AI policy

1. Introduction

Artificial intelligence (AI), including generative AI tools, offers opportunities for creativity, analysis and efficiency in our PR, communications and social media work. However, these tools also introduce risks relating to accuracy, transparency, confidentiality, rights, and reputational impact. As an agency, we are committed to using AI responsibly, ensuring that all outputs maintain the trust of clients, media partners and the public.

This policy sets out how AI may and may not be used in our work. It must be followed by all staff, freelancers and contractors operating on behalf of the agency.

2. Core principles

All AI use in the agency must:

• Never undermine trust with clients, journalists, the public or stakeholders.

• Always involve transparent disclosure and effective human oversight.

• Align with our values of accuracy, impartiality, fairness and respect for privacy.

• Comply with UK GDPR and all legal requirements on data protection, copyright and commercial rights.

• Avoid introducing misleading, fabricated or biased content into our work.

• Prioritise human judgment at all times.

• Only use AI tools with clear terms governing data security, retention and training use.

AI is an aid; it is never a substitute for professional editorial, strategic or creative judgement.

3. Editorial oversight and accountability

• Any use of AI in creating, shaping or presenting client content must have active editorial oversight and senior-level approval.

• A named senior staff member must be accountable for each AI deployment.

• Oversight should match the tool's significance:

  •   High-level oversight for idea generation and curation.

  •   Detailed, line-by-line oversight for analysis or content shaping.

• AI outputs must always be treated as if produced by a junior team member: checked for accuracy, tone, suitability and risk.

• AI outputs must never be published or shared externally without documented human review.
  
  

4. Transparency

• Where AI has meaningfully contributed to the development of content, this must be disclosed to clients and, where relevant, to the public.

• Disclosure should be proportionate:

  •   Required where AI influenced substance or style.

  •   Not required for trivial uses such as spellchecking or grammar support.
    
    

5. Accuracy, bias and verification

• All factual claims produced or informed by AI must be independently verified by a human before publication.

• Staff must actively check for algorithmic bias, omissions, outdated information or AI-generated "hallucinations".

• AI must not be used to summarise legal, regulatory, or financial information without secondary human research and verification.

• Generative AI must never be used to imitate the style of journalists, clients, competitors or public figures.
  
  

6. Rights, privacy and data protection

• Staff may not input confidential client information, unpublished materials, media lists, journalist contact details, embargoed releases or personal data into tools that do not guarantee data isolation and GDPR-compliant processing.

• AI tools may not be used to create content that risks infringing copyright, commercial rights or creative ownership.

• Plagiarism, mimicry of identifiable authors, or generation of outputs derived from protected works is prohibited.

• Image-generation tools must only be used where licensing rights are clearly documented (e.g. Adobe Firefly's royalty-free indemnification terms).
  
  

7. Internal guidance and approvals

• Staff must consult a senior editorial or strategy lead before deploying AI for any new use case.

• High-risk uses may be referred to a small advisory group (editorial, strategy, data/privacy).

• AI risk assessment must consider:

  •   Reputational impact

  •   Accuracy and editorial integrity

  •   Legal and privacy risks

  •   Client expectations

  •   Audience sensitivity
    
    

8. Acquisition and third-party tools

• Any third-party AI tools used for client work must go through internal authorisation.

• Vendor assessment must check:

  •   Data retention policies

  •   Whether user inputs are used for model training

  •   GDPR compliance

  •   IP ownership of outputs

• Staff should work with clients to ensure their own AI use aligns with this policy.
  
  

9. Application-specific rules

The following permitted and prohibited uses combine the current policy with operational guidelines:

9.1 Permitted uses of AI (with human refinement)

• Creative ideation, angles, formats and brainstorming.

• Drafting structures or outlines for human-written content.

• Conducting desk research, initial synthesis or competitor summaries.

• Analysing conversations, sentiment or engagement trends using tools such as Breeze AI within HubSpot.

• Supporting AEO/GEO analysis, schema reviews and visibility assessments.

• Content curation and personalised recommendations under human editorial oversight.
  
  

9.2 Prohibited or restricted uses

• Direct, unedited AI generation of published client content (press releases, articles, blogs, social posts, thought leadership).

• Using AI to write journalist-facing copy such as pitches or statements.

• Inputting media lists, personal data or confidential client material into public tools.

• Creating content that imitates specific journalists, executives or competitors.

• Producing commentary on political, legal, regulatory, crisis or sensitive topics.

• Making claims of authorship or expertise on behalf of clients where AI was involved.
  
  

9.3 AI-detection tools

Tools such as ZeroGPT and SEMrush originality checker may be used where appropriate, but human editorial judgement remains the definitive assessment of originality and suitability.

10. Evolution of the Policy

This policy will be reviewed regularly as AI capabilities, industry standards and legal requirements evolve. Staff will be informed of updates and given training as needed.