Citizen Auditing and the AAAP Protocol: An Experimental Accountability Layer for AI Platforms and Democratic Digital Infrastructure

Citizen Auditing and the AAAP Protocol: An Experimental Accountability Layer for AI Platforms and Democratic Digital Infrastructure

An invitation to discuss, test and challenge the proposal

As artificial intelligence and large digital platforms become increasingly influential in public life, one question deserves more attention:

Who audits the institutional behaviour of the systems that increasingly mediate our information, communication and democratic participation?

This proposal is intentionally experimental.

It does not assume that large technology companies are inherently untrustworthy, nor does it assume that public institutions will always provide the necessary oversight. Instead, it starts from a simpler principle:

Trust should be complemented by traceability.

From citizens as users to citizens as observers

Digital platforms have transformed citizens from passive audiences into active participants in public communication. However, citizens generally remain outside the institutional processes that determine how their signals, complaints, evidence, appeals and contextual information are processed.

A citizen may report an issue to a platform, receive a standard response, appeal a decision, or receive no meaningful response at all.

In most cases, these events remain isolated.

Citizen Auditing proposes to change this.

The citizen becomes not only a user of a digital service, but also an observer of institutional behaviour.

The objective is not to compel an institution to agree with the citizen.

The objective is to create a structured, chronological and verifiable record of what happened.

This means documenting, where appropriate:

• the original signal;
• the context in which it was submitted;
• the institution's acknowledgement or non-acknowledgement;
• routing and escalation;
• requests for clarification;
• evidence provided;
• institutional decisions;
• appeals;
• resolution or non-resolution;
• repeated patterns over time.

Importantly, non-response is also an observable institutional event.

So are automated responses, incorrect routing, repeated disclaimers, unresolved appeals and changes in policy or behaviour.

The purpose is not to label an institution as “good” or “bad”.

The purpose is to make institutional behaviour observable over time.

The AAAP Protocol

As an experimental companion to Citizen Auditing, we propose an AAAP Protocol: a structured accountability layer designed to preserve the traceability of consequential interactions between citizens and high-impact digital institutions.

The protocol can be explored around four basic functions:

Accountability → Audit → Appeal → Provenance

The precise implementation should remain open to discussion and testing.

The essential idea is that a consequential digital interaction should not disappear into an opaque interface, an automated workflow or an isolated customer-service ticket.

Where appropriate, it should leave an evidence trail.

This could create a new layer between individual citizens and large digital institutions:

Citizen → Signal → Evidence → Institutional Process → Decision → Appeal → Outcome → Institutional Memory

The resulting dataset could then be analysed longitudinally.

Why does this matter for AI?

The acceleration of AI capabilities will increase the volume and complexity of decisions made or assisted by automated systems.

But technological intelligence and institutional intelligence are not necessarily the same thing.

An AI system may become increasingly capable while the institution deploying it remains unable to:

• properly route external signals;
• preserve context;
• verify claims;
• remember previous interactions;
• learn from recurring failures;
• explain consequential decisions;
• provide meaningful avenues of appeal.

This creates a potential democratic asymmetry:

Increasing technological intelligence without increasing institutional accountability may increase the distance between citizens and the systems that affect them.

Citizen Auditing therefore asks a complementary question to the question of how quickly AI can be developed:

Can the institutions deploying increasingly powerful AI systems also become more observable, verifiable and capable of learning?

Meta as a possible stress test

Large platforms such as Meta provide an interesting experimental environment for this question.

The historical development of social media demonstrated both the democratic potential of large-scale communication platforms and the risks associated with attention-driven information systems, algorithmic amplification, advertising incentives and political communication.

This does not require us to conclude that a particular company is inherently anti-democratic.

Instead, it raises a testable governance question:

Can a platform of enormous social and economic scale make its consequential institutional behaviour sufficiently traceable for independent observation and civic auditing?

The proposal therefore does not ask citizens to “trust” or “distrust” Meta.

It asks whether the interaction between citizens and such platforms can become sufficiently documented to allow longitudinal evaluation.

That distinction is fundamental.

From complaint to evidence

Traditional citizen complaints are often evaluated individually.

Citizen Auditing introduces another level of analysis:

What happens when thousands of individual interactions become structured longitudinal evidence?

Patterns may become visible:

• recurring failure to route;
• repeated use of standardised responses;
• systematic gaps between declared policy and observed practice;
• changes following regulatory intervention;
• differences between jurisdictions;
• improvements or deterioration over time;
• recurring points where citizen signals disappear from the institutional process.

At that point, the objective is no longer to win an argument with an institution.

The objective is to build institutional memory where institutional memory is otherwise fragmented or absent.

An experimental proposition

We propose this as an open experiment rather than a finished standard.

The methodology could be tested by researchers, civil-society organisations, academics, regulators, journalists, technology companies and citizens.

Questions for discussion include:

1. What should constitute an auditable digital event?
2. Which evidence should citizens be able to preserve?
3. How can provenance and authenticity be established?
4. How can privacy and data minimisation be protected?
5. How should automated responses be recorded?
6. How can institutional non-response be represented without turning absence of evidence into an accusation?
7. Could longitudinal datasets reveal institutional learning?
8. Could such datasets complement existing AI governance and regulatory mechanisms?
9. Can an independent civic audit layer improve accountability without becoming another centralised authority?

These are deliberately open questions.

The proposal is therefore offered to the Apply AI community for discussion, criticism, experimentation and refinement.

We would particularly welcome challenges to the model.

If the methodology is flawed, where is it flawed?

If it creates new risks, what are they?

If it is useful, where should it be tested?

And if it can contribute to democratic digital governance, what institutional safeguards would be necessary?

The underlying principle

The central principle is deliberately simple:

Don't optimize for response. Optimize for traceability.

A response can disappear.

A record can be compared.

A single complaint can be dismissed.

A longitudinal pattern is much harder to ignore.

The ambition of Citizen Auditing is therefore not to replace democratic institutions, regulators or independent journalism.

It is to explore whether citizens themselves can contribute to a structured evidence layer that helps society observe how powerful digital institutions actually behave over time.

AI may increase the intelligence of machines.

The next democratic challenge may be increasing the observability and learning capacity of institutions.

This proposal is an invitation to experiment with that missing layer.

Selected References from the Longitudinal Archive 27

This proposal is part of an ongoing longitudinal research and experimentation process. The following publications provide methodological, technical and empirical background for the concepts discussed above:

• AAAP Technical Specification v0.1 — technical specification of the proposed accountability and traceability protocol.
• AAAP as External Observation Layer — explores AAAP as an external, evidence-based observation layer between citizens and high-impact institutions.
• The Empirical Turn in Citizen Auditing — develops the transition from individual civic intervention toward structured empirical observation.
• Citizen Auditing Method Generates Its Own Evidence — examines how citizen auditing can generate structured evidence through documented institutional interactions.
• Institutional Learning Record — develops the idea of recording institutional responses longitudinally in order to evaluate whether institutions learn from recurring signals.
• Future Antecedent — explores how present-day records can become antecedent evidence for future institutional and governance analysis.
• The Report Card of 500 Institutions — extends individual observations toward comparative institutional measurement and benchmarking.
• AAAP in Context — situates the AAAP approach within the broader development of digital accountability and AI governance.

These publications are not presented as a finished theoretical framework or as authoritative conclusions. They constitute an evolving living research record.

The purpose of making the underlying record available is precisely to allow others to inspect the methodology, challenge its assumptions, reproduce parts of the experiment, identify weaknesses and propose improvements.

Traceability of the proposal

The broader Longitudinal Archive 27 provides the chronological research context from which this proposal emerged.

Rather than presenting Citizen Auditing and AAAP as an isolated conceptual intervention, the archive makes it possible to examine how the ideas developed through successive observations, experiments, institutional interactions and methodological revisions.

This is itself part of the proposed methodology:

Don't optimize for response. Optimize for traceability.

The archive is therefore offered not simply as supporting material, but as part of the experiment itself.