisation licenses, and on what terms. Ask IT or whoever holds the Microsoft or Google contract. If the answer is “none”, that is your first finding, and the AI Use Policy template is where to start the conversation.
2. Run Monday’s task in that tool. Coverage you already have, one page out, every line checked by you before it goes anywhere. Then a new task each day, and ask the tool for a better prompt each time.
3. Write down what you put in. One line in the log for each use. The habit is the point.

Every Applied piece follows the same shape (The Brief, The Story, The Practice) and is made the same way: my argument and examples, drafted with Claude, every claim checked against its source, and a final edit by me. This one: 70 claims checked, 10 corrected or reworded before publication.

Changelog

• Q4 2026 edition: first published 27 October 2026. Next scheduled refresh: January 2027.