> Corrections
Correction log
When we get something wrong, we say so here, in public, with the figures. A benchmark whose author corrects quietly is not a benchmark.
This log covers the Gr8 Index and the GR8ER2 AI Visibility Standard. It opens at CORR-002: CORR-001 is an internal retraction from our own research notes that was never published on this site, so there is nothing public to correct.
The measurement data behind every figure in the Gr8 Index is published as a dataset: download the CSV ·schema and recompute formula. It is counts and positions only — no answer text — and it recomputes all 67 published Gr8 Scores exactly.
> CORR-002 · 11 November 2026
CORR-002 — Correction to tie reporting in the Gr8 Index
In short. GR8ER2 is correcting how its Gr8 Index reported statistical “ties” between brands. The original method grouped brands together too readily, so some brands that are in fact clearly separated were shown as “tied.” The method has been corrected. No brand’s score has changed — only the way ties and the margin of error are reported. The data underlying the Index is the publicly observable output of four independent AI assistants, which anyone can reproduce by asking the same questions; the measurement data behind every figure, and the method, are published — and the complete answer log is retained for independent verification on request.
What this notice is
This is a correction to how the GR8ER2 Gr8 Index reported statistical ties between brands. It is issued by GR8ER2 and applies to the Gr8 Index v1 (published October 2026), across all four categories. The figures are presented as measurements of independently observable AI output — not as findings, allegations or opinions about any institution.No Gr8 Score has changed; this correction affects only the reporting of ties and intervals, not any underlying measurement.
What was wrong, in plain terms
The original Index grouped brands into “tie bands” using a method that chained overlapping intervals: if brand A’s interval overlapped B’s, and B’s overlapped C’s, then A, B and C were reported as one tie — even where A and C were, in fact, clearly separated. The effect was to overstate how many brands were statistically indistinguishable, and in some categories to collapse a whole league into one wide “tie band.” That is a conservative error in one direction — it understated the separations the data actually supports — but it is an error, and it is corrected here.
What changed in the method (Standard v1.2)
v1.2 is a reporting-layer correction only (a minor version under the Standard’s governance; scores remain directly comparable to the prior version):
• Ties are now pairwise, never chained. Two brands are reported as separated only when the 95% confidence interval of the paired difference between them excludes zero.
• The interval now states exactly what it measures: run-to-run variation of this panel, on this question set, on the measurement date, computed within each assistant.
• Panel sensitivity is disclosed for each score, and any ranking that reverses if a single assistant is dropped is flagged.
• Every figure is labelled with its surface and scope: API Panel · 4 engines × 8 runs · 1 day · same-day repeats.
• A distinction is drawn between “clearly ahead of” (every assistant agrees on the direction) and “ahead on panel mean only” (the panel average separates them but the assistants disagree on which is ahead). Only the former is stated as a clear ordering.
How and when the data was obtained
On 1 October 2026, GR8ER2 put a fixed, published set of buyer questions to four independent AI assistants — ChatGPT (OpenAI), Claude (Anthropic), Gemini (Google) and Perplexity — asking each question eight times per assistant, logged out and set to South Africa. Every answer returned was recorded in full. The per-run measurement data from which every figure is computed — each brand’s mention, citation and position, by engine and by run — is published as a dataset, so that any figure in the Index can be independently recomputed; and the complete answer log is retained and available for independent verification on request. Any figure can also be re-tested directly by putting the same questions to the same assistants. GR8ER2 did not generate the content of those answers; it recorded the publicly observable output of the independent AI systems. The method and the verbatim questions are published in full: the Standard.
The source of the underlying information
The information underlying this Index — which brands the AI assistants name, and how often — is produced by those independent, third-party AI systems, and the institutions are identified by reference to publicly available regulatory records (banks and insurers on the Prudential Authority register; medical schemes on the Council for Medical Schemes register). GR8ER2 does not independently investigate, collect or assert any facts about the institutions named. Its role is to observe, record and analyse that publicly observable third-party output. Accordingly, any statement about a named institution reflects what those independent systems produced on the measurement date — not an assertion or investigation by GR8ER2.
The restated ties — regulated institutions first
Each statement below reports a measured difference in how often the AI assistants named each brand, on the measurement date, to the published question set. These are measures of AI visibility — how often a brand appears in AI answers — and are not measures of any institution’s products, services, financial standing or conduct.
Rounding. Gr8 Scores are shown rounded to whole numbers; paired differences and intervals are computed from unrounded values.
Retail banking
Originally, the five brands from Nedbank down to Bank Zero were reported within a single tie band. They are not tied: Nedbank (33) is separated from Bank Zero (24) — a paired difference of +9.3 points (95% interval 7.6 to 11.0). Restated, the category resolves into nine groups (largest two brands); 94% of brand pairs separate cleanly, with Capitec, First National Bank and Standard Bank each standing alone at the top.
Short-term insurance (car & household)
Originally, eight brands from King Price down to Budget Insurance were reported within a single tie band. They are not tied: King Price (39) is separated from Budget Insurance (19) — a paired difference of +19.9 points (95% interval 17.9 to 21.9). Restated, the category resolves into ten groups (largest three); 92% of pairs separate cleanly, with OUTsurance standing alone at the top.
Medical aid / health insurance
Originally, four brands from KeyHealth down to Profmed were reported within a single tie band. They are not tied: KeyHealth (11) is separated from Profmed (5) — a paired difference of +5.8 points (95% interval 4.6 to 6.9). Restated, the category resolves into nine groups (largest two); 98% of pairs separate cleanly, with Discovery Health standing alone at the top.
AI-visibility / digital-marketing / SEO agencies
Originally, 26 of the 28 brands were reported within a single tie band. Restated, the category resolves into sixteen groups (largest six); its two extremes, which were in one band, are 20.0 points apart (95% interval 18.6 to 21.5), and 87% of pairs separate. Conflict note (see below): GR8ER2 competes in this category, so this restatement has been independently recomputed and the category carries a standing conflict disclosure.
The complete restated tables, with every brand, interval, panel-sensitivity range and tie grouping, are published alongside this notice: the Gr8 Index.
What is measured fact, and what is GR8ER2’s interpretation
The measured facts are the counts and the paired statistical comparisons, drawn directly from the published measurement data. Where GR8ER2 describes one brand as “clearly ahead of” another, that is stated only where every assistant agrees on the direction and is supported by the published figures and method; where the assistants disagree, it is labelled “ahead on panel mean only” and is not presented as a settled ordering. Any such description is GR8ER2’s interpretation of the referenced data and should be read as such.
The limits of this measurement
• It is a single-day snapshot. All runs were taken on one date; they are same-day repeats. This measures how the AI assistants answered on that date and has not yet been re-tested over time. A score should be read as “as at this measurement,” not as a settled or permanent position.
• It is the API Panel. Figures are measured through the four assistants’ public interfaces with web search enabled; they do not claim to represent every consumer-facing AI surface.
• The assistants disagree. Where a separation depends on the panel average while the assistants split on direction, it is labelled as such and not stated as a clear ordering.
• This is an analysis and interpretation of independently observable AI output, offered in that context.
Conflict disclosure
GR8ER2 is an AI-visibility consultancy. It has no clients among the banks, insurers or medical schemes named in this Index. It does compete in the AI-visibility/agency category; that category is retained only with an independent recomputation and this standing disclosure.
Sources and referencing
The figures are derived entirely from the recorded answers of the four AI assistants. The per-run measurement data from which they are computed is published and recomputable: the measurement dataset. The complete answer log is retained and available for independent verification on request. The verbatim question set and the full method are published: the Standard. The named institutions are identified from their public regulatory records — the Prudential Authority register (banks and insurers) and the Council for Medical Schemes register (medical schemes). The complete restated tables are published alongside this notice: the Gr8 Index.
Disclaimer
This publication and this correction have been prepared from publicly observable AI-assistant output and publicly available information. The underlying data is the output of independent, third-party AI systems and public records; GR8ER2 reports and analyses that information as found and does not warrant the accuracy or completeness of the underlying third-party information. Sources are identified wherever practicable and the underlying measurement data is published so that any figure can be independently recomputed, and the complete answer log is retained for verification on request. Any analysis, interpretation or commentary represents GR8ER2’s interpretation of that information and should be read in that context. The figures measure AI visibility — how often brands appear in AI answers — and are not statements about any institution’s products, services, financial standing or conduct.
Produced to the GR8ER2 AI Visibility Standard v1.2. Correction log reference: CORR-002.
See the restated figures
Every brand, interval, panel-sensitivity range and tie grouping, restated under v1.2.