Skip to content
Klariton Docs

34. Sales Intelligence — Company Visits & AI Referral

GEO / Agentic Reach shows which AI bots read your pages. Sales Intelligence shows the other half: which real companies visit your site — and which of them arrived through an AI interface. Anonymous private visitors do not produce a company account.

You find it under Market Analysis → Sales Intelligence (the Account Intelligence board). The board is LIVE. Connecting a visit data source is a guided setup today, not self-serve — see “Where the companies come from” below.

Where the companies come from

Section titled “Where the companies come from”

Two possible sources, both company-level (no personal names, no raw IP in persistence):

  • Reach Beacon (server-side). A server-side hook on your own domain reports page views; Klariton resolves the company from the visitor context. No client script, no cookies. Setup is guided together with the Klariton team.
  • Visitor-identification webhook (optional, e.g. Snitcher). A signed webhook delivers recognized companies into the board. Setup is guided.

Use one source per domain for company visits. Two parallel sources make the same company appear twice.

Visits from your own organization are excluded on purpose: your own traffic is not a lead. The board says so in a hint, so a comparison with an external tool (which does list you) does not look like missing data.

AI referral: who came through an AI

Section titled “AI referral: who came through an AI”

For every visit the source (referrer) is evaluated. If it comes from an AI interface — ChatGPT, Perplexity, Gemini, Claude, Copilot — the visit is marked as an AI referral. That is how the board answers “which real company came BECAUSE it found us through ChatGPT” — the ROI evidence for your AI visibility work.

Crawl is not referral. When a user asks an AI about your brand, the AI may fetch your page live. That fetch is a bot read and counts as a visibility signal in Agentic Reach — it is NOT a visit and not a lead. Only the click of a human afterwards is the AI referral. The two are counted separately.

Honest limit: some AI interfaces strip the referrer. Those visits show up as direct traffic. AI referral numbers are therefore a lower bound, not an exact count.

Quality instead of noise

Section titled “Quality instead of noise”

Raw “company recognized” lists are noisy: bounces, bots, employees shopping privately. The board separates signal from noise in three ways:

  • Journey phases, classified automatically. Every visited page is assigned a phase — Discover, Problem, Budget, Compare, Decide, Support — based on the page itself. Buying-near activity (for example cart or decision pages) surfaces as the “Buying-near signals” KPI and as the account’s phase. You do not maintain a page list for this.
  • Weak reverse-IP identities are downgraded. IPs that resolve to networks, ISPs or cloud/datacenter providers (Telekom, AWS, Google Cloud, …) are not company offices, even when reverse-IP lookup returns a company-like name. They are labeled “Network/Provider” and hidden behind the “All incl. weak reverse-IP” toggle — visible for reach checks, never counted as qualified accounts.
  • Truth labels on every claim. Account facts carry their evidence level: observed, derived or externally enriched. The board shows signals and probabilities, not causality, and does not fill gaps with estimates.

A lead is a recognized company with buying-near behavior — not “company recognized” alone.

B2B vs. B2C. On a B2C shop a company visit is often an employee shopping privately — not a lead. Buying-near phases on business pages are the separator. For B2B offerings almost every solidly recognized company visit is worth a look.

Reading the board

Section titled “Reading the board”

KPI tiles at the top: Signals, Accounts, Anonymous visitors, Buying-near signals, AI referral, Cart w/o purchase. The list shows recognized companies first; anonymous journeys stay visible below them, so no fake accounts are invented when identity evidence is missing.

Per account you see: signal status (Visit / Active / Lead / Purchase), account status (New / Warm / Active / Customer / Disqualified), identity confidence and source (First-party / Event metadata / Reverse-IP), visit velocity (rising / stable / cooling / new), journey phase, last visit and — where applicable — the AI source. Filters: signal status, source, phase, persona, plus company/domain/page search.

Sort by engagement and phase, not by raw recognition. When the list is capped, the board says “N of M displayed” instead of pretending completeness.

Privacy

Section titled “Privacy”

Company-level only. No personal names, no raw IP addresses in logs or persistence. Person-level signals (for example from tagged email clicks) are a separate legal question and are not shown unfiltered. See Security & Data Protection.

Section titled “Related”