Your customers are getting different answers.

Problem

One question. Four different answers.

Four teams update four surfaces at different times. Plumb reads them side by side, so the contradiction becomes visible before a customer acts on the wrong answer.

At least three of those are wrong for somebody. None of them belongs to a department. That is the whole problem.

Solution

Read every surface as one document.

Plumb reads the website, structured data, chatbot, profiles and ads as one business record. It marks the contradiction, shows the evidence, and tells you which surface to fix.

1. Websiteevery page it can reach, rendered, not just fetched prices, hours, claims, contacts, links
2. Structured dataJSON‑LD, rendered and flattened what you tell machines you are
3. Chatbotasked the questions a customer actually asks what it says when nobody is watching
4. Social profilesbios and the links in them, where the network lets us read the hours nobody updated
5. Google Business Profilehours, category, address, phone the one customers actually see
6. Adslive ad copy, through the Meta Ad Library claims you are paying to amplify

A surface we could not read is a finding, not a silence. Instagram served us a login wall, your ad token expired, your Google listing does not name your domain. Each one comes back as its own line in the report, at low severity, saying Nothing in this report describes that surface, so treat it as unchecked rather than clean. The score carries the same fact next to it: how many surfaces there were and how many were actually read. A number that does not say what it is a number about is how you end up with a hundred out of a hundred on a site nobody could open.

Logic

The model reads. The engine decides.

Models extract the language. Deterministic engines decide whether two claims conflict. The same input produces the same result, and every finding points back to the source sentence.

What we suppress

Then it refuses to pad the number.

Most scanners count the same defect once per page. Plumb groups repeated detections into one finding, keeps every instance, and shows exactly why each duplicate was folded.

1. 36detections
2. →
3. 18distinct defects
4. 18folded in

1. Heading levels skip × 10Rolled into a single site-wide finding for heading-order.
2. Form field with no label × 6Rolled into a single site-wide finding for form-field-label.
3. JavaScript errors on load × 1Rolled into a single site-wide finding for console-error.
4. Control with no accessible name × 1Rolled into a single site-wide finding for control-name.

Every folded instance is listed in your report with the finding it was folded into, because a number you cannot audit is just a number we deleted.

Folding is housekeeping, and we will not sell it as rigour. The stronger thing is a detection disproved: stood up against the evidence, found wanting and struck, with the reason published where the finding would have been. On this scan that happened zero times. All 18 were roll-ups. The report says so in those words rather than reporting one suppression figure and letting you read refutation into it.

One detection was neither confirmed nor folded: a link that never answered us at all. It is reported at low severity, and what the product says about it is what it can defend: the host did not answer our automated request rather than serving a page. We cannot tell a bot filter from a dead link on that alone, so this is reported as unmonitorable rather than broken. The previous version of that sentence ended “until a normal browser identity also fails”, and no second opinion with a normal browser identity was ever taken. It has been deleted from the engine and from this page.

05 · what it costs

What it costs to have someone own this.

Every plan reads the same surfaces and preserves the same evidence. Price changes only the scan frequency, property count, and API access. Seats and roles are included.

1. Watch

   $39/ month

   4 trial credits · 40 each paid month

   • One property, every surface
   • Weekly scan, plus a scan whenever you ask
   • The full suppression log
   • Diffs between scans
   Start 7-day trial

2. Studio

   $199/ month

   30 trial credits · 300 each paid month

   • Everything in Watch, on ten properties
   • Daily scan
   • Reports you can hand a client
   Start 7-day trial

3. Agency

   $549/ month

   120 trial credits · 1,200 each paid month

   • Everything in Studio, on forty properties
   • API keys and signed webhooks
   • The only tier where the API answers
   Start 7-day trial

Every subscription trial lasts seven days and carries the full feature, cadence, seat, and property limits of the selected tier with 10% of its monthly scan credits: 4 on Watch, 30 on Studio, or 120 on Agency. Stripe requires a card at checkout, shows $0 due today, and automatically charges the selected monthly price after seven days unless you cancel first.

Checkout, recurring billing, credit top-ups, invoices, and the customer portal are hosted by Stripe. Plumb never receives card numbers, and a return from Checkout grants nothing until Stripe's signed event arrives. Your ads and Google profile still need an integration credential we hold, not a plan you buy: both surfaces are read for every tier or for none, and where the credential is missing the scan marks that surface unread rather than clean.

The builder

Four vendors, and nobody who owns the whole surface.

Plumb exists because of a failure its author kept watching happen from the inside, in one business after another, and could not fix from where he was standing.

Kaan Can Guven

Founder and chief executive, Plumb

He also co-founded and runs Clap Digital, a digital team working across eight countries and more than ten sectors. Plumb is separate from it, and Clap Digital is where the problem was found. A business buys its website from one place, its ads from another, its listings from a third, and every one of them reports that its own piece is fine. The contradictions do not live inside any of those pieces. They live in the gaps between them, which is nobody’s job.

So a customer reads one closing time on the site, a different one on the Google profile, and a third from the chatbot. A claim the marketing team stopped being allowed to make two years ago is still sitting on a landing page nobody opens. None of it is anyone lying. It is five people who updated five things in five different years, and no one who has read them side by side since.

That is the whole of what Plumb is for. It reads every surface a business publishes on at once (site, structured data, chatbot, listings, ads) and marks the places where they contradict each other, the legal exposure, and the things that are simply broken. It does not grade any one vendor’s work. It is the thing that finally looks at all of it together.

And because he sells the kind of work this tool inspects, it is built to be suspicious of itself. Every finding is attacked before anyone sees it, and the number Plumb reports is deliberately smaller than the number it found. A scanner written by people who make the thing being scanned should be assumed to be marking its own homework, so this one is built on the assumption that it cannot be trusted to.

Before Clap Digital: Amazon, and Economics at the University of Toronto. The record is public at linkedin.com/in/kaancanguven, and anything on this page is his to answer for at payments@clapdigital.co.