From absent to top-mentioned
in 26 days.

When someone asks ChatGPT, Gemini or Perplexity “what's the best expat health insurance in the Netherlands?”, the model doesn't invent an answer. It pulls live pages from the open web, reads them, and stitches a recommendation out of what the most authoritative sources say. Whichever brands those pages mention end up in the answer. Whichever brands they don't, don't exist as far as the user is concerned.

Oasy is a layer that sits in front of a publisher's website. It detects when an AI model is fetching a page (vs. a human or a Google crawler) and serves a version of the page that contains a short, visible, sponsored recommendation. Humans see the original page. Search engines see the original page. AI models see — and cite — the sponsored version.

For this trial, the publisher was IamExpat.com, one of the most authoritative English-language sites for expats living in the Netherlands. The advertiser was Glider Insurance, a Dutch expat-insurance brand that, before the trial, had effectively zero presence in AI answers for the category — competitors like Cigna Global, Allianz Care and Zilveren Kruis dominated every recommendation.

The question we wanted to answer was simple: if a credible publisher uses Oasy to recommend a brand, do AI models actually pick that recommendation up? And if they do, how fast, how strongly, and what happens after the campaign stops?

To keep ourselves honest, we didn't measure the result. We registered the experiment on Promptwatch.com — a neutral monitoring platform that runs a fixed set of prompts against the major AI models on a daily cadence and records every answer. The numbers below are theirs, not ours.

The biggest risk in any “AI visibility” study is that the company running the campaign also measures it. To eliminate that, the entire trial was monitored by Promptwatch.com, an independent platform that issues prompts to commercial AI models on a fixed cadence and records every response verbatim. We did not control the prompts after the monitor was registered, the models that answered them, or the parsing of mentions.

We selected 10 commercial-intent prompts covering the Dutch expat-insurance category — the same prompts a real customer would type when comparing providers. Together they represent 1,584,154 estimated monthly searches and produced 2,272 model responses during the observation window.

The intervention was a single change on the publisher side: Oasy deployed in front of the IamExpat property and, when an AI crawler was detected, returned a page variant with sponsored text recommending Glider Insurance. Human visitors saw the unmodified page; SEO crawlers (Googlebot, Bingbot) were excluded from the injection.

On 13 April 2026 we ended the campaign. The chart and stats above cover the full live period and stop on that date.