The Pitot Index® is designed to measure how leading artificial intelligence systems evaluate and recommend brands. Its methodology is built around privacy, independence, standardisation, comparability and transparency.
Pitot does not access, purchase or analyse private user conversations with AI systems.
The Index does not depend on personal chat histories, individual user prompts or behavioural data. Its purpose is to measure the behaviour of AI systems, not the behaviour of the people using them.
Pitot's measurement does not require personally identifiable information or private user data.
Prompts used for Index measurement are created specifically for the research process and are independent of individual users.
This allows the Pitot Index® to measure AI brand perception without observing or profiling the people interacting with AI.
There are fundamentally different ways to measure brands within AI.
Observational measurement starts with real-world user behaviour. It examines what people ask AI, which brands appear in those conversations, how frequently they are mentioned and which sources AI systems cite. This can provide valuable insight into AI visibility and user demand.
But observational measurement also introduces the characteristics of the observed population into the measurement itself. Geography, language, demographics, cultural context and differences in user behaviour can influence which questions are asked and therefore which brands and attributes are measured.
Pitot uses controlled measurement.
ARES-C establishes a standardised evaluation framework independently of individual user behaviour and applies that framework consistently across brands and AI models.
The distinction is important:
Observational measurement asks what is happening in AI conversations.
Controlled measurement asks what happens when AI systems are subjected to the same test.
Both can produce useful intelligence, but they measure different phenomena.
Pitot uses the controlled approach because the objective of the Index is comparability: comparing brands against brands, models against models, cultural perspectives against cultural perspectives, and one measurement period against another.
Comparability requires consistency.
Brands within an Index are evaluated using a controlled and standardised measurement framework. The underlying questions, evaluation criteria and scoring methodology are designed to minimise differences introduced by the test itself.
The objective is simple:
Change the brand or the AI model — not the rules of the test.
This enables meaningful comparison between brands, models and measurement periods.
No single LLM represents artificial intelligence as a whole.
Different models are trained on different datasets, developed by different organisations and influenced by different architectures, policies, languages and information environments.
For this reason, the Pitot Index® measures brands across nine leading AI models, rather than treating the output of one platform as representative of AI generally.
Individual model observations are combined through the ARES-C methodology to produce a broader measurement of AI brand performance.
The Index therefore seeks to identify not simply what one AI system says about a brand, but the degree to which that assessment is shared across the wider AI ecosystem.
Each of the nine models runs under a pinned version. The model actually served is recorded separately from the model requested.
This is how the Index detects when a provider changes what sits behind an API name without announcing it. Substitutions are detected, dated and registered, and the record is read across them accordingly.
A measurement is only comparable over time if the instrument can prove what it measured with.
AI systems do not exist in a culturally neutral environment.
Their training data, languages, development environments and information ecosystems can influence how brands, products and organisations are represented and evaluated.
Pitot treats this variation as something to measure, rather than noise to eliminate.
By applying a consistent evaluation framework across models originating from different technological and cultural environments, ARES-C can identify areas of convergence and divergence.
This provides an additional dimension to the Index:
Where does AI broadly agree about a brand — and where does that assessment change across models and cultural perspectives?
Pitot does not interpret model origin as proof of a particular cultural bias. Differences are measured empirically and reported where the evidence supports them.
The Pitot Index® does not attempt to infer AI brand performance from what users happen to ask.
Search frequency, prompt popularity and brand mentions can be valuable measures of visibility and consumer interest, but they answer a different question.
Pitot asks:
When brands are evaluated under comparable conditions, how do AI systems assess and recommend them?
This separation prevents variations in user populations, query behaviour or geographic sampling from defining the underlying measurement framework.
The Pitot Index® is unprompted. The models are never told which brands to discuss.
They surface, rank and characterise brands on their own, and the Index records what they surface. It measures salience — whether the AI systems bring a brand up at all, and how they place it. This is what makes the Index uninfluenced and independent: there is nothing in the instrument for a brand to be inserted into.
Pitot’s commissioned assessments — Pulse and the Audit — are prompted: the commissioning brand is named, and the instrument measures how the systems characterise it when asked directly.
The same brand will score profoundly differently under the two modes. That is the method, not an error.
An unprompted ranking is never compared with a prompted assessment — by Pitot, or in anything Pitot publishes.
ARES-C is named for what it reads. Every measured brand is assessed on five dimensions, and a brand’s score is their combination — not a count of mentions.
Four dimensions are asked. The fifth — agreement — is what the answers reveal when placed side by side.
Being visible to AI and being recommended by AI are not the same thing.
A brand can appear frequently in AI-generated answers while being evaluated less favourably than its competitors.
The Pitot Index® is therefore designed to go beyond measuring mentions, citations or share of voice. ARES-C examines the comparative strength of brands within AI-generated evaluation and recommendation.
This distinction becomes increasingly important as AI systems move from answering questions to assisting with — and potentially acting within — purchasing and decision-making processes.
The Pitot Index® is a weekly measurement. It is not a quarterly one.
ARES-C closes one benchmark cycle every week. Each cycle runs to a fixed weekly schedule and is identified by its week (for example, 2026-W38). Every measured brand in every sector is re-measured in every cycle, on the same instrument, in the same order. Nothing is sampled between cycles and no sector is measured on a different clock from any other.
The published record — standings, scores, movement and the news wire — is refreshed as each cycle closes. Movement on the Index is therefore week-on-week movement, and a sector’s history is a continuous series of weekly cycles rather than a handful of periodic snapshots.
Reporting is quarterly; measurement is not. The in-depth industry reports are published four times a year and cover the full quarter’s record — every cycle in the quarter, brand by brand, with the analysis and the data integrity statement. The quarterly report is the publication of analysis written on top of the weekly record. It is not the rate at which brands are measured.
Where the cadence itself changes, that change is a material methodological change and is recorded in the register described below, exactly as a change to the instrument would be.
Nothing in the live record is smoothed, imputed or carried forward. Each weekly cycle is published exactly as the instrument measured it.
A sector’s cohort is re-established in every cycle by unprompted salience. A brand at the cohort boundary can therefore leave the table one week and return the next: that is the models disagreeing at the edge, and the record shows it rather than hiding it.
The quarterly industry reports read the same record through a rolling multi-cycle window, so a reported rating never rests on a single week.
An Index becomes valuable when movement over time means something.
Pitot therefore prioritises methodological consistency between cycles. Changes to prompts, model configurations, scoring methodology, the measurement cadence or other material components of ARES-C are controlled and documented.
This allows movements in the Index to be interpreted against a stable baseline rather than being unknowingly produced by changes in methodology.
Where a material methodological change affects comparability with previous measurements, Pitot will identify that limitation.
Because the instrument is frozen between cycles, its noise can be measured directly.
Across tens of thousands of consecutive week-to-week comparisons — same brand, same question, same model, unchanged instrument — the most stable models in the basket change their committed answer in fewer than one case in thirty. The most volatile change it in roughly one case in five to eight.
Reported movement is only meaningful against an instrument’s own noise floor — which is why Pitot measures its own.
The Pitot Index® is intended to function as an independent benchmark.
A brand's position is determined by the ARES-C measurement process. Commercial relationships, report purchases or participation in Pitot services do not influence Index scores or rankings.
Paid access provides additional intelligence and analysis. It does not provide a mechanism to alter the underlying measurement.
Every completed cycle is validated, checksummed and frozen into an immutable archive before publication. The published record cannot be edited after the fact.
Data flows one way, from the instrument outward. No public system can reach the instrument.
Operational anomalies — a model retired by its provider, a voice failing a cycle — are detected, dated and kept in the instrument’s permanent internal record alongside the full audit trail.
Pitot does not publish the prompt texts, their number, the persona statement, the label sets, the score mappings, or the operational record behind the archive.
These are the instrument’s core. Published in full, they would let anyone run a lookalike whose numbers appear comparable while being nothing of the kind — which damages the one thing a baseline is for.
Everything needed to judge whether the measurement is controlled and repeatable is on this page. The rest stays in the vault.
Bona fide researchers and auditors can request supervised access to the full methodology and record, under agreement.
Pitot does not measure what people ask AI about brands.
Pitot measures what AI says about brands when brands are evaluated under the same conditions.
By combining controlled measurement, standardised evaluation, cross-model comparison and analysis of cultural variance — without relying on private user conversations — the Pitot Index® aims to provide a consistent global baseline for understanding how artificial intelligence evaluates and recommends brands.
