From an afternoon with GeoEasy to GIG's own AI-visibility toolset
UAE buyers have stopped searching and started asking. When someone asks ChatGPT "what's the best car insurance in Dubai?", the answer names two or three insurers and moves on. If GIG is not one of them, the buyer never learns it existed — and unlike a Google results page, that answer is one we can influence but cannot see. This is the story of how we made it visible.
An afternoon with Azam and GeoEasy
Azam sent us GeoEasy, and it was exactly the right thing to be looking at. We spent an afternoon with it — and by the end of the day we had stopped evaluating a tool to rent and started building one to own: AXA-aware, GIG-specific, about 250 lines of Python we can read and you can keep.
What that afternoon turned into is everything below — a working measurement engine, a reverse-engineering of Visiblie's pilot, a market forecast, a prices-into-Gemini playbook, and the prototypes and insights that fell out along the way.
The shift
AI answer engines now sit between the buyer and the brand. The shelf has moved: where a buyer once scanned a page of results, they now read a single synthesised answer that names a few brands and stops. For an insurer that is the whole game — to be named in that answer at the moment of intent. The problem is that, unlike search, almost none of it shows up cleanly in analytics. We have to make it visible before we can manage it.
GIG's own analytics already carry a "Generative AI Sessions" report. In the 28 days to 17 June it logged 2,336 sessions arriving from an AI answer — but look at where they came from:
Almost all of it is ChatGPT, because ChatGPT links out. Gemini, Perplexity and Copilot barely register — not because GIG is absent from them, but because they answer the buyer without ever sending a click. Our probe shows GIG is cited across all of them; the traffic simply never arrives to be counted. GA4 can see the click; it cannot see the answer. That gap is the reason a deliberate measurement tool has to exist.
What we measure — the honest numbers
The early passes flattered GIG by counting AXA legacy as GIG's own and reading a single lucky pass. Run 3 is the rigorous one: 100 Motor queries across five answer engines, each run three times — 1,500 calls — with AXA counted separately and a 95% confidence interval on every figure.
How we got there
That number is lower than our first pass — and every step down was rigour removing inflation, not GIG losing visibility. Run 1 counted AXA as GIG and a brand "mentioned anywhere"; Run 2 split AXA out; Run 3 added the weak parametric engines, three repeats and a confidence band.
And the blended 51% hides the real finding. Split the engines by what they read and a chasm opens: on the conversational LLMs answering from memory — ChatGPT, Claude — GIG is named only ~18%, with AXA filling the space (94% on Claude); on the engines that read the live web — Perplexity, Google AI Overviews — GIG is named 62–79% and its own pages are cited 66–76%. The full working, engine by engine and intent by intent, is in Run 3.
By intent, GIG owns branded and competitive ground (87%, 81%) but falls away on the high-intent service moments — claims (35%) and eligibility how-tos like black points and traffic fines (29–39%). Those are buyers in-market now, and they are exactly where GIG's content is thin. That is the build list.
Reading Visiblie
The same afternoon's work let us do something a rented dashboard never could: take a vendor's number apart. Visiblie's pilot put GIG at a 12.6% AI citation share; ours puts GIG at 51% included. The full reconciliation is here — and the honest finding is that the two mostly measure different things and roughly agree once normalised (51% presence ÷ ~5 brands per answer ≈ 10% share, close to their 12.6%). What our tool caught that theirs did not is the AXA blind spot. The win is not a bigger score; it is that we can decompose any score on demand.
The tool
It is one owned engine — about 250 lines of Python, open to read — that asks every answer engine the questions a GIG buyer would ask, week after week, notes whether GIG and its rivals are named, credits AXA back to GIG, and scores the share of the answer we win. It runs on five engines — ChatGPT, Claude, Perplexity, Google AI Overviews and Google organic — and reads both what the models remember and what the live web says. The full capability stack — what each engine unlocks, and what the tool still cannot see — is on the tool page.
Where buyers are going
ChatGPT carries 96% of GIG's captured AI traffic today — but its share of the assistant market is falling as Gemini and Google AI Overviews rise, and on our forecast Gemini overtakes ChatGPT around 2028. That matters because Gemini is grounded in Google's index, where GIG already ranks — so GIG's SEO equity reasserts itself in AI answers. The model-share forecast sets out the scenario; the prices-into-Gemini playbook is the concrete move it points to.
What it needs to be operational
As it stands the tool is a credible weekly read, not yet a board-grade score: four engines, a single pass, a yes/no on whether GIG is named and linked. To operationalise it we would run each question several times so the number carries a confidence band rather than noise, widen beyond Motor to the other lines, capture the full citation set so we know whose content is winning, and pull GIG's crawler logs to confirm the engines even fetch our pages. That is the whole of the gap — and none of it requires renting anyone's platform.