Verification infrastructure for every claim in your catalog.
Merchants submit the evidence behind each claim. CITAQ evaluates it, records what holds as a signed, tamper-evident fact, and publishes it for any system to query.
CITAQ verifies the evidence merchants submit for the claims they make. It is not a certification body, and it does not aggregate consumer reviews or reputation scores.
Fernhollow Trail Bottle 750ml
- BPA-free — Tritan copolyesterVerifiedAccredited lab report
- Vacuum-insulated — 12 h cold retentionVerifiedManufacturer test data
- Capacity — 750 mlVerifiedProduct specification
- Dishwasher safe — top rackPendingAwaiting document
- Recycled content — 40% post-consumerUnverifiedEvidence no longer holds
Sample read · the record asserts only the claims whose evidence holds
Illustrative — not a live account
- GS1 Digital Link
- W3C Verifiable Credentials
- EU Digital Product Passport
- Model Context Protocol
THE RECORD
One verified record. Everything else reads from it.
An assistant asked whether a jacket is waterproof does not read your spec sheet. It weighs text — and a heavily upvoted forum thread outweighs a manufacturer's plain sentence. The loss is not a bad review. It is a hedge, and you never see it happen.
Fernhollow Trail Bottle — 750ml
$38.00
Double-walled 18/8 stainless steel.
Holds 25 oz.
Looks fine. Nothing appears to be missing.
VERDICT: APPROVE_CONDITIONAL — 2 blocking gaps
Illustrative — not a live account
HOW IT WORKS
How a claim becomes a fact.
Five steps. Every one of them leaves a record you can inspect.
THE DASHBOARD
Find the gap. Upload once. Watch it hold.
One document can be linked to multiple product lines, multiple SKUs and multiple claim types at once — so a single upload closes the gap for every product it touches.
Catalog readiness
Citation Readiness Score — catalog scope
One coverage figure for the whole catalog, aggregated across every claim state in it. Individual products are not scored — each one resolves to Verified, Unverified, or Pending.
Illustrative — not a live account
Interactive preview — illustrative data, not a live account
Evidence gaps, ordered by what they block
The inventory lists every gap in the catalog and orders CITAQ's own output by structural impact — which gap holds back the most claims — rather than alphabetically.
A claim-state map for the whole catalog
Every claim across every product, in the state it currently resolves to, on one surface.
Risk flags that carry their own way out
Each flag links directly to the resolution path for it, so a flag is a next step rather than a warning you have to interpret.
Suppressed claims, shown rather than hidden
Claims the evidence does not currently support are visible to you, with what they are waiting on.
Bulk analysis across the catalog
Run the same evaluation over the whole catalog at once and see which documents would close the most gaps.
Catalog health over time
Catalog readiness is recorded as it moves, so you can see the trend rather than a single reading.
Continuous monitoring
A drop in catalog readiness raises an alert.
Evidence-constrained copy
Draft product copy that stays inside what the evidence actually supports. It does not replace human copywriting.
Multi-platform sync
Renewing one lab certificate updates every connected platform at once, instead of once per storefront.
Running catalogs for other people? The agency view manages many at once.
THE READ INTERFACE
A fact, its evidence, and how good that evidence is.
The payload says what a product is and what it is not. Whoever reads it applies their own threshold.
Illustrative — not a live account
A signed record, independently checkable
Any system capable of cryptographic verification can confirm this record independently — without asking CITAQ.
Evidence levels L0 through L5
Every claim response carries the level of the evidence behind it, so the reader can apply its own confidence threshold rather than inheriting ours.
Per-product state, in one word
Verified, Unverified, or Pending. A product's state is a state, not a grade, and never a number.
Attribute locking
Your verified spec cannot be overwritten by a marketplace's AI — including Amazon Rufus and Shopify Magic.
Identity anchor and drift detection
The record is the fixed reference point, and CITAQ detects when a live listing diverges from it.
Consensus-poisoning defense
Repetition across sources is not evidence. Volume of agreement cannot promote an unsupported claim into the record.
A GS1 Digital Link that resolves to the record
The identifier on the product resolves to CITAQ's endpoint — not to the storefront that sells it.
Webhooks for the events that matter
Governance, drift monitoring, evidence and verification, compliance and audit — each a subscribable category.
Named tools on the endpoint
verify_product_claim: checks one claim against the evidence behind it.get_evidence: returns the artifacts a claim rests on.validate_certificate: checks a submitted certificate.get_verified_claims: returns the claims that currently hold.get_canonical_product: returns the record itself.
The same record is available over a documented API for systems that would rather call it directly.
THE RUBRIC
The scoring is published. Check our work.
Every dimension, its weight, and what moves it — written down before you buy anything.
How a page is actually retrieved
- static fetch
- browser render
- stealth render
- paid tier A
- paid tier B
- Attribute Completeness82
- Entity Definition78
- Ambiguity71
- Clarity76
- Consistency84
- Intent Clarity67
- Archetype Match80
One product, resolved end to end
Findings are ordered by impact rather than alphabetically — that is CITAQ ordering its own output, not ordering products against each other. Some fields need no evidence at all: fixing them requires data standardization, not a document.
The same method, applied to any page
A field-level comparison of data completeness between product pages — present, missing, or ambiguous. It compares the completeness of data, never the quality of products, it produces no ordered list, and it is visible only to the merchant who ran it.
Read the full scoring rubric — every dimension and its weight.
PUBLIC SCAN
Point it at a product page. See what a machine sees.
A structured read of how machine-readable a product page is — the attributes that are missing or ambiguous, broken out by area, and the specific fix for each, with examples.
It fetches the page the way an automated reader does — GPTBot, PerplexityBot, ClaudeBot, Google-Extended, CCBot — and reports what resolves and what does not.
Every finding names the attribute, why it is ambiguous, and what to write instead.
Up to 20 products per access code.
Access is by code. Join the list to get one — there is no self-serve code path.
STANDARDS
Built on the formats regulators and retailers already use.
Digital Product Passport output, GS1 Digital Link identifiers, W3C Verifiable Credentials, evidence graded to accredited-laboratory level.
Digital Product Passport output
EU markets require a Digital Product Passport. CITAQ emits one from the record you already maintain.
GS1 Digital Link identifiers
The identifier scheme retailers and regulators already read, resolving to the verified record.
W3C Verifiable Credentials
Evidence is expressed in the open credential format, so it travels without a CITAQ-shaped adapter.
Active revocation
When evidence stops holding, the record stops asserting it — automatically and continuously, not on a review cycle.
Jurisdictional claim fencing
A claim that is legal to make in one market is not asserted in another.
Per-market readiness
The record tracks which markets a product's evidence is currently sufficient for.
Evidence is graded by what stands behind it, up to and including a report from an accredited laboratory. CITAQ evaluates the evidence a merchant submits and records what it supports. It is not a certification body, and a Verified state is not regulatory approval, conformity assessment, or a guarantee of how a product performs.
THE WIDER PLATFORM
The record is the foundation. This is what builds on it.
A store, a settlement layer, auctions, disputes, agent identity — all reading the same verified record, and all architecturally forbidden from changing it.
A record can be read by any client, and nothing in the request says which agent asked or what it was allowed to do.
Agents present a signed credential, so every read against the record is attributable to an identified agent on your own audit trail.
Illustrative — not a live account
A marketplace that cannot lie to you, because the layer that sells is architecturally forbidden from touching the layer that verifies.
Agent identity and credentialing
An agent acting on a buyer's behalf carries a credential of its own, so the record knows who is asking and on whose authority.
Protocol compatibility
The record is served over the protocols agent systems already speak, rather than requiring them to adopt a CITAQ-specific one.
A selling layer that cannot touch the verifying layer
Every commerce system is constitutionally forbidden from mutating claim content or altering claim verdicts. The store is a derivative of the verification system, not the other way around.
Transactions, auctions, standing orders, bundles
Each is defined against the same verified record, and none of them may change what that record says.
Disputes resolved against the record
A dispute is settled against the verified claim record rather than against whichever party argues better.
Forensic audit streaming
The chain of what was asserted, on what evidence, at what time, is streamable for audit.
Deterministic product comparison
Products are compared on the verifiable basis of their claims — a comparison of what is evidenced, never a recommendation and never a ranking.
A settlement layer for machine-to-machine payment
A unit of account for agent-mediated settlement. It is not a blockchain token, not a stablecoin, and not a cryptocurrency.
A registry for digital goods
Non-physical products carry the same claim-and-evidence structure as physical ones.
Seller reputation, at seller scope
Reputation is held against the seller, never expressed as a score on any individual product.
Agent behavioural signals
The behaviour scored is the agent's, not the product's.
Inventory propagation
An inventory change propagates to the surfaces reading the record, without a manual re-publish per storefront.
Start with the claims you already make.
You do not change the product. You record the evidence behind what you already say about it.