AI Transparency & Limitations Notice

1. Summary

Effective Clerking uses artificial intelligence to help you draft documents faster. It does not, and cannot, do your job for you.

Every document the system produces is a draft. Nothing is finalised, circulated or sent without you reviewing it and clicking to approve it.

AI output can be wrong. It can be inaccurate, incomplete, or state things that were not in your notes. You must check it.

Your content is not used to train AI models. Our AI provider is contractually prohibited from doing so, and we will not engage any AI provider that does not accept the same prohibition.

The compliance check is not legal advice. It checks your draft against a checklist that you maintain, for documentation evidence only.

You must tell people when AI helped produce a document you circulate. We give you wording below.

2. You are interacting with an AI system

Effective Clerking includes an artificial intelligence system. When you use the drafting, compliance-check, action-extraction or statement-extraction features, your text is processed by a large language model.

We tell you this here, and the Service tells you again at the point of use, because you are entitled to know when you are dealing with AI rather than a person.

3. Which AI providers we use

3.1 Current providers

Primary provider — all AI features

Additional or contingency providers

None currently engaged. If one is added, it appears here and in our Sub-processor List before it processes any customer content, and the process in section 3.3 applies.

3.2 How the AI is called

Only from our own server-side code. Our API keys are never exposed to your browser, and your browser never talks to an AI provider directly.

3.3 If we add or change an AI provider

We may engage an additional or alternative AI provider — for continuity if our primary provider is unavailable, for a specific feature, or because a better option becomes available.

Any AI provider we engage must meet all of the following before it processes any customer content, and we will not use one that does not:

A written data processing agreement containing the terms required by Article 28(3) of the UK GDPR.

A contractual prohibition on training models on our customers’ content.

A published retention position, with the shortest retention the provider offers.

A lawful transfer mechanism under Chapter V of the UK GDPR where processing takes place outside the UK.

Published, current independent security certification.

A published sub-processor list.

Before an AI provider begins processing customer content we will:

add it to our Sub-processor List and to this notice;

give customers at least 30 days’ notice by email, as required by clause 5.3 of our Data Processing Agreement;

honour the right to object in clause 5.4 of that agreement;

complete a transfer risk assessment where processing takes place outside the UK; and

record the assessment in our AI Governance Policy.

Where an urgent change is needed for continuity — for example our primary provider suffers an extended outage — we may make the change first and notify as soon as practicable, with the reason. Your right to object is not affected.

4. Exactly where AI is used

The AI is not used to make decisions about people, to score or evaluate any individual’s performance or conduct, to decide who gets access to anything, or to generate anything that is sent without your approval.

5. What AI output cannot be relied on to be

Accurate. Large language models can produce text that is plausible and wrong. They can state a decision that was not taken, attribute a remark to the wrong person, or introduce a detail that was not in your notes.

Complete. Something in your notes may not make it into the draft.

Consistent. The same notes submitted twice may produce differently-worded drafts.

Original. Similar inputs can produce similar output for different users. We make no representation that output is unique to you or free from resemblance to output generated for others.

Legally sufficient. No output has been checked against any statutory requirement, and none of it constitutes legal advice.

Where the system is unsure or your notes are ambiguous, it is instructed to insert a [CLARIFY: …] marker rather than guess. Treat every such marker as a task for you, not a cosmetic artefact.

6. What you must do — this is not optional

These obligations flow from our own agreement with our AI provider and are binding on you under clause 8 of our Terms of Service. They apply equally to any AI provider we engage in future.

6.1 A qualified professional must review every AI-assisted document before it is finalised

Governance documents affect individuals and carry legal consequences. Our AI provider’s usage policy requires that where AI is used to provide advice, recommendations or subjective decision-making that directly affects individuals, a qualified professional in that field must review the content before it is disseminated or finalised.

Two of the domains that policy treats as high-risk are directly relevant to clerking: legal — interpreting articles of association, a scheme of delegation, quorum or governor eligibility — and employment, where minutes record a headteacher appraisal or a disciplinary matter.

For Effective Clerking that qualified professional is you, or the governance professional in your organisation. The Service is built to require an explicit approval action on every AI-touched surface. Do not attempt to bypass it, and do not treat approval as a formality.

6.2 You must disclose that AI was used

Where you present AI-assisted output to other people — governors, trustees, staff, or the public — you must disclose that AI helped produce it. This is a requirement of our AI provider’s usage policy, and it is good governance practice in its own right.

Suggested wording, which you are free to adopt:

This document was drafted with the assistance of Effective Clerking, an AI-assisted clerking tool, from notes taken at the meeting. It has been reviewed and approved by the clerk before circulation.

For minutes going to a board for approval, a footer line is usually sufficient. For anything published, put it where a reader will see it.

6.3 You must check facts before relying on them

Our AI provider requires us to tell you, and we do: factual assertions in output should not be relied upon without independently checking their accuracy, as they may be false, incomplete, misleading, or not reflective of recent events or information. It is your responsibility to evaluate whether output is appropriate for your use, including where human review is appropriate, before using or sharing it.

6.4 You must not use the AI features for prohibited purposes

Your use is subject to our Acceptable Use Policy and, through it, to our AI provider’s usage policy as it applies from time to time. In particular you must not use the Service to produce anything presented as legal advice, or to make or support decisions about an individual’s employment, without independent professional review.

7. What we do with your content, and what we do not

We do not train models on your content. Under Anthropic’s Commercial Terms of Service, Anthropic may not train models on customer content submitted through its API. This is a contractual prohibition, not a setting we chose — and, as section 3.3 says, it is a condition of engaging any AI provider.

Retention by our current AI provider. Anthropic’s published position is that it deletes API inputs and outputs from its backend within 30 days of receipt or generation. Where content is flagged by Anthropic’s automated trust and safety systems as a possible usage policy violation, Anthropic may retain inputs and outputs for up to 2 years, and trust and safety classification scores for up to 7 years.

We send only what is needed. For a drafting request, the AI receives your notes for that meeting, the school’s initials register, the selected style template and the open actions list. It does not receive your other clients’ data, your account details, your billing information, or anything from another customer.

We do not use your content to improve our product beyond providing the Service to you, except in the form of aggregated statistics that cannot identify you or any individual.

8. Human oversight is designed in, not bolted on

The Service is architected so that:

no document changes state without an explicit human action;

no email is sent without an explicit human action;

no action is added to a tracker, or closed, without an explicit human action;

no compliance flag resolves itself — you dismiss or address it;

no scraped or extracted text reaches a document until you have reviewed it;

once a document is sent it is locked, and a correction creates a new version under the same reference number rather than silently overwriting the original.

Every one of those steps is recorded in an audit log with the actor and timestamp. If you are ever asked to demonstrate that a human, not a machine, was responsible for a set of minutes, the audit log is your evidence.

This also matters legally. The Data (Use and Access) Act 2025 substituted new Articles 22A to 22D into the UK GDPR, restricting solely automated decisions that have a significant effect on people — and restricting them further where special category data is involved. The human approval gate is what keeps this Service outside that regime. It is a compliance control, not a preference, and we will not remove it or make it bypassable.

9. Marking AI-assisted documents

We identify AI-generated draft output as such within the Service.

If you publish or circulate AI-assisted minutes or agendas, your own review and editorial responsibility for the final text is what makes the document yours. Keep a record of that review — it is good governance, it is what the approval audit trail evidences, and it is what your disclosure under section 6.2 rests on.

10. Our AI governance

We maintain an internal AI Governance Policy covering provider selection and the minimum requirements in section 3.3, model selection, prompt change control, evaluation of output quality, human-oversight requirements, incident handling for AI-specific failures, and review of our AI providers’ terms. It is aligned with the ICO’s guidance on AI and data protection and with DfE’s Generative AI: product safety standards for edtech suppliers. A summary is available to customers on request.

11. Questions, and telling us when it gets something wrong

If the Service produces output that is materially wrong, misattributed, or unsafe, please tell us at support@effectiveclerking.co.uk. We log AI output complaints, review them, and use them to improve our prompts and guardrails. We would rather hear about it than not.

For questions about this notice: support@effectiveclerking.co.uk.