IBM · Filed Jan 31, 2025 · Published Aug 6, 2026 · verified — real USPTO data

IBM Patents a System That Surrounds Your Real Searches With AI-Generated Decoys

Every time you search for something online, you're leaving a trail that advertisers and data brokers can follow. IBM has patented a way to muddy that trail by having an AI generate convincing fake searches that travel alongside your real ones.

IBM Patent: AI-Generated Decoy Searches Hide Your Real Intent — figure from US 2026/0228435 A1
Figure from the official USPTO publication.
See all 5 drawings from this filing ↓
Publication number US 2026/0228435 A1
Applicant International Business Machines Corporation
Filing date Jan 31, 2025
Publication date Aug 6, 2026
Inventors David James Beymer, Dinesh C. Verma, PAWAN RAGHUNATH CHOWDHARY, Swanand Ravindra Kadhe
CPC classification 704/9
Grant likelihood Medium
Examiner GODBOLD, DOUGLAS (Art Unit 2655)
Status Docketed New Case - Ready for Examination (Mar 4, 2025)
Document 20 claims

How IBM's fake-search privacy shield actually works

Imagine you spend an afternoon searching for symptoms of a medical condition. By the end, any company watching your search history has a pretty good idea of what you're dealing with. That kind of profiling is exactly what this IBM patent is designed to prevent.

The system generates a stream of fake searches, called decoys, that run alongside your real ones. The clever part is how those decoys are built: instead of random noise, the AI pays attention to how your real searches are shifting over time and mimics that same pattern of drift in a completely unrelated topic. So if your real searches move from "back pain" toward "surgery options," the decoy searches might shift in a parallel way through, say, gardening topics.

The result is that anyone watching your query history sees two plausible-looking stories at once. They can't easily tell which one is you.

How the AI mirrors your search pattern in a fake topic

The patent describes a system that tracks the direction and distance between consecutive real search queries using a technique called vector embeddings. A vector embedding is a way of turning a word or phrase into a list of numbers that captures its meaning, so that concepts close in meaning end up close together in a mathematical space.

When you go from searching "lower back pain" to "herniated disc symptoms," the system calculates the gap between those two points in that meaning-space. That gap is called a difference vector. Think of it as an arrow pointing from one idea to another.

The system then takes that same arrow and applies it to a completely different starting point in an unrelated topic area, the previous decoy query. The result is a new decoy query that has moved the same conceptual distance in the same conceptual direction as your real query, just in a different subject entirely.

This process repeats for each new real query you make, so the decoy sequence stays internally coherent and convincing. An observer watching both streams would see two evolving, believable lines of inquiry, with no reliable way to determine which is genuine.

We find one patent like this every day. Get the best of each week in your inbox, free →

What this means for search privacy and data brokers

Search history is one of the most valuable data assets advertisers and data brokers collect. A sequence of searches can reveal health concerns, financial stress, relationship problems, or political views. Most existing privacy tools either block queries entirely (breaking functionality) or add random noise (which is statistically easy to detect and strip out).

IBM's approach is different because the decoys aren't random. They're structurally identical in shape to your real searches, just anchored to a different topic. That makes them much harder to dismiss statistically. For everyday users, this kind of protection could eventually show up in browsers, VPNs, or enterprise privacy tools, anywhere that search-query logging is a concern.

Editorial take

This is a genuinely thoughtful privacy patent. The insight that decoys should mirror the *pattern* of real queries, not just add random junk, is the kind of detail that separates a real engineering idea from a defensive filing. Whether IBM ships this as a product or licenses it is another question, but the core concept is worth paying attention to.

There are more where this came from

We read every patent application Big Tech publishes and send you the ones worth knowing. Plain English, free, every week.

The drawings

5 drawing sheets from US 2026/0228435 A1 · click any drawing to enlarge

Patent filing page

Source. Full patent text and figures from the official USPTO publication PDF.

Editorial commentary on a publicly published patent application. Not legal advice.