The ACA Framework: authority and credibility as measures of algorithmic representation
The methodological framework created by Sonia Yánez Blum as an independent researcher and applied by Blum Digital PR to measure, analyse and manage how language models represent an organisation.
The ACA Framework (Authority and Algorithmic Credibility) was developed by Sonia Yánez Blum through independent research to measure organisational representation in language models. It works across two output dimensions — algorithmic authority and algorithmic credibility — and four analytical pillars: Traceability, Narrative Coherence, Depth and Currency. The result is the ACA-Score™: an index comparable across models and trackable over time.
What the ACA Framework measures
Language models — ChatGPT, Claude, Gemini, Perplexity — answer queries about organisations based on patterns in their training data. That answer is not controlled by the organisation: the model constructs it from external sources, mentions across different contexts, and the coherence between them. The ACA Framework measures the result of that process.
The measurement works across two independent dimensions: how present the organisation is as a reference (authority) and how accurate the model’s description of it is (credibility). The two dimensions are separable because their causes are different and their solutions differ accordingly.
An organisation may have high authority — the model mentions it frequently as a sector reference — but low credibility if the description it produces is imprecise or outdated. The reverse also occurs: high credibility in the description but low presence as a reference. The ACA Framework distinguishes the two problems because they require different interventions.
The two dimensions of the ACA-Score™
Algorithmic authority
Algorithmic authority measures the weight with which the model incorporates the organisation into its responses. It evaluates whether the model mentions it when asked about its sector, whether it mentions it first or as a secondary reference, whether it attributes statements or positions to it, and whether it cites it as an information source on the topics where it should be a reference.
Algorithmic authority is not the same as popularity. An organisation may generate many media mentions yet have low authority in models if those mentions are not the type of signal that models weight as indicators of sector specialisation.
Algorithmic credibility
Algorithmic credibility measures the accuracy and coherence with which the model describes the organisation when it includes it in a response. It evaluates whether what the model says is factually correct, whether the attributes it assigns correspond to the organisation’s actual positioning, whether the narrative is coherent with what the organisation communicates about itself, and whether there are incorrect, outdated or confused statements.
Low credibility does not always come from an obvious factual error. It sometimes comes from a generic description that fails to capture specific attributes, or from a narrative that correctly combines some data but contextualises them inadequately.
The four pillars of ACA analysis
The ACA-Score™ is not calculated directly from the two output dimensions. It is built from four analytical pillars that structure the diagnosis of underlying causes.
- Traceability
- Evaluates whether the statements the model makes about the organisation can be traced back to verifiable sources. A representation without traceability — where the model describes attributes with no basis in identifiable sources — signals that the model is inferring without sufficient data. Absence of traceability is the most frequent cause of factual errors in algorithmic representation.
- Narrative Coherence
- Evaluates whether descriptions of the organisation across different sources are consistent with each other. When sources are contradictory or fragmented, the model generates a synthesis that may mix attributes, periods or different positionings. Narrative Coherence measures the degree to which the organisation has a consolidated public narrative that the model can process consistently.
- Depth
- Evaluates the quantity and quality of information available about the organisation in the sources the model processes. A shallow representation — where the model mentions the organisation but cannot describe what it does with precision — indicates insufficient Depth. Depth does not depend solely on the volume of owned content: it depends on presence in relevant external sources that develop the organisation’s activity and positioning in detail.
- Currency
- Evaluates the degree to which the model’s representation reflects the organisation’s current state. Language models have training cut-off dates, but they also weight the mass of historical data against recent data. An organisation that has evolved its positioning may have an outdated representation even after publishing recent content, if the historical volume exceeds the recent volume in the corpus the model processed.
The complementary frameworks: RICFE, FACE and GEAC Protocol
The ACA Framework operates alongside three additional frameworks that structure correction, visibility and governance work.
- RICFE — Role, Instruction, Context, Format, Emotion/Goal (Academia ARP)
- A strategic prompting framework for communication teams created by Academia ARP. Blum Digital PR applies it as a methodological complement: it structures how to design and apply diagnostic prompts to analyse algorithmic representation systematically, ensuring consistency across measurement cycles and comparability of results across different models.
- FACE — Sources, Amplifiers, Catalysts, Echoes (academic research)
- An algorithmic reputation propagation model from academic research. It describes how an organisation’s digital signals travel through AI-mediated systems and emerge as content cited, paraphrased or represented in generative outputs. Blum Digital PR applies it as an algorithmic visibility framework: it determines which types of presence in external sources feed model representation and how that presence can be developed sustainably.
- GEAC Protocol — Governance and Ethics in Algorithmic Communication
- Created by Sonia Yánez Blum as an independent researcher. It governs the ethical use of AI in communication teams: defines validation criteria for language model outputs, human oversight procedures and editorial accountability frameworks. It is the foundation for AI governance projects for agencies and communication departments, and provides alignment with the EU AI Act and ISO 42001.
How the four frameworks are applied in a project
All Blum Digital PR projects begin with the AI Diagnosis, which applies the ACA Framework to produce the organisation’s initial ACA-Score™ in the main models: ChatGPT, Claude, Gemini, Perplexity, and any other model relevant to the client’s sector.
The diagnosis delivers:
- The initial ACA-Score™ in authority and credibility per model
- Analysis of the four pillars (Traceability, Narrative Coherence, Depth, Currency)
- Identification of the root causes of the current representation
- A 90-day action plan with structured prompting protocols
- An initial assessment of algorithmic propagation signals (FACE)
From the diagnosis, the relevant service lines are activated: algorithmic reputation work if the main problem is model representation, governance if the team needs AI usage protocols, training if editorial criteria is the main gap, or crisis management if there is an active and persistent erroneous representation.
At 90 days, the ACA-Score™ measurement is repeated to evaluate the effect of the actions taken. The cycle is continuous because models update, the sector evolves and the organisation’s narrative changes.
See your organisation’s ACA-Score™
The first step is the AI Diagnosis: an executive report in 5–10 working days with the initial ACA-Score™ in authority and credibility for the main models, analysis of the four pillars and a 90-day action plan.
