Commercial evidence with a visible unknown bucket

AI Referral & Assisted Conversion Measurement

Identifiable AI-referred visits are followed through on-site behavior and into approved assisted revenue paths, with the evidence boundary visible at every step.

Retained referrers and campaign markers are reconciled with consented behavior and approved conversion records. Direct, last-touch, assisted and unattributed journeys keep separate definitions, and observed association stays an association. You can see which identifiable AI-referred visits continue into approved behavior and assisted revenue records. Unidentifiable journeys remain in a visible unknown bucket. Analytics and growth leads deciding whether the next investment belongs in collection, attribution, the landing page or follow-up.

Figure tracing a path from an AI answer card through a visit to a conversion flag

Some of the 500+ brands we've worked with

See all references
  • DenizBank
  • Shell
  • Sanofi
  • Arabam.com
  • Sporx
  • Amazon
  • BMW
  • Hyundai
  • PepsiCo
  • Red Bull
  • Decathlon
  • MediaMarkt
  • Bayer
  • EY
  • KPMG
  • GE
  • 3M
  • Domino’s
  • Lexus
  • Trendyol
  • Hepsiburada
  • Yandex
  • Pegasus Airlines
  • LC Waikiki

Trace only approved joins

Agree the rules before connecting records

Taxonomy and consent boundaries are agreed before records meet. Session data connects to outcomes only after the event definitions, conversion windows and join rules have named approval.

How we hold ourselves to it

  • Identified traffic only
  • Unknown stays unknown
  • Association isn't causation
  • Windows fixed before comparison

  1. Freeze the measurement dictionary

    We draft the source rules from retained examples and put every exclusion on the record. The dictionary is not frozen until every exclusion carries a reason and a named owner on the record.

    Versioned referrer, event, window and evidence-state specification.

  2. Validate collection coverage

    We replay representative journeys across tags, consent states, events and approved joins. The known collection gaps are accepted in writing before any coverage figure is used in a comparison.

    Coverage report for tags, consent states, events and joins.

  3. Classify referral sessions

    The frozen rules then run against every retained source signal, and ambiguous cases stay visible. A sampled review of each session state has to agree with the classification before the ledger is published.

    Session ledger with direct, campaign, ambiguous and unknown states.

  4. Reconstruct approved journeys

    Eligible events connect to conversion records inside the fixed windows and the authorized identity rules. No journey enters the path table until it clears the fixed window and the authorized identity rule.

    Path table connecting entries, events and eligible conversion records.

  5. Separate direct and assisted outcomes

    Each observed layer is calculated against the set of sessions it actually applies to. Each metric is released only after the owner confirms which decision it may and may not support.

    Metric table where every number names the sessions it was counted against, with no two definitions overlapping.

  6. Publish limits and next actions

    We write down the coverage gaps, the non-results and the decision each owner can take from them. The interpretation ships only when the coverage gaps and the non-results are stated alongside it.

    Decision report for instrumentation, landing pages or no action.

A replayable commercial record

Outputs another analyst can reproduce

Another analyst can replay the source rules, exclusions, joins and formulas. Unknown records remain beside identified journeys, where their effect on the revenue total is visible.

  • Measurement dictionary

    Every chart and query cites it.

  • Referral-session ledger

    Raw examples support sampled review.

  • Assisted-journey table

    Unmatched records stay visible.

  • Decision report

    No unsupported revenue attribution.

  • Identified AI referral count

    The headline number for identified traffic. It isn't total AI influence.

Start with source evidence

When an AI-referred journey can be inspected

A retained referrer or approved campaign marker gives the investigation a defensible starting point. The team then keeps direct, last-touch and assisted paths separate before deciding whether collection, the landing page or follow-up needs attention.

A good fit when

  • AI referrer labels disagree — Retained examples lack a tested taxonomy.
  • Journeys cross sessions or systems — Approved CRM joins still leave matching limits.
  • Investment is blocked — The task measures, but doesn't optimize, the page.
  • A visit has an approved AI source marker — Landing-page topic alone cannot identify the platform.
  • Site behavior is mixed with source — The report keeps them separate.
  • Assisted paths blur into last touch — The fixed window must keep both definitions apart.

Better handled as other work when

  • Prompt visibility is the question — Answer monitoring owns model exposure.
  • Unknown AI visits need a total — This work leaves visits unknown.
  • Direct traffic is assigned to AI — No retained source supports it.

People who watch how AI cites a brand

GEO work starts with recording what AI answers actually say about a brand today, then moves to the parts you can influence. The consultants below work on the specific capability this page covers.

Tools we use

Tools behind this work

  • Google Analytics

    the consented session and conversion record every AI-referral journey gets reconciled against

  • Similarweb

    the channel-level classification a retained referrer string is checked against before reconciliation

  • Google Tag Manager

    implements the frozen measurement dictionary as live events, under the approved consent states

  • BigQuery

    holds the raw session and conversion records the approved joins are replayed against

  • Looker Studio

    renders the unknown bucket beside the identified journeys, where its effect stays visible

  • Jupyter

    runs the journey-reconciliation join as a documented, rerunnable notebook

Unknown traffic stays labeled unknown

Use the journeys that kept their source

Retained referrer samples, current analytics coverage and approved conversion definitions set the boundary. We'll trace only the journeys those records support and mark exactly where the trail ends.
Trace AI referral journeys

Questions we hear

Does this capture every conversion influenced by AI search?

It reports identifiable referrals and approved assisted paths. Some referrers are stripped. Direct traffic may have no source evidence. Unmatched journeys remain unknown.

What counts as an AI referral?

A retained referrer or campaign marker must match the versioned taxonomy because a landing page about AI does not identify the traffic source.

Does an assisted conversion prove incrementality?

No. It shows an observed prior touch inside a fixed window. It doesn't show what would have happened without that touch.

Can this run without consented analytics?

Only inside the lawfully observable scope. Missing identities stay missing, and the work does not widen tracking to improve the number.