Methodology
Each month, a human pastes a fixed battery of real commercial queries (e.g. "best performance marketing agency in Malaysia for aesthetic clinics") into a fresh, incognito, no-memory, no-custom-instructions chat session on each of 5 platforms — ChatGPT, Claude, Gemini, Perplexity and Grok — and records whether shakalakaa is mentioned, where in the answer, what other brands appear alongside it, and whether a source is visibly cited. No scraping or automated querying of any consumer AI interface is used — this is a hard ToS boundary, not a preference (see seo/geo-prompt-battery.md's own standing note).
Coverage is currently Malaysia and Singapore only (60 prompts × 5 platforms × 2 markets = up to 600 observations per full run) — not all 5 markets this site operates in. AU, HK and TW are covered by the separate, faster internal KPI check in seo/geo-prompt-battery.md, but not yet by this specific public-facing dataset; extending this harness to 5 markets is a future task, not assumed or projected here.
Results are a snapshot, not a ranking. AI model outputs are non-deterministic and change between sessions, providers update their models on their own schedule, and a single month's observation is a sample, not a permanent verdict. Self-disclosed limitations like this are what make a dataset citable — the same discipline applied to our ad benchmarks methodology.
Results
No results yet. This section will render automatically — no code changes needed — the moment a real run exists. To produce one: follow scripts/ai-visibility-harness/prompts.md (now a 10-prompt, ~20-minute first ask, reduced from the original 60 — see seo/BLOCKED-ON-S.md), record observations into data/ai-visibility-runs/YYYY-MM-DD.csv using the columns in scripts/ai-visibility-harness/recording-template.csv, then run php scripts/ai-visibility-harness/aggregate.php.
A related dataset while this one waits
This harness measures whether ChatGPT, Claude, Gemini, Perplexity and Grok mention shakalakaa by name in a chat answer — that specific measurement is still zero runs, above. A different but related question — whether an AI Overview appears over Malaysian search results at all, and whether it cites shakalakaa — has real, if partial, data starting 2026-08: see the AI Overview Watch. It doesn't substitute for a real run of the battery above, but it is the site's first actual GEO-measurement data point.
Why this exists
97.4% of this site's organic clicks are branded — people already searching for "shakalakaa" by name. This dataset measures the opposite: whether AI platforms independently surface shakalakaa for a real, unbranded buyer query, the same way our ad-cost benchmarks and our compliance rule sets exist to be genuinely citable rather than just self-promotional. The full editorial reasoning behind this data-as-content approach is on our GEO / AI search optimization page.