The decision behind the search.
An assistant can mention you without generating a click. A visitor can hear about you through AI and arrive later by typing your address. A form submission may be unqualified. These are different events, and adding them together creates a misleading result.
A practical way to approach it.
01 / Define success before the campaign
Agree what a qualified inquiry, demo, booking, or purchase means. Identify the source of truth and who will verify lead quality.
02 / Keep the funnel stages separate
Record question-level visibility, website visits, actions, and sales independently. Use campaign tagging and referral data where available, supported by a simple customer self-report field.
03 / Review costs and limitations
Compare media spend and service costs with the outcomes you can reasonably attribute. Discuss assisted influence separately from directly observed conversions and avoid treating correlation as causation.
Platform guidance: Google Search Central and OpenAI: Ads in ChatGPT. Paid campaigns must satisfy current eligibility and policy requirements.
Questions to bring to the brief.
Use these as starting points and replace the bracketed terms with the specifics of your business.
- How many qualified inquiries can we actually verify?
- Which sources are observed, self-reported, or unknown?
- What evidence would justify expanding this test?
What the work can include.
We agree the scope around your starting point and priorities.
- Definitions for qualified conversions.
- Campaign and referral tracking recommendations.
- A funnel report with attribution limitations.
- A decision log for continued testing.
Before you get started.
Is an AI mention a lead?
No. It is a visibility observation. A lead requires a separate action and should be qualified against your business criteria.
Can you attribute every sale to AI?
No. Referral loss, delayed visits, and multiple touchpoints limit attribution. We show what the available evidence supports.
How often should we review results?
Use a cadence that fits traffic and sales-cycle length. Avoid drawing conclusions from a tiny sample or changing several variables before enough evidence is collected.