Technology, data and AI

Writing an AI use policy for your team

A practical AI use policy for UK restaurants and pubs: what staff may put into AI tools, what stays out, approved tools, checks on outputs and a template.

Ava research teamPublished 7 min read

On this page
  1. Why a small venue needs one
  2. The traffic-light list
  3. What the policy should contain
  4. The three checks that matter most
  5. Show staff how to strip a prompt
  6. When something goes wrong
  7. Security basics to include
  8. Rolling it out without a lecture
  9. Guest-facing AI belongs in the same document
  10. Frequently asked questions
  11. References

Key takeaways

  • Your team is probably already using free AI tools on their phones; a policy brings that into the open rather than banning it.
  • The core of the policy is a traffic-light list of data: green to use freely, amber with care, red never.
  • Anything an AI tool writes about allergens, licensing or employment terms must be checked by a named person before it is used.
  • Keep it to two pages, brief it at a pre-shift meeting and review it every six months.

Why a small venue needs one

AI use has spread fast across UK business. In September 2026, 30% of all UK businesses used at least one AI technology, up from 9% three years earlier[1], and 29% of businesses with fewer than 10 employees were using it[2]. In accommodation and food, 8.9% were using text generation tools by late September 2026[3]. Those figures count businesses that know they use AI. Staff writing rota messages, social captions or supplier emails with a free chatbot on their own phone often do not show up in anyone's count.

Share of UK businesses using at least one AI technology
Share of UK businesses using at least one AI technologyComparison. All businesses, Sept 2023: 9%; All businesses, Sept 2026: 30%; Under 10 employees, Sept 2026: 29%.9%All businesses, Sept 2023All businesses, Sept 2023: 9%30%All businesses, Sept 2026All businesses, Sept 2026: 30%29%Under 10 employees, Sept2026Under 10 employees, Sept 2026: 29%Share of UK businesses using at least one AI technologyComparison. All businesses, Sept 2023: 9%; All businesses, Sept 2026: 30%; Under 10 employees, Sept 2026: 29%.9%All businesses, Sept 2023All businesses, Sept 2023: 9%30%All businesses, Sept 2026All businesses, Sept 2026: 30%29%Under 10 employees, Sept 2026Under 10 employees, Sept 2026: 29%

Source: Office for National Statistics, 2026[1]

Show the data
ItemValue
All businesses, Sept 20239%
All businesses, Sept 202630%
Under 10 employees, Sept 202629%

The risk is not the tool itself. It is a supervisor pasting a guest complaint, with name, phone number and a description of an allergic reaction, into a free service that keeps the text. Or a menu description generated by AI that drops the word 'peanut'. A policy sets the boundary before that happens, and it gives staff permission to use tools that genuinely save time.

The traffic-light list

Put this table at the top of the policy. It is the part people will remember.

Example traffic-light list for a 40-cover restaurant; adapt the examples to your venue.
ColourData or taskRule
GreenMenu descriptions without allergen claims, social captions, event ideas, recipe scaling, translating a public notice, drafting a job advertUse any approved tool; a manager reads before publishing
AmberReview replies, supplier emails, rota templates, training quizzes, summaries of meetingsApproved tools only; remove names and contact details first
RedGuest names with phone numbers or emails, allergy or health details, card or bank details, staff grievances, disciplinary notes, CCTV images, passwordsNever enter into any AI tool unless it is a system the business has contracted for that purpose

The red list follows the law. Health details are special category data under UK GDPR, and card data has its own security standards. A free chatbot on a personal account has no contract with your business, so you cannot say where that data goes or how long it stays.

What the policy should contain

  1. Purpose: one sentence on why the business uses AI and who the policy covers, including agency staff and contractors.
  2. Approved tools: a named list, with which account to use. Business accounts only; no personal sign-ins for work tasks.
  3. The traffic-light list above.
  4. Checking outputs: who must approve AI-written content before it reaches guests, suppliers or the public.
  5. Guest-facing AI: how the business tells guests when they are dealing with an AI assistant, and how they reach a person.
  6. Recruitment and staff decisions: AI may help draft adverts but must not reject applicants or set hours without a manager reviewing each case.
  7. Mistakes and incidents: what to do if red data was entered by accident, and who to tell within 24 hours.
  8. Ownership and review: the named owner and the next review date.

Sample clause: 'Any description of a dish's ingredients or allergens produced with an AI tool must be checked against the kitchen's allergen matrix by the head chef or duty manager before it is printed, posted or read to a guest.'

The three checks that matter most

Most AI mistakes in hospitality fall into three groups. Build the checks into the policy as named responsibilities, not general advice.

  1. Allergen and food safety claims

    AI tools produce fluent text that can omit or invent ingredients. Nothing about allergens, cooking temperatures or food safety goes out without a check against your own records.

  2. Legal and employment wording

    Licensing conditions, tipping policies, contracts and disciplinary letters are drafted or checked by the owner or an adviser. Tools often describe rules that apply in another country.

  3. Public replies

    Review replies and social posts are read by a manager before posting. Never confirm a guest's booking details, health information or a staff member's name in a public reply.

Show staff how to strip a prompt

Telling people to remove personal data is easier when they see it done. Put one before-and-after example in the policy, using a real type of task from your venue.

Example. Before: 'Write a reply to Sarah Patel, 07700 900123, who says her son had a reaction to the satay at table 12 on Friday and our waiter Tom ignored her.' After: 'Write a calm, short reply to a guest who says a child had a reaction after a dish on Friday and felt the server did not take her seriously. Offer a call with the manager. Do not admit liability or mention health details.'

The stripped version still gets a useful draft. The manager then handles the real complaint by phone, records the incident in the allergen log and follows your complaints procedure. The AI tool never sees the guest's name, number, child or the staff member involved.

The same habit works for rotas (use role names such as 'Server A' instead of people's names), supplier disputes (remove account numbers) and meeting notes (summarise the decision, not who said what).

When something goes wrong

People will make mistakes, and the policy should make it easy to own up quickly. Spell out the steps so a supervisor at 11pm knows what to do.

  1. Stop and record

    Note what was entered, into which tool, when and by whom. Take a screenshot if you can.

  2. Delete where possible

    Delete the conversation in the tool and, for business accounts, ask the supplier to remove it from logs.

  3. Tell the owner within 24 hours

    The owner decides whether it is a personal data breach that must be reported to the regulator, which has to happen within 72 hours of becoming aware of a reportable breach.

  4. Fix the cause

    Was the tool unapproved, or the data list unclear? Update the policy or training rather than blaming the individual.

Security basics to include

In 2025, 30% of UK food or hospitality businesses identified a cyber breach or attack in the previous 12 months[4]. AI tools add logins and stored conversations, which add risk. Include a short security section:

  • Two-step login on every business AI account.
  • Chat history turned off, or set to the shortest period, where the tool allows it.
  • No sharing of manager logins between shifts.
  • Leavers removed from AI tools on the same day as the till and the booking system.
  • Downloads of guest lists never uploaded to an AI tool to 'tidy them up'.

Rolling it out without a lecture

Hospitality has one of the lowest training rates of any sector: only 42% of accommodation and food SME employers provided training in 2025[5]. A two-page policy works only if people hear it. Use a ten-minute pre-shift slot.

  1. Ask first

    'Who has used an AI app for anything at work this month?' Most hands go up. Thank them; the policy is about doing it safely.

  2. Show the traffic lights

    Print the table and stick it inside the office door and by the pass.

  3. Give two real examples

    One green: a captions draft for the weekend special. One red: a complaint email with the guest's allergy details.

  4. Name the checker

    Say who approves allergen wording and who approves public replies on each shift.

  5. Get a signature

    Add the policy to the onboarding pack and keep signed copies with the staff handbook.

Where AI is used inside the business, it mostly affects admin. Among UK businesses with 10 or more staff that use AI, 38% said administrative and clerical roles were most affected[6], and the most used tools were text generation (21%) and visual content creation (19%)[7]. That is where your policy will get the most use: emails, captions, rotas and menus.

Guest-facing AI belongs in the same document

If the business uses an AI assistant on the phone or website, the policy should say what it may handle and what it must pass to a person. Write it as a short list: opening times, directions and booking requests can be handled; allergen decisions, complaints, refunds and anything about a guest's safety go to a named manager. If you use Ava, its routing settings should match this list exactly, so that what staff are told and what the assistant does are the same thing.

For the data protection side of these tools, see AI and UK GDPR: what the ICO expects and the DPIA guide for guest tech. For a channel-by-channel view of what can be automated, read automating booking enquiries across phone, web and chat.

Frequently asked questions

Should we ban staff from using AI tools?

A ban is hard to enforce when the tools are free on every phone. A clear list of approved tools and forbidden data is more realistic and protects guests better.

Can staff use AI to reply to online reviews?

Yes, as an amber task: remove names and details first, and a manager reads the reply before it is posted. Never confirm health details or booking information in public.

Who should own the policy in a single-site venue?

The owner or general manager. Name a deputy for each shift who can approve allergen wording and public posts.

How often should the policy be reviewed?

Every six months, and whenever you add a new tool or a supplier adds AI features to software you already use.

References

  1. Office for National Statistics (2026). Business insights and impact on the UK economy: 8 October 2026 (opens in a new tab) ↑

  2. Office for National Statistics (2026). Business insights and impact on the UK economy: 8 October 2026 (opens in a new tab) ↑

  3. Office for National Statistics (2026). Business insights and impact on the UK economy, wave 165 dataset (AI sheets) (opens in a new tab) ↑

  4. Department for Science, Innovation and Technology (2025). Cyber security breaches survey 2025 (opens in a new tab) ↑

  5. Department for Business and Trade (2025). Longitudinal Small Business Survey 2025: SME employers (businesses with 1 to 249 employees) (opens in a new tab) ↑

  6. Office for National Statistics (2026). Business insights and impact on the UK economy: 8 October 2026 (opens in a new tab) ↑

  7. Office for National Statistics (2026). Business insights and impact on the UK economy: 8 October 2026 (opens in a new tab) ↑

Book a demo

See Ava handle your calls.

Book a short demo. We’ll walk through your calls and workflow, and agree what Ava should take care of.

Book a demo