How Plumb verifies business information.

Plumb is business information consistency software. It reads public surfaces together, reduces what it finds to sourced claims, applies deterministic rules, tries to disprove every finding, and reports the evidence that survives. This page states what that means, including what a scan did not read.

What a scan can read

Four surfaces need no customer credential and are attempted on every scan: the website, its rendered JSON-LD structured data, an available website chatbot, and public social profiles linked from the site. Two more are conditional: public Google business information through the Google Places API, when configured and matched back to the customer’s domain; and available ad copy through Meta’s Ad Library, where the API and market provide coverage.

Plumb does not treat an unavailable integration as agreement. Login walls, missing credentials, robots instructions, access refusals, incomplete discovery and market limits remain visible as coverage information.

The five-stage pipeline

1. Collect. Render reachable website pages in a browser and read the configured public surfaces without evading access controls.
2. Extract. Turn visible statements and JSON-LD into typed claims such as prices, hours, phone numbers, addresses, social identities and measurable marketing claims. Every claim retains its source URL and surface.
3. Decide. Normalize equivalent formats, compare like with like, apply the rule catalogue and assign stable finding identities. The verdict path is deterministic: the same collected input produces the same decision.
4. Verify. Attack each candidate finding with checks designed to find scanner mistakes, duplicates, unsupported comparisons and already-resolved evidence.
5. Report. Show the findings that survived, the evidence behind them, what was folded or disproved, what changed since the previous scan and what remained unread.

The decision boundary

Language models may eventually help interpret unstructured wording or phrase a suggested fix. They do not decide whether a contradiction or rule breach is real. Normalization, comparison, scoring, suppression and finding identity sit below a deterministic boundary. That makes a verdict reproducible and keeps a fluent sentence from becoming its own proof.

Evidence before volume

A finding keeps the exact sentence or value that produced it and the surface where it was read. Repeated copies of one defect are folded into one stable issue rather than counted as separate proof. A verifier can disprove a candidate entirely; that result is recorded separately from simple deduplication because tidying a list is not the same as proving it wrong.

One example shaped this rule. An early contrast pass on Plumb’s operator site produced 391 candidate failures. The scanner had read an ancestor background instead of the layer a person could see. After verification, one finding survived. A smaller number was the more accurate product.

Coverage travels with the score

A score without its denominator is not trustworthy. Every report carries how many surfaces were expected, how many were read and which remained unchecked. A scan that reads nothing is not scored as clean. A partial scan can still produce useful findings, but its unread surfaces are stated next to the result rather than hidden in an internal log.

Reference facts and change over time

Customers can declare the facts they consider authoritative. Plumb can then distinguish a contradiction from a deliberate change and point to the surface that still publishes the old value. Scheduled scans keep stable finding identities so a team can see what appeared, persisted or disappeared instead of receiving a new unconnected list each time.

What Plumb does not claim

• It does not guarantee legal, accessibility, security, advertising or regulatory compliance.
• It does not say an unread surface is clean or that an API covers data it does not expose.
• It does not change Google listings, social profiles, websites or private ad accounts on a customer’s behalf.
• It does not currently measure how often a brand appears in ChatGPT, Gemini, Claude or Perplexity answers.
• It does not guarantee that every public page can be discovered or read.

Who the method is for

Plumb is built for founders, operators, marketing teams, agencies and multi-location businesses whose public information is split across vendors and channels. The method is industry-neutral: a price, opening time, phone number, policy, claim or service promise can drift in software, retail, professional services, hospitality, property services or any other business with more than one public surface.

See the broad operating patterns Plumb is built for, or read how the same evidence boundary applies to factual AI answers.

Try the method on your own site

An account can run one public sweep a month without a card. Paid Watch, Studio and Agency plans offer a seven-day trial through Stripe; a card is required and the selected monthly plan starts automatically after seven days unless cancelled first. See the plans or create an account.

Authorship and questions

This methodology is maintained by Plumb founder Kaan Can Guven and implemented by Clap Digital, Co. The public professional record is at LinkedIn. Questions and corrections can be sent to payments@clapdigital.co.