ai source integrity
See what they’re teaching AI about you.
w3p is the AI source integrity platform. web proof / we protect. We watch the web that agents read, and we prove what was engineered into it.
There are two webs now.
The web people read: rendered pages, pixels, layout, the version you QA and publish. You know exactly what it says, because you wrote it.
The web agents read: raw HTML, HTTP headers, metadata, comments, llms.txt, structured data, feeds. Nobody reviews it. Anyone can write into it. And increasingly, it decides what AI says about you.
the record so far
The poisoning is not hypothetical. It is already priced.
Independent research teams, platform security units, and national broadcasters have all documented the same pattern: content is being engineered for machine readers, and machine readers are repeating it.
Dates are publication dates. Every entry cites a public source. Full ledger with links on /evidence.
method
How w3p works
One pass through the same archive: what agents received, what was engineered into it, what the evidence proves, how it comes down.
See
Render the web the way agents receive it: dual-view reads of pages, headers, feeds, and agent-only planes.
Find
Detect engineered content: hidden instructions, planted narratives, cloaking, citation farming, poisoned metadata.
Prove
Build the Evidence Pack: source, timestamp, crawler view, propagation trace. A finding you can act on, not an alert you can ignore.
Neutralize
Takedown playbooks, platform reports, remediation guidance, and a continuous watch so it does not come back.
the system
One archive. Five surfaces.
Every surface writes to the same evidence archive, the asset that compounds. Ships in phases, behind evidence gates.
A deep read of the sources teaching AI about your brand: your domains, your category’s citation graph, the pages your buyers’ agents actually consult. One scan, one verdict, one Evidence Pack.
status · in developmentContinuous watch. The archive learns your baseline, so a new hidden instruction or a fresh planted page is a diff, not a discovery problem.
status · plannedThe deliverable, not a dashboard: timestamps, dual-view captures, attribution trail, propagation analysis, and a takedown path your counsel or agency can execute.
status · first deliverableThe feed: source-integrity signals for GEO platforms, digital-risk vendors, insurers, and legal teams. Distribution through embedding, not dashboards.
status · researchWhite-label source integrity for agencies. Your clients, your report cover, our archive and verdicts underneath. If you run a GEO or crisis-PR practice, this is the service layer designed for you.
status · design partners
Free utility. Does your page look different to a human, a search crawler, and an AI crawler? The first honest answer most sites have never had.
Free utility. Which sources are feeding today’s AI answers about you? The citation graph behind the answer, rendered as evidence.
Everyone watches what AI says. We watch what AI reads.
Forty-plus tools monitor answers. Almost none monitor sources. Answer monitoring is symptom-checking. By the time the answer is wrong, the source worked. Source integrity is the discipline of watching the second web: finding what was engineered before it becomes an answer, and proving it well enough to remove.
What compounds
the archive // cannot be backfilled
Every sweep, every dual-view diff, every citation shift, timestamped and kept. A competitor can copy our detector in a quarter. They cannot copy three years of what the web taught.
the verdict // a referenceable fact
We sell verdicts, not alerts. Confidence-scored findings with provenance, calibrated against a labeled corpus, so “a w3p finding” means something to counsel, insurers, and platforms.
the embeds // distribution as product
Intel feeds into the platforms that already own the customer: GEO suites, DRP vendors, underwriters. We are the layer they integrate, not the dashboard they compete with.
One portfolio, three disciplines
geolify
Prepares what you want machines to read: audit, simulate, verify your presence inside AI answers.
outflow
Teaches machines to speak for you. The outbound side of the agentic web.
w3p
Guards what machines are made to repeat: source integrity for the second web.
A Stelixx venture. See the first light, build what endures.
Where we are
exists today
- A research corpus covering attack taxonomy, live incidents, detection design, and buyer demand
- A published evidence methodology: what we count, what we ignore, how we score confidence
- The archive schema: every finding lands as structured, timestamped evidence
- This site, the evidence ledger, and an operating thesis you can audit
still being built
- Sweep: the engine that turns a domain into an Evidence Pack
- Patrol: continuous watch on top of the archive
- Intel: the feed and partner API
- First campaign disclosures: our own findings, published
We are looking for design partners.
GEO and AEO agencies, digital-risk vendors, insurers, legal teams, and the brand operators who already suspect something is wrong with their AI answers. Bring us your hardest case.