Most people search tools are glorified lookups. You type an email, they query a few static databases, and they hand back whatever rows happen to match. The moment the trail forks — a username that does not match the email, a profile photo taken somewhere revealing, a breach record pointing at a second address — a flat lookup stops cold. Revealer's AI Deep Search was built to keep going. It is an autonomous, agent-driven people intelligence engine that treats a single identifier not as a query but as a seed, then recursively pivots across a subject's entire digital footprint until the leads run dry.
This is the difference between an AI people search and a database scrape. Deep Search does not return rows. It runs an investigation.
What AI Deep Search Actually Is
At its core, AI Deep Search is a recursive AI OSINT system: a multi-agent loop that ingests one starting signal — an email, a username, a phone number, a name — and expands it into a graph of correlated identities. Each new fact the engine discovers becomes a fresh starting point for the next pass. Breach-derived emails surface usernames, usernames surface profiles, profiles surface photos and bios, and bios surface yet more handles. The loop keeps pivoting until additional passes stop producing new, high-confidence signal.
We call the output people intelligence rather than search results for a reason. A search result is a row. Intelligence is a synthesized, sourced, confidence-scored picture of a person assembled from many weak signals that no single source could have produced alone.
The Recursive Multi-Agent Loop
The engine is not one model answering one prompt. It is a coordinated set of autonomous AI agents, each specialized, each feeding the others.
Seeding the investigation
Every investigation starts with normalization. The seed identifier is parsed, validated, and classified — is this an email, a handle, a phone number, a real name? That classification routes the seed to the right entry agents. An email goes straight to breach correlation; a username goes to handle-semantics decoding and cross-platform enumeration.
The pivot graph
From there the loop runs in passes. On each pass, the orchestrator hands the current set of known signals to the specialist agents, collects whatever new identifiers and artifacts they return, deduplicates them against what is already known, scores the new edges, and decides whether any of them are strong enough to seed another pass. This is recursive search in the literal sense: the output of one iteration is the input to the next, and the graph grows breadth-first across a subject's accounts rather than stopping at the first match.
The key engineering decision is knowing when to stop. Recurse too shallowly and you miss the second-order accounts that make an investigation valuable. Recurse forever and you drown in noise and false positives. Deep Search uses confidence thresholds on every new edge to prune aggressively, so the graph expands toward the subject and away from coincidental collisions.
Breach-Derived Identifier Pivots
The richest pivots come from breach and stealer-log data. When the engine has an email, it cross-references it against a continuously updated corpus of breach records. Those records routinely expose the other identifiers a person used — an old username, a recovery phone number, a secondary email, a password pattern that itself hints at handles elsewhere.
Each of those becomes a new seed. A breach-derived username that a flat search would never connect to the original email gets fed straight back into the loop, where cross-platform enumeration turns it into live profiles. This is where Deep Search routinely finds accounts the subject themselves may have forgotten they created.
Decoding Handle Semantics
Usernames are not random. People reuse stems, append birth years, swap a zero for an O, or carry the same alias across a decade of platforms. AI Deep Search runs a handle-semantics pass that decodes these patterns — separating the meaningful stem from the noise, recognizing common mutation rules, and generating high-probability variants to test across platforms.
That semantic understanding is what lets the engine connect coolmike_88 on one service to cool.mike on another and mike1988 on a third, then assign a calibrated confidence to each link rather than treating them as unrelated strangers.
Live Web-Grounded Search Across 200+ Platforms
Static databases go stale. To stay current, Deep Search self-directs live, web-grounded searches across 200+ platforms — social networks, forums, marketplaces, developer sites, gaming services, and niche communities. The agents do not run a fixed checklist; they decide which platforms are worth querying for this subject based on the signals already gathered, then ground their findings in real, retrievable sources rather than model recall.
Web grounding matters because it anchors every claim to something verifiable. The agent is not guessing that a profile exists — it is reading the live page and capturing the evidence.
Multimodal Image Forensics
When the loop surfaces profile imagery, a multimodal forensics layer goes to work. The engine analyzes photos for environmental and location signals: signage and language, architecture, vegetation, license-plate formats, interior cues, and other context that hints at where a subject lives, works, or travels. It also compares faces across profiles to strengthen or weaken identity links. A single photo can collapse weeks of ambiguity, and Deep Search treats imagery as a first-class signal rather than decoration.
Neural Correlation and Confidence Scoring
This is the pass that turns a pile of artifacts into intelligence. A neural correlation layer cross-references every signal the loop produced — emails, handles, phones, profiles, images, breach records — and scores how strongly they belong to the same person. Reinforcing evidence raises confidence; contradictions lower it.
Crucially, nothing is presented as fact without a number attached. Deep Search produces a confidence-calibrated dossier, so an analyst can instantly separate a near-certain identity link from a speculative one. That calibration is what makes the output usable in real work rather than a wall of maybes.
Source Provenance: Every Claim Is Traceable
An intelligence product you cannot audit is a liability. Every node and edge in a Deep Search dossier carries full source provenance — which breach, which platform, which live page, which image a given conclusion came from. You can trace any claim back to its origin, verify it independently, and defend it. Provenance is what separates AI OSINT you can act on from a black box that simply asserts.
Why Recursive Beats Flat Search
A traditional AI people search answers one question and stops. Deep Search asks the next question automatically — and the one after that. The recursive loop, breach-derived pivots, handle-semantics decoding, live web grounding, multimodal forensics, and neural correlation are not separate features bolted together; they are stages of one continuous investigation that compounds. Each layer makes the next one sharper, and the result is a picture of a digital footprint that no single lookup, however good its database, could assemble.
That is the whole idea behind people intelligence: stop returning rows, and start delivering an answer you can trust and trace.
Want to see recursive people intelligence run on a real identifier? Explore Revealer's AI Deep Search and watch a single seed expand into a fully sourced dossier.