At Extract Summit 2025, Domagoj Marić asked the web scraping community to look harder at what it was collecting. His talk, "Scraping a Synthetic Web: Dead Internet Theory Meets Web Data Extraction", began with an idea that still sounded faintly like an internet joke: what if the web was becoming less human?
Then he started researching it. The joke stopped being funny.
The 2026 Imperva Bad Bot Report found that automated bots accounted for more than 53% of web traffic in 2025. For Domagoj, AI customer delivery manager at Pontis Technology, that number sits inside a larger problem involving trust, provenance, fraud, and the ease with which generative AI can support scams or manipulate public conversation.
When I caught up with him ahead of Extract Summit 2026, he described his latest idea as a spiritual successor to that 2025 talk. Last year, he looked at a web crowded with synthetic activity. This time, he is turning the lens toward the humans inside it.
AI turns a scattered footprint into a profile
Open-source intelligence (OSINT) uses publicly available information to answer questions about a person, company, or event. None of the pieces needs to look alarming on its own: an email address on one platform, a conference photograph on another, or a personal detail exposed in a data leak. The danger appears when those pieces are connected.
AI makes that connection faster and easier to scale. Combined with web search and scraping, it can help someone build a convincing profile for a phishing attempt, social engineering campaign, deepfake, or voice-cloning attack.
"AI is just an upgrade on the whole thing," Domagoj said. OSINT and social engineering existed long before the current generative AI wave, but the cost and skill needed to use them at scale have fallen sharply.
That is why this subject matters to the web scraping community. Search sits at the heart of OSINT, and modern large language models increasingly arrive with web search or scraping capabilities. The same lead-generation platform built to help a sales team find a contact can become one input in a much less benign workflow.
The same tools can show you what is exposed
Domagoj is careful to frame this as a dual-use problem. Scraping can collect information that somebody never expected to see assembled. It can also help people defend themselves.
An individual can search for the information available about them or their family. A security team can examine the public footprint of its employees and company. The tools are similar; the intent is different. Seeing the full picture is often the first step toward reducing the risk.
This is the thread connecting Domagoj's 2025 talk with his proposed follow-up. The first examined whether the material we scrape is still trustworthy, a theme that stood out in Zyte's five takeaways from Extract Summit 2025. The new idea asks what our own public data can reveal when AI joins the dots.
The conversations you cannot get from documentation
For Domagoj, that exchange of perspectives is also the reason to attend an event in person. Technical documentation and articles are available from home. What he values is meeting people working on similar problems, hearing how they approached them, and letting those conversations change his own view.
That experience helped shape this new angle: from the bots filling the web to the human footprints left among them. As Zyte's recent post on why Extract Summit is worth leaving your desk for puts it, the best material may not happen on stage.
_HFpro5d6k3.png&w=256&q=75)
_E4PyVpfAxa.png&w=256&q=75)
-(1).png&w=1920&q=75)
-(1)_VZGHqxCgXV.png&w=1920&q=75)
