Putting the buying details in one place
For our first website experiment, AI helped bring the Nyonya kuih buying details together in one short introduction. The effect on search still needed time and evidence.
On 28 October, after the website measurement checks, I approved a small experiment on our Nyonya kuih page. The idea was to put the local buying information into one short, visible answer. AI helped prepare the wording and the checks for it. We kept this first change to the introduction on one page.
The information was already on the website. The old introduction described ingredients and preparation; the pickup location, rotating selection and order confirmation details appeared in different places. The new paragraph brought those details together for someone asking where to buy Nyonya kuih in Sydney. We were reorganising information we had checked, without adding a new promise about what the kitchen could supply.
The wording needed care. It said the kitchen offers Nyonya kuih, also spelled kueh, rather than saying we make every item in the selection. The menu at the time included a partner-supplied kuih. It directed readers to the dated menu for the listed selection, box size, price and pickup time, and to ask the kitchen about future availability if no batch was listed.
The same paragraph explained the request process. A reader prepares and reviews a text request, then sends it from their own messaging app. The kitchen’s reply confirms the order. The website does not send the request or take payment. Keeping that explanation beside the buying details mattered to the accuracy of the page, whether or not the experiment later changed search traffic.
The approved change went live that evening. The agent checked the actual page and its mobile layout, while preserving the existing buying controls, images and other page features. The experiment record also set out what to observe before looking at results: a fixed search-click measure, an initial period to check search-engine adoption, and a later comparison window. It left room for an inconclusive answer.
At that point, we had a verified content change. We did not yet have evidence that search engines had adopted the new text, that AI answers cited it more often, or that it brought more orders. AI had helped turn a broad wish to improve the website into one specific change with a record we could return to. The next step was to observe what happened, keeping the limits of that comparison visible.
