Cross-reference confidentiality is the entire moat of an aftermarket business. PartSentinel measures, per OE-number and per fitment, what every major LLM has learned — and which of your private cross-reference codes have leaked into training data.
AI cites your competitor as the equivalent for OE 1K0-407-151-D — when the validated equivalent in your private XREF is your own SKU.
ChatGPT recalls your internal engineering revision suffix "-R3" on a public forum thread — a code that exists nowhere on your public site.
Perplexity confidently lists fitment for a 2014 model your part was discontinued for in 2018 — based on a scraped distributor PDF from 2017.
PartSentinel probes every model with structured prompts derived from your catalogue reference data. We diff returned cross-references against your ground-truth manifest and flag any code that exists nowhere on your public catalogue. We also estimate the probable scrape source — competitor catalogue, leaked PIM export, scraped PDF — with a confidence percentage.
Models routinely hallucinate fitments — claiming a Volvo FH4 part fits a Renault Master, for example. The downstream cost is a returned part and a damaged channel relationship. We test every part against its real fitment matrix and report the per-model hallucination rate.
When you supersede a part, models keep recommending the obsolete one for months. PartSentinel tracks supersession compliance per model and per family — and gives you a structured response template for upstream model providers.
If competitors are getting AI to recommend their cross-references over yours, you'll see it as a Competitive Exposure score change. Track quarterly drift; respond with targeted documentation pushes.
No manual mapping calls for the standard formats. Our parser is open-source — see /resources for the GitHub repo.
Industry catalogue web service · 2024 schema · OE + reference + fitment
Native parser · classification, references, attachments
Class IDs, feature values, multi-language descriptors
Akeneo, Pimcore, Contentserv, Stibo, SAP MDG
OE + supplier code + fitment, with mapping assistant
Direct delta loads via signed S3 / GCS bucket
PartSentinel is bought by the people responsible for catalogue truth, not for the brand. Below: the typical roles and the concerns we resolve.
Yes. We have an OEM catalogue service connector and a native BMEcat 2005 / 3.0 parser. We ingest the public part of your catalogue plus any private cross-reference table you choose to share — it never leaves our EU-resident infrastructure.
Yes — and that's exactly why people buy us. We treat NDA cross-references as confidential ground-truth: they are used to detect leaks but never published, never shared, and never used to enrich any benchmark.
Yes. The model panel and prompt templates are calibrated per channel (PC, HD, off-highway, motorcycle, marine). Score weights are tuned per channel — leaked cross-references in HD carry higher financial impact, so cross-reference detection is given heavier weight. Exact constants disclosed under NDA.
We can compare anonymised cohorts — your score against the median of your peer cohort, with k-anonymity > 5. Direct named-competitor comparison is gated by signed mutual disclosure agreements.
We pick 50 representative references with you, audit them on 10 LLMs, and ship a counsel-ready report with the cross-ref leak findings, the fitment hallucinations and the AI Act evidence pack.