For the Guild of Fine Food · Ahead of 17 July

AI Policy and Data Security.

Everything the Guild needs in one place: your AI policy ready to publish, the one-pager for the wall, and a glanceable read on Claude's features and where the edges sit for the Guild. Designed to be read in 10 minutes, or fed into Copilot or Claude if you want to interrogate any of it.

Prepared forTortie Farrand & the Guild team
ByMary Kemp, AI Potential
Plans in scopeClaude Team plan · paid Copilot

The AI policy, ready to publish.

Short, plain and confident. It lives on your website, linked from your privacy policy. The highlighted parts are yours to fill.

For your website · linked from your privacy policy

Artificial Intelligence at the Guild of Fine Food

How we use AI, in service of the fine food community.

Policy owner: [role, e.g. Operations Director]  ·  Last reviewed: July 2026  ·  Reviewed every six months

For nearly forty years, the Guild of Fine Food has championed independent food and drink. Trust is the foundation of everything we do. This page sets out exactly how we use AI, for our team and for everyone we serve.

Our principles

Human-led. People make every decision that matters at the Guild. AI supports our work; our team directs it, checks it and takes responsibility for it.

Secure. We use business-grade AI tools on accounts the Guild controls. Our data stays ours and is never used to train AI models.

Transparent. We are open about where AI supports our work.

Accountable. One person at the Guild owns this policy, the whole team is trained, and we review it every six months.

How we use AI

AI helps our team with research, drafting and administration, so more of our time goes to tasting, judging, publishing, training and the people we serve. Every tool is approved before anyone uses it, and nothing connects to our systems without a yes from our technical lead.

Your data

The personal information of our members, producers, delegates and readers stays within our own systems, handled in line with UK GDPR and our [link to privacy policy].

Our awards

Every product entered into Great Taste and the World Cheese Awards is tasted, discussed and judged entirely by human experts. That is permanent. Technology helps with the paperwork, not the palate.

Our publishing

The judgement and recommendations in Fine Food Digest and across our publishing come from our editorial team. Where AI has played a real part in something we publish, we say so.

Questions

Ask us at [email address] and a person will answer.

The water cooler version.

One A4 page, up by the coffee machine and around the office. Add the names and print it.

AI at the Guild
The short version. Our full AI policy has the detail; this is how we work day to day.

Good to go

  • Use Copilot for any work task. Be signed in to your Guild account and check the green shield has its tick. Then use it as much as you like.
  • Use Claude if you have a seat on the Guild's Team plan and you are signed in to it.
  • Ask AI to draft, summarise, brainstorm and prepare. You review everything before it goes anywhere.
  • Stuck? Ask Copilot first. "Have you asked Copilot?" is the house question.
  • Share what works. Your wins become the whole team's wins.

Ask first

  • New tools, plugins or connections. Speak to [technical lead] before connecting anything to anything.
  • Sensitive information. Data from Allen, member and producer personal details, judging scores before results day, and Guild financials stay inside our own systems. If in doubt, ask [policy owner].
  • Guild work happens on Guild accounts. Personal and free AI accounts are for personal life.
AI drafts. You decide.

Spotted something unexpected? Tell [first call] straight away and we sort it together.

Finish both documents in one sitting

Copy the prompt, paste it into Copilot or Claude with both documents underneath, and it will interview you to fill every gap. Send me the results and I'll give them a final once-over before they go live.

You are helping the Guild of Fine Food finalise two AI documents pasted below: a public AI policy for our website and a one-page version for our office walls. We run Great Taste, the World Cheese Awards and Fine Food Digest.
Interview me one question at a time to fill every bracketed placeholder: the policy owner role, the technical lead, the first call for anything unexpected, the privacy policy link, and the contact email. Then ask whether any wording needs adjusting to match how the Guild speaks.
When the interview is complete, return both finished documents with every placeholder resolved. Keep the structure, tone and length exactly as they are, and keep every line positive and confident.
Here are the documents:
[PASTE BOTH DOCUMENTS HERE]

What the Claude Team plan means for your data.

This is the bit a lot of teams worry about and a lot of vendors overcomplicate. In plain English.

The Guild's Claude seats sit on the Team plan, the business-grade version. That changes the security picture significantly compared with free or personal accounts. Three things to know up front.

Your data isn't used to train Claude

Nothing the team types into Claude on the Team plan is used to train Anthropic's models. Your prompts, files and outputs stay yours.

Enterprise-level security

Anthropic holds SOC 2 Type II and ISO 27001 certifications, with encryption in transit and at rest. Their trust portal is public, so you can check it yourselves.

Admin controls

Your admins manage the seats, see workspace-level usage and stay in charge of who can do what. The Guild holds the keys.

So for everyday work, the team can paste in real material the same way they'd put it in a Word document on a work laptop. The Team plan does the heavy lifting on data security. Copilot on your paid accounts gives you the same standard inside the Microsoft world.

The places that stay inside your own systems are exactly the ones on the wall poster, your red list:

If a task seems to need red-list data, talk to [policy owner] first. There is usually a safe way to do it. The rest of this page is about the features, not the platform.

Two ways in, and what the verdicts mean.

The 10-minute read

Start with the at-a-glance table below. Each feature has a verdict against the four-level key. Then read the feature cards for the ones that matter to you. Each card has the upside, the watch-out, and a specific verdict for the Guild.

Feed it into Claude

Copy the page (or just one section) and paste it into Claude with the prompt below. Then ask things like "for the marketing team specifically, where should the edges sit?" or "which of these would you enable first?".

I'm pasting in a security report on Claude's features written for the Guild of Fine Food, the organisation behind Great Taste, the World Cheese Awards and Fine Food Digest. We work with member and producer data, pre-announcement judging results and a CRM called Allen that runs our business. We're on the Claude Team plan with paid Copilot alongside. I want you to act as an experienced AI risk advisor and help me think through how this applies to my specific work.
After I paste the report, ask me one question at a time about my role and the kind of work I do, then give me a tailored "what to use freely, what to use with care, what to hold off on" recommendation for my work specifically.
Here's the report:
[PASTE THE REPORT CONTENT HERE]

What the verdicts mean

Every feature carries one of four verdicts. Here's what each one means in practice.

Go

Turn this on for everyone with a seat. No special caveats. Default safe under the Team plan.

Use with care

Useful, but has a specific consideration, usually about who sees what or what to decide before enabling. Worth a 5-minute conversation.

Essential governance

Set up before anyone logs in: admins, expectations, the policy on the wall. The foundation everything else stands on.

Hold for now

Still maturing or more than the Guild needs right now. Revisit once the day-one value is banked.

The Claude features that matter.

Scan view. Detailed cards follow.

Feature What it does in one line Verdict
Claude (chat)Standard one-to-one conversation with ClaudeGO
ProjectsWorkspaces with persistent knowledge, shareable across the teamGO
ArtifactsThe side panel where Claude builds pages, documents and toolsGO
MemoryClaude remembers preferences and context across chatsGO
Incognito (Temporary Chat)Nothing stored, nothing remembered, nothing in historyGO
Web searchLive web search inside ClaudeGO
File uploadsPDFs, docs and spreadsheets analysed in chatGO
SkillsCustom specialised behaviours Claude knows when to useGO
Voice modeTalk to Claude with your voice, get spoken or written repliesGO
Claude DesignTurn a brief into polished visuals, slides and one-pagersGO
ConnectorsClaude plugged into your inbox, calendar and other toolsCARE
SharingShare a chat or project with a colleagueCARE
Claude Code & CoworkClaude builds software and works across files on a computerCARE
Admin controlsYour admins govern who's on, who's admin, what's allowedESSENTIAL
Computer Use & DispatchClaude operates a computer like a human, even sent from your phoneHOLD

The upside, the watch-out, and the verdict.

Each card is two minutes' read. Skip what doesn't apply.

1. Claude (the chat)
GO
The default daily use. Open Claude, type a prompt, get a response.
Why it helps Fast, accessible, low friction. On the Team plan your prompts aren't used to train models and your data stays inside the Guild's workspace. Brilliant for copy, content and external communication, where it outshines Copilot.
Where it bites It will make things up with complete confidence. It doesn't know it's lying, it's trying to be helpful. Ask it what it assumed, tell it to flag uncertainty, and fact-check anything heading for print.
The verdict for the Guild Default daily tool for the seat holders. Real material is fine for everyday work. Red-list data stays in your own systems.
2. Projects
GO
Workspaces with persistent knowledge files and instructions. Personal, or shared across the team.
Why it helps Upload the Guild's context once (tone of voice, brand guidelines, Fine Food Digest style, event templates) and everyone with access gets consistent outputs. One project per recurring job works beautifully.
Where it bites Anyone with project access sees the knowledge inside. Choose what goes into team-shared projects deliberately. Personal projects stay private to the user.
The verdict for the Guild Two layers: shared projects for organisational knowledge (voice, templates, programme info), personal projects for individual work.
3. Artifacts
GO
The side panel where Claude produces pages, documents, diagrams and small tools you can preview, edit and download.
Why it helps Turn a need into a one-page resource or a polished document in minutes. Member one-pagers, event pages, producer guides, internal trackers. The page you're reading is one.
Where it bites Whatever you fed into the prompt can show up in the output. Review before sharing externally, the same as everything else the Guild publishes.
The verdict for the Guild One of the biggest unlocks. Build pages for members, producers and the team. Standard "would I send this externally?" check before sharing.
4. Memory
GO
Claude remembers your preferences, working style and recurring context across separate conversations.
Why it helps Less repeating yourself, better outputs over time. Claude learns how you write and what you care about, all inside the Team plan's protections.
Where it bites Personal preference more than risk. Some people prefer a fresh start each time, and users can edit or clear their own memory whenever they like.
The verdict for the Guild On by default. Users control their own memory.
5. Incognito (Temporary Chat)
GO
A mode that doesn't save the chat, doesn't update Memory, doesn't appear in history.
Why it helps The right mode for one-off sensitive work, or anything you'd rather not have referenced later. Easy to switch on.
Where it bites No history means nothing to go back to, and Memory doesn't learn from it.
The verdict for the Guild The team's "if in doubt" setting for anything sensitive that still sits within the rules. Red-list data stays out of AI tools either way.
6. Web search
GO
Claude searches the web live for current information and fact-checking.
Why it helps Fresh research with sources cited. Sector news, market data, background on producers and partners, all checked on the fly.
Where it bites Pages on the web can carry hidden instructions designed to nudge AI off course (prompt injection). Treat web-sourced output as a research assistant's first draft.
The verdict for the Guild Use freely. Verify facts before they go into Fine Food Digest or anything published.
7. File uploads
GO
Drop a PDF, document, spreadsheet or image into Claude for analysis.
Why it helps Analyse a long report in seconds. Extract themes, build summaries, draft from minutes, compare versions. A huge time-saver for desk work.
Where it bites Same rules as everywhere else: red-list files stay in the Guild's own systems.
The verdict for the Guild Use freely for standard internal work, with the red list as the line.
8. Skills
GO
Custom specialised behaviours Claude knows when to use, written for the Guild's recurring jobs.
Why it helps The best prompt anyone has written for a magazine feature or an event plan becomes how it always gets done, whoever's asking. Your best practice, codified.
Where it bites Skills are only as good as the brief behind them. Review them every few months so they stay aligned with how the Guild actually works.
The verdict for the Guild Build them as recurring needs surface. You already have someone in the building who can, and I'll help.
9. Voice mode
GO
Talk to Claude with your voice, get spoken or written replies.
Why it helps Three times faster than typing and brilliant for thinking out loud. A great accessibility win too.
Where it bites Voice input transmits the same way as text, so the same data rules apply. Quieter rooms work better.
The verdict for the Guild Encourage it. Talk, don't type. Same rules as text.
10. Claude Design
GO
Describe what you need and Claude builds the visual: designs, slides, one-pagers, prototypes. Refine by conversation, comments or direct edits.
Why it helps Polished visual work without a designer's day. Event visuals, member one-pagers, deck mock-ups, campaign concepts. Load the Guild's brand once and everything comes out on-brand, with export to Canva, PDF, PowerPoint or a web page.
Where it bites Whatever you feed in can appear in the output, so the usual review applies before anything goes external. And your editorial standards decide where AI-assisted visuals do and don't belong in Fine Food Digest.
The verdict for the Guild Use freely, especially in marketing and events. I use it constantly. Mock up first, then decide what deserves the human polish in InDesign.
11. Connectors
USE WITH CARE
Claude plugged into your other tools: inbox, calendar, files, note-takers and more.
Why it helps The biggest productivity unlock. Claude preps your morning briefing from your real calendar and inbox, drafts replies in context, and pulls together everything you need before a call.
Where it bites Each connector opens a new door into your systems. An email or document can carry hidden instructions meant for the AI rather than for you (prompt injection). Without an approval habit, the picture sprawls.
The verdict for the Guild Enable deliberately, not by default. Every connection goes through [technical lead], one at a time, with Claude set to ask before it acts. Allen stays unconnected until we've talked it through properly.
12. Sharing
USE WITH CARE
Share a single chat with a colleague, or share a whole project across the team.
Why it helps Spreads good patterns fast. Newer users see what good AI work looks like, which is exactly how the team levels up together.
Where it bites Sharing a chat shares everything in it, including content earlier in the thread you might have forgotten about.
The verdict for the Guild Share freely inside the workspace. Quick mental check before sharing externally: would I send this exact chat by email?
13. Claude Code & Cowork
USE WITH CARE
Claude builds working software and works across files and folders on a computer. Already powering the World Cheese paperwork tool.
Why it helps The Guild effectively has a dev team in the building. Repetitive processes become tools: producer paperwork, internal forms, the jobs that used to eat someone's afternoon. Cowork handles the bigger desk jobs, like a finished pack built from a folder of sources.
Where it bites These modes act on real files and real systems. There are settings that let them run without checking in; Guild work stays on the ask-first settings, and Cowork only gets the folder for the job, not everything.
The verdict for the Guild Keep building, it's already paying for itself. Ask-first settings always, and anything that touches member or producer data gets a once-over against the policy before it goes live.
14. Admin controls
ESSENTIAL GOVERNANCE
Your admins manage users, permissions and what's available to the team.
Why it helps The Guild stays in control. Workspace-level visibility of usage, seats added and removed in minutes, and evidence for the upgrade decisions when the heavy users hit their limits.
Where it bites The team should know admin visibility exists, in writing, before they start. Openness up front is what keeps the trust intact.
The verdict for the Guild Set up before launch. Keep the admin team tight, [admins], and say openly in the policy and onboarding what admins can and can't see.
15. Computer Use & Dispatch
HOLD
Computer Use lets Claude operate a computer like a human: clicking, typing, navigating. Dispatch sits on top, so you can send Claude a job from your phone and it completes it on your desktop while you're away.
Why it helps Highest theoretical productivity unlock. Repetitive screen-based admin delegated entirely, even from the train.
Where it bites Highest-risk feature and still in research preview. Claude acts in real systems, so mistakes have real-world consequences. Currently Mac only, and most of the Guild is on Windows.
The verdict for the Guild Not yet. Bank the value from everything above first. We'll revisit together once the team is flying and it lands on Windows.

What to do, in the order to do it.

If you read nothing else, this.

Before anyone logs in

  • Confirm your admins and your named roles. Policy owner, technical lead, first call. One name per job.
  • Publish the policy and put the one-pager on the wall. Everyone knows the rules before they start.
  • Land the red list. Allen data, member and producer personal details, pre-announcement judging results, financials. Inside Guild systems, always.

Turn on freely (the day-one stack)

  • Claude chat, Projects, Artifacts, Memory, Incognito, Web search, File uploads, Skills, Voice mode, Claude Design. Everyone with a seat gets value from day one. The Team plan protections do the heavy lifting, and Copilot covers everyone else.

Turn on with care

  • Sharing. Free inside the workspace, quick check before anything external.
  • Claude Code and Cowork. Keep building on ask-first settings, with a policy once-over for anything touching member or producer data.

Decide together

  • Connectors. Which tools Claude gets plugged into, in what order, through your technical lead. Allen stays unconnected until we've talked it through. This is a conversation for the workshop on 17 July.

Hold for now

  • Computer Use and Dispatch. Bank the rest of the value first. We'll revisit together when the time is right and it reaches your Windows machines.