ICF AI guidelines and the coaching platform
This page is for coaches evaluating whether the platform aligns with their obligations under the ICF AI guidelines. The table below maps each guideline to how the platform handles it, including where it fully meets the requirement and where the coach's own judgment and processes are needed.
The International Coaching Federation published their Acceptable Use of Artificial Intelligence (AI) and Client Protection Guidelines in September 2025.
Alignment summary
| ICF Guideline | Platform Approach | Status |
|---|---|---|
| Appropriate use of AI (session preparation, frameworks, professional development, pattern analysis) | AI analyses transcripts against ICF competency rubrics, seeds development areas, and produces cross-session synthesis. All appropriate uses per the ICF guidelines. | Meets |
| Inappropriate use of AI (sharing client info, clinical diagnoses, replacing human judgment, undisclosed AI, AI communicating with clients) | Local providers share nothing externally. AI does not diagnose or replace judgment. Assessments are clearly labelled as AI-generated. AI does not communicate with clients. | Meets |
| Sequestered AI models (recommended for client data) | Three local providers (Ollama, LM Studio, llama-server) run entirely on the coach's machine. Cloud providers are also available. The coach chooses. | Meets |
| No PII in open AI models | Local providers keep everything on the machine. Cloud providers receive full transcripts including names. PII stripping before cloud submission is on the roadmap. | Partially meets |
| Avoid PII in all AI unless authorised | Local providers process data within the coach's own environment. The coach is the data controller. Cloud providers receive transcripts when used. | Partially meets |
| Confidentiality safeguards | Local providers maintain full confidentiality. Cloud providers transmit transcript data to the provider's API for analysis. | Partially meets |
| Informed consent (transparent AI use, coaching agreement clause) | Compliance log documents every AI analysis automatically. Informing clients and obtaining consent is the coach's responsibility. | Partially meets |
| Professional liability (coach responsible for all decisions) | AI is supplementary. Competency scores include transcript evidence for the coach to verify. The platform does not make decisions for the coach. | Meets |
| Intellectual property (disclose AI-generated content) | AI assessments are clearly labelled as AI-generated within the platform. External disclosure is the coach's responsibility. | Meets |
| Quality assurance (review and validate AI content) | Every competency score comes with specific transcript evidence. The coach can verify each assessment against the actual session moment. | Meets |
| Document your approach (written AI policy) | Immutable compliance log of every AI analysis: when it ran, which session, which provider. Automatic and searchable. | Meets |
| Regular review (assess AI usage periodically) | Compliance log with filter and search enables review of all AI activity over any period. | Meets |
| Professional development (continuous ethical education) | Cross-session synthesis and development areas support reflective practice. Ethical AI education is external to the platform. | Partially meets |
| Data security (reputable platforms, strong security) | All coaching data stored locally. HTTPS for all external communication. No telemetry, no analytics, no crash reports. | Meets |
| Data protection compliance (GDPR, CCPA, certifications) | Privacy policy covers GDPR and Australian Privacy Principles. Local-first design means the coach is the data controller. No formal SOC 2 or ISO 27001 certification. | Partially meets |
| Client-centred approach (prioritise client needs, adjust if uncomfortable) | AI is entirely optional. The coach controls whether to use it, which provider, and which sessions to analyse. The platform works fully without AI. | Meets |
10 meets, 6 partially meets, 0 does not meet.
The "partially meets" items fall into two categories: PII handling when using cloud AI providers (addressed by using local providers or by the planned PII stripping feature), and coach responsibilities that no tool can fully automate (informed consent, ethical education, formal compliance certifications).
Sequestered AI models
The ICF distinguishes between open AI models, where data travels to external servers, and sequestered models, where data stays within the coach's own environment. For anything involving client data, they recommend sequestered.
The coaching platform supports three local AI providers that qualify as sequestered: Ollama, LM Studio, and llama-server. All three run on the coach's machine. The transcript, the names, and the entire analysis stay local. Nothing leaves the machine.
Cloud providers (OpenAI, Anthropic, xAI, Perplexity, Mistral, Groq, Google Gemini) are also available for coaches who prefer them. With cloud providers, the coach holds their own API keys and controls which sessions are analysed. See AI setup for configuration.
Personally identifiable information
The ICF says never to input personally identifiable information into AI systems, and recommends using generic descriptions when seeking AI assistance with client scenarios.
With a local AI provider, nothing identifiable reaches any external service. The transcript, including all names, stays on the coach's machine throughout the analysis.
With a cloud AI provider, the full transcript is sent to the provider's API for analysis. This includes any names spoken during the session. The coach controls which provider receives the data and holds their own API keys, but the transcript is not anonymised before submission.
PII stripping before cloud AI submission is on the roadmap. This feature will automatically mask client names and identifying information in transcripts before sending them to cloud providers, allowing coaches who use cloud AI to align with the ICF's PII guidance without switching to a local model.
Privacy Mode is a separate feature that masks client names in the platform's user interface for safe demos, screenshots, and screen sharing. It does not affect what is sent to AI providers.
Compliance logging
The ICF asks coaches to document how they use AI and be transparent with clients about their approach. The platform keeps an immutable compliance log of every AI analysis: when it ran, on which session, with which provider and model, and what type of analysis was performed. This log is automatic, searchable, and cannot be edited or deleted.
Coaches can use this log to demonstrate their AI usage to supervisors, mentors, or credentialing bodies. See the complete feature list for details.
Quality assurance
The ICF says coaches must review and validate AI-generated content before using it with clients. AI should inform thinking, not replace professional judgment.
Every competency assessment in the platform comes with specific transcript evidence. Each ICF competency marker is scored and linked to the exact moment in the session transcript where the behaviour was observed. The coach can click through to the transcript, play back the recording from that point, and verify the assessment against what actually happened.
The platform also supports manual competency tagging, where the coach tags transcript segments against ICF competency markers during their own session review. Both AI-generated and manual tags are visible side by side. See AI analysis modes for how standard and deep analysis work.
Coach responsibilities
The following areas from the ICF guidelines require action from the coach that the platform supports but cannot automate:
- Informed consent. The ICF recommends adding a clause to coaching agreements about AI use. The compliance log provides the documentation, but informing clients and obtaining their consent is the coach's responsibility.
- Intellectual property disclosure. AI assessments are clearly labelled within the platform, but if a coach shares AI-generated content externally, disclosure is their responsibility.
- Professional development. The platform's cross-session synthesis and development areas support reflective practice, but ethical AI education and continuous learning are pursued outside the platform.
- Data protection compliance. The platform's local-first architecture and privacy policy cover GDPR and Australian Privacy Principles, but formal certifications (SOC 2, ISO 27001) are not applicable to a single-developer, local-first product. Coaches in regulated jurisdictions should review the privacy policy and terms of service against their obligations.
ICF guidelines document
The full ICF document is available at:
Acceptable Use of Artificial Intelligence (AI) and Client Protection Guidelines
This application is not affiliated with, endorsed by, or sponsored by the International Coaching Federation (ICF). ICF is a registered trademark of the International Coaching Federation.