What you need to know about ARA.

ARA stands for The ARA Index. It is an independent audit that measures how visible, legible, and citable a public figure's legacy is to the AI systems your audience is already using — ChatGPT, Gemini, Claude, Perplexity, and others.

We score each figure across five proprietary dimensions and produce a ranked score out of 100, benchmarked against their peer cohort. The result is a clear picture of how AI sees the figure today, and a prioritised roadmap for improving it.

AI is now the first point of enquiry for a growing share of research, teaching, and reputation-shaping activity. When a student, journalist, or curator asks a language model what to think about a figure — they get an answer. That answer is not random. It reflects the structured and unstructured record AI was trained on and continues to retrieve from.

Figures whose estates and archives have invested in machine-legibility appear in those answers, richly and accurately. Figures who haven't are being flattened into the nearest cliché, or defined by their critics. The stakes rise further as AI agents become the default authors of biographical content, classroom material, and cultural commentary.

There are now three distinct disciplines, and it matters which one you're doing.

SEO optimises for search crawlers — links, keywords, technical signals. It determines whether machines can find pages about the figure.

GEO (Generative Engine Optimisation) optimises content for AI retrieval — structuring articles, essays, and archival material so that language models surface them in responses. It's SEO's next layer.

ARA works upstream of both. It audits and optimises the legacy itself — the record, the narrative architecture, the positioning, the semantic identity, and the emotional footprint — the foundations that determine what any content or page-level optimisation can achieve. A figure with weak legacy architecture will hit a ceiling on GEO no matter how well their content is structured. ARA finds and fixes that ceiling.

SEO asks: can machines find the record? GEO asks: can machines retrieve the content? ARA asks: do machines understand, trust, and correctly represent the figure? These are different questions at different levels of the stack.

Our Synthetic Recall Test (A3) runs live queries across four platforms: ChatGPT, Perplexity, Gemini, and Claude. We run a standardised set of recall queries for each cohort and record which figures are cited, across which platforms, and in what context.

Our other four dimensions — Structural Readiness, Semantic Clarity, Emotional Residue, and Voice — draw on a broader analysis of how AI systems represent figures across their training data and retrieval behaviour.

We score each figure across five proprietary dimensions, each worth up to 20 points for a maximum of 100:

Each dimension uses a distinct technical methodology:

A1 — Structural Readiness is assessed through automated auditing of biographical schema completeness, archival digitisation coverage, primary source availability, and the density and coherence of canonical reference works.

A2 — Semantic Clarity is measured by probing AI models directly with identity and positioning questions — "Who was [Figure]?", "What did they stand for?", "What is their legacy?" — and comparing the outputs against the historical record. Specificity, accuracy, and differentiation are all scored.

A3 — Synthetic Recall Test runs a defined set of recall queries through ChatGPT, Perplexity, Gemini, and Claude under a standardised protocol: fresh session, no system prompt, default model settings, verbatim response recording. Citations are coded against a consistent rubric and mapped to a recall heatmap.

A4 — Emotional Residue analyses the sentiment distribution of figure-adjacent content across editorial, academic, and archival corpora likely to have influenced AI training. It also probes how models characterise the figure's emotional and moral identity — the vocabulary, associations, and affect they reproduce unprompted.

A5 — Archetype, Personality & Voice evaluates phonetic clarity and name resolution in voice contexts, consistency of personality descriptors across AI outputs, and how reliably a figure's tone is reproduced when AI systems are asked to write in their voice.

Tiers describe a figure's overall ARA standing within a scored cohort. They are relative to the study cohort, not absolute thresholds. The four tiers are:

Awesome — cohort-leading machine presence. AI consistently finds, understands, and cites this figure with depth.

Strong — above-average AI readiness with identifiable gaps. Well-remembered but not dominant.

Average — AI knows the figure but doesn't reliably cite them. Structural or narrative work needed.

Weak — AI either misrepresents the figure or passes them over. Significant intervention required.

Cohort scores are benchmarked within their peer set, not across the full index. A score of 72 among US Presidents means something different from 72 among 20th-century writers — the recall dynamics, AI query patterns, and archival norms differ by cohort.

Across-cohort comparisons should be treated as directional rather than exact. Where we have run multiple waves in the same cohort, wave-on-wave comparisons are fully valid.

Each A3 query set is designed before testing begins and held constant across all platforms and figures in the study. Queries represent real recall occasions derived from research, teaching, and curation contexts — not prompts engineered to produce particular outputs.

We run each query through a standardised session protocol (fresh context, no system prompt, default model settings) and record responses verbatim. Citations are coded by a consistent rubric: present, absent, or not applicable based on query relevance. Platform totals and query totals are reported transparently in every deliverable.

Yes. AI models are updated continuously — some through fine-tuning, some through retrieval augmentation, and some through full retraining. This means that A3 results for a given figure can shift between waves without any action by the estate or archive, simply because the model has ingested more (or different) training data.

This is precisely why re-auditing matters. A figure with strong A3 performance in one quarter cannot assume that result holds later without re-testing. Sustained machine presence requires ongoing investment in the inputs that inform AI training.

A4 is one of the more nuanced dimensions. It evaluates what a figure's life and afterlife have left behind in the data AI systems trained on — not just whether they are known, but whether they are known well.

We assess the sentiment distribution of figure-adjacent content across editorial, academic, and archival corpora. We also probe how AI models characterise the figure when asked open-ended questions: the vocabulary, associations, and affect they reproduce. High A4 scores reflect figures whose lives have generated rich, specific, and durable cultural material — not just biographical mentions.

Yes, indirectly. Modern editorial coverage — obituaries, retrospectives, documentaries, features — enters training data and shapes how AI systems describe a figure. Where a figure's estate or advocates have driven substantive modern coverage, the residue shows up in A2 and A4.

What ARA does not do is treat coverage volume as inherently positive. A large volume of critical or contested modern coverage can suppress a figure's ARA rather than lift it, if it displaces the figure's own voice or record.

Because the data that shapes AI outputs is being written right now — and it reflects what estates, biographers, and advocates have done over the past several years, not what they plan to do next quarter.

AI systems don't discover figures in real time. They synthesise understanding from the cumulative record of what has been published, cited, archived, and linked to. A figure whose estate starts building machine-legible material today will see those signals enter AI training cycles over the coming months. One that waits will be inheriting a version of the figure shaped by whoever moved first.

The second driver is the urgency of the underlying shift. AI is already the first point of enquiry for a meaningful share of research, teaching, and biographical work. The next wave — agentic AI that generates biographies, essays, and cultural commentary autonomously — is moving from experiment to product. In that world, the figures machines don't know, trust, and cite in depth are the figures who quietly disappear from the record.

It depends on which dimension you're improving. Structural changes (A1) move fastest — corrected schema, newly digitised archives, and improved reference coverage can show up in AI outputs within weeks of implementation.

Semantic and emotional dimensions (A2, A4) move on a longer cycle. They reflect what AI systems have absorbed across their training data, which updates over months. A sustained editorial and archival programme will show meaningful improvement over two to three quarters — not two to three weeks.

We map each recommendation to a timeframe in the audit delivery, and we recommend re-auditing every six months for actively managed legacies. The score you get today is a baseline, not a ceiling.

Structural (A1) changes can be reflected in AI outputs within weeks of implementation — new archival digitisation, corrected schema, and expanded reference coverage have measurable near-term effects.

Semantic and emotional dimensions (A2, A4) move more slowly. They reflect what AI systems have absorbed across their training data, which updates on months-long cycles. A sustained editorial and archival programme will show meaningful A4 improvement over two to three quarters.

We recommend re-auditing every six months for actively managed legacies, and annually for tracking purposes.

ARA operates on two levels.

The public index — available at araco.ai — covers cohorts of comparable figures across published studies. It shows each figure's overall ARA score, tier, and per-dimension breakdown. This is independent research published by ARA as cultural intelligence. If a figure is in a published cohort, their standing is already visible.

A commissioned study goes significantly deeper. It is private by default and includes everything in the public index plus:

· Extended A3 testing — more recall occasions, more query variants, platform-specific analysis
· Full per-figure findings narrative: Strengths, Vulnerabilities, Opportunities, and Watches
· A prioritised recommendations roadmap with implementation timeframes
· Cohort-level intelligence on AI positioning dynamics
· Wave-on-wave tracking if re-auditing

The public score is real. The commissioned study is everything you need to act on it.

A standard cohort study takes three to four weeks from scoping to delivery. This includes query design and platform testing (A3), structural and semantic analysis across all figures, findings synthesis, and production of the full report and recommendations deck.

Rush timelines (two weeks) are available for an additional fee. Ongoing monitoring programmes operate on a continuous basis with quarterly snapshots.

Our standard study covers a cohort of 6–10 figures. The benchmarked context is a core part of the value — knowing a figure's score in isolation tells you less than knowing where they stand against the peers AI is citing instead.

Single-figure audits are available on request. They are best suited to estates seeking a baseline before entering a full cohort study, or to figures in cohorts we have already covered.

Very little. We need:

· The cohort you want assessed
· The figures you want in the cohort (or we can recommend a standard peer set)
· Your commissioning contact and any internal briefing context

We do not require access to private archives, unpublished material, or estate systems. Our methodology is entirely external — we assess how AI systems represent the figure based on publicly available information and live query behaviour.

ARA operates on two tiers, and the privacy model is different for each.

The public index is independent research — ARA-initiated cohort studies published as cultural intelligence. Any figure in a published cohort has a public score. This is intentional: the index exists to demonstrate what algorithm readiness looks like in practice, and to give estates a baseline before they decide to go deeper.

Commissioned studies are private by default. The extended findings, recommendations roadmap, and full per-figure narrative are delivered exclusively to the commissioning client and not shared with other estates in the study, the public, or third parties.

The short version: the public score is visible to anyone. Everything commissioned beyond it is yours alone.

Three things underpin ARA's credibility:

Live testing, not modelling. The A3 Synthetic Recall Test is conducted live across ChatGPT, Perplexity, Gemini, and Claude using a standardised protocol. We record which figures AI systems actually cite, on which platforms, in which contexts. This is not estimated or inferred — it is observed.

Transparent methodology. Every dimension is documented, every scoring rubric is explicit, and every A3 query is disclosed in the deliverable. You can see exactly how each number was arrived at. Read the methodology.

Benchmarked context. A score only means something relative to what's possible and what the cohort is achieving. All ARA studies include full peer scoring — you see where the figure stands, not just what they scored.

If the cohort is already in the public index, you can start there — the figure's score, tier, and dimension breakdown are already live at araco.ai. That's the baseline.

When you're ready to go deeper, submit an audit request — tell us the figure, cohort, and any context about timeline or brief. We'll be in touch within two business days to scope the commissioned study, confirm what's already been covered publicly, and outline what the full private engagement adds.

We don't need access to private archives or internal systems. Our methodology is entirely external — we assess how AI systems represent the figure based on publicly available information and live query behaviour.

The public index is free. If a figure is in a published cohort, their score, tier, and per-dimension breakdown are available at araco.ai at no cost.

Commissioned studies — which include extended A3 testing, the full findings narrative, a prioritised recommendations roadmap, and private delivery — are priced per engagement. Pricing is scoped based on cohort complexity, the size of the peer set, and whether the study includes implementation support or ongoing wave tracking.

A standard commissioned study is priced to be comparable to a mid-range research project — meaningful, but well within a foundation, publisher, or institutional budget. Get in touch and we'll send you a scoping proposal with full pricing within a week.

Studies are typically commissioned by estates, foundations, presidential libraries, museums, publishers, biographers, and the communications teams around living cultural figures. We work with clients who take the AI readiness question seriously — usually those who have already seen AI-driven retellings compress a figure into a caricature, or who want to understand the exposure before it does.

Yes, under a formal partnership agreement. Agencies, publishers, and cultural institutions can integrate ARA methodology into their client offering as a licensed capability. We work with a small number of partners to ensure quality is maintained. Get in touch to discuss partnership terms.

ARA's core product is the audit and the roadmap — we identify what to do, prioritise by impact, and explain the mechanics of why each intervention works.

Implementation support is available as a separate engagement for clients who want hands-on assistance executing specific recommendations — particularly around biographical schema, archival digitisation, and narrative positioning. Ask about this when you request your audit.