AdUp case studies: what we found and fixed on client websites
Real tasks from our work and results in numbers. We don’t name clients, only their industry and scale.
Paid orders were being cancelled and analytics counted purchases twice
The task
The owner saw many cancelled orders in the CRM, while Google Ads reported purchases that never happened. There was no way to judge the ads.
What we did
- We matched cancelled orders against the LiqPay payment system: of 70 orders cancelled in 30 days, 65 had actually been paid.
- We found the cause: payment confirmations weren’t reaching the site, so orders were cancelled automatically.
- We added a safeguard: before cancelling, the site checks the payment in LiqPay itself.
- We rebuilt purchase tracking in GA4 and Google Ads and connected the Facebook pixels.
Results
Lead forms were silent for three months while bots loaded the server
The task
The store got few leads from its forms, the admin panel was slow and spam was piling up in the database.
What we did
- We found a server rule that had broken the "Custom quote", "Become a partner" and "Quick order" forms since at least late July, and fixed it.
- We stopped a bot that tried to register about 40 times an hour, each attempt tying up the server for 3–4 seconds.
- We blocked spam sent through forms bypassing JavaScript and removed what had built up.
- We investigated a case when the store took 48 seconds to respond: a full PHP cache. We increased it and speed returned.
Results
GEO optimisation: ChatGPT and Perplexity couldn’t even open the store
The task
The owner wanted the store’s products and categories to appear in ChatGPT and Perplexity answers and in Google’s AI Overviews. First we had to find out what AI services could actually see.
What we did
- The audit showed that a Cloudflare setting blocked ChatGPT, Claude, Perplexity and other AI bots with a 403 error. We changed the rule, and the bots now get the full page.
- We found duplicate product markup: every page had two identical Product blocks. We kept one and added the brand.
- We wrote FAQs for 14 categories in Ukrainian and Russian: 5–6 questions with direct answers, a visible block and FAQPage markup.
- We sent each product’s brand to the Google Merchant Center feed, taking into account that the catalogue is multi-brand.
Results
GEO optimisation: 191 completed projects that neither Google nor AI knew about
The task
The manufacturer has hundreds of installations in sanatoriums, hospitals and rehabilitation centres, but the project pages were made of photos only. There was no text at all for search engines or AI services.
What we did
- We opened the site to ChatGPT, Perplexity, Claude and Google-Extended bots in robots.txt and added an llms.txt in Ukrainian and English.
- We added Organization and Product markup to every equipment page, including UAH prices that had only existed as text inside a tab.
- For each of the 191 projects we identified the equipment from photos and the city, wrote a short description, added venue markup and "Where installed" links from product pages.
- We added FAQs to 5 key categories in three languages and alt text to images — none of the 1,206 files had any.
Results
A slow website and a catalogue Google couldn’t see
The task
The site loaded slowly on phones, and the energy storage catalogue was built in React, so search engines didn’t see it as regular text.
What we did
- We deferred reCAPTCHA until the first interaction and removed an unnecessary GTM plugin.
- We set up caching and lazy loading of images.
- We moved the catalogue to plain server-rendered HTML so Google can read every model.
- We added "Use cases" and "How to get a system" sections and FAQ markup.
Results
Want the same for your website?
An audit of your website, ads or AI visibility is free. Implementing fixes is paid separately once the estimate is agreed.
Get a free audit →