Measure AI search pipeline by connecting recorded referrals and buyer-reported discovery to the opportunities in your CRM. Keep those records alongside the answers AI gives about your company, then review both. You can show where there is evidence of influence without claiming that every untracked visit or new deal came from AI.
We want a report that a marketing lead can explain deal by deal. That’s why we start with the buyer’s evidence and a consistent counting rule. A citation alone tells you very little about the commercial outcome, and an increase in Direct traffic leaves the source unresolved.
What metrics should you track to measure AI search pipeline?
Track recorded AI referrals, buyer-reported discovery, and the opportunities linked to either source. Use separate measures for share of voice, recommendation rate, accuracy, and sentiment to understand the answers buyers encounter. We treat AI search as an influence channel, so the report needs both commercial records and evidence of what changed in those answers.
| Measure | Counting rule | What it tells you |
|---|---|---|
| AI referral sessions | Sessions with an identified AI referral source; keep the source and reporting dates. | Recorded visits from that source. It doesn’t capture every visit influenced by AI. |
| Buyer-reported AI discovery | Distinct leads that name an AI assistant when asked how they first heard about you. | The buyer’s account of discovery, which you can retain and review. |
| Opportunities with AI evidence | Distinct opportunity IDs linked to a recorded referral or a buyer’s account. | The deals you can inspect when discussing AI’s contribution. |
| Open pipeline with AI evidence | Sum the current values of those distinct open opportunities, using one currency and value basis. | Potential contract value associated with the evidence. Open pipeline isn’t revenue. |
| Closed-won value with AI evidence | Report closed-won contracts in their own total, under the same inclusion rule. | Signed value associated with AI evidence. Keep recognized revenue and cash receipts in their own financial reports. |
A recommendation and a citation can occur in the same answer. Use a field for each. The first concerns which supplier AI suggests; the second shows a source it displays. Neither establishes that a buyer saw that particular answer or later became a customer.
For answer quality, retain the response, date, platform, prompt, and evaluation notes. Our guides to finding AI visibility gaps and checking inaccurate product descriptions explain how to investigate what you find.
How do you capture AI referrals and self-reported attribution?
Start with the acquisition information your analytics records and an open question on your enquiry form or sales call. Preserve the buyer’s wording, then connect the lead to its CRM opportunity. These are complementary records: the first describes an observed visit, while the second describes what the buyer remembers.
Record identifiable visits
Review source and medium for visits from AI services, keeping the landing page and visit date. Where your existing tracking can connect a submitted enquiry to that visit, preserve the source on the lead. Test the journey through any separate booking domain before relying on it; a visit count alone can’t tell you which opportunity it produced.
Keep unknown sources unknown. Google defines Direct traffic as traffic without a clear referral source. Missing referral information has several causes, so assigning the entire category to AI would invent an attribution signal.
Ask about discovery and later research
Use “How did you first hear about us?” for discovery. On a sales call, follow with “What did you use to research or compare the options?” if that detail is useful. A buyer might discover you through a colleague and later use an AI assistant to compare vendors. Keep both answers rather than replacing the original source.
Save the wording, the date it was captured, and a source category you can report on. Add a platform or prompt only when the buyer provides it. Self-reported attribution is primary research into your funnel, but recall can be incomplete. A blank response stays blank.
Keep a small record for each opportunity
Your CRM record needs the opportunity ID, company, creation date, stage, value, currency, observed referral evidence, and the buyer’s discovery and research answers. Link the supporting lead records. If several contacts belong to the same opportunity, their evidence belongs on that one deal; each contact isn’t a new pipeline amount.
How do you calculate AI search pipeline without double counting?
Use the opportunity ID to combine evidence before adding deal values. An opportunity supported by both a recorded referral and a buyer’s answer appears once in the combined total. Keep its evidence labels so the team can inspect the source of the attribution and distinguish acquisition from later influence.
The example below is fictional. All four opportunities are open at the same reporting date, and every amount is unweighted annual contract value in US dollars. It illustrates the calculation; these are not VisibleIQ client results.
| Opportunity | Annual value | Evidence linked to the deal | Include? |
|---|---|---|---|
| A | $24,000 | Enquiry linked to a recorded AI referral; discovery answer blank. | Yes |
| B | $36,000 | Buyer names an AI assistant as the discovery source; no recorded referral. | Yes |
| C | $18,000 | Recorded AI referral and buyer-reported AI discovery. | Yes, once |
| D | $12,000 | Direct visit; buyer’s discovery source unknown. | No |
Recorded-referral opportunities total $42,000: A plus C. Buyer-reported opportunities total $54,000: B plus C. Adding those totals would count C twice. The combined open pipeline with AI evidence is $78,000, calculated as $42,000 + $54,000 – $18,000, or by summing A, B, and C once each.
D remains outside that total because the source is unknown. If a buyer later supplies useful evidence, record the update and its date. When C closes, move it out of the open-pipeline figure and into the closed-won report for the appropriate period. Don’t keep it in both.
We’d label the $78,000 as “open pipeline with AI evidence.” That label preserves what the records establish. Proving how much additional pipeline AI created would require a separate causal study; this example is a way to document associations that your team can inspect.
How do you report AI search results each month?
Put the commercial totals beside a consistent record of the answers buyers encounter. State the reporting period, opportunity inclusion rule, and prompt set so changes can be interpreted. We keep this monthly report small enough that a reader can inspect the underlying deals and responses when a number needs explaining.
| Report item | Definition to record | Review question |
|---|---|---|
| Share of voice | For this report: your brand’s answer-level mentions divided by all tracked brands’ answer-level mentions. Count each brand once per response and keep the competitor set fixed. | Is your presence changing relative to those competitors? |
| Recommendation rate | Valid responses that recommend you divided by all valid responses to the fixed buying-question set. Report failed runs separately. | Are buyers more likely to see you suggested for a relevant need? |
| Accuracy | Verified-correct factual claims divided by all checkable claims reviewed. Show unresolved claims separately and keep the review rule consistent. | Are pricing, product, and positioning claims current? |
| Sentiment | Positive, neutral, negative, or mixed assessments among responses that discuss you, using a written review rule. | How are you framed when you appear? |
| Commercial evidence | Qualified enquiries, distinct opportunities, open pipeline, and closed-won value under the rules above. | Which records support the connection to sales? |
Keep model versions, prompt wording, run frequency, and included markets in the report notes. If the test changes, mark a new baseline. Counts of brands mentioned and shares of mentions use different denominators; choose a definition and carry it through rather than switching between them.
Use Google Search Console for Google visibility and clicks, and analytics for what happens on the site. Google explains why clicks and sessions don’t match exactly. Branded search growth can add context, but it can’t identify the source of an individual opportunity.
There is still open territory in measuring influence before a visit. We can improve the evidence by asking better questions and preserving it carefully. We can’t fill missing records with assumed conversions. A smaller number with traceable opportunities is more useful when deciding where to invest.
For the wider delivery approach, see our AI Search Optimization service and published SaaS case study.
Frequently asked questions
Start with the records you can maintain consistently, then improve the detail as the sales cycle develops. A report should make its evidence and gaps easy to inspect. These questions cover the practical decisions that affect whether the numbers remain comparable from one reporting period to the next.
Can a small team start without an attribution platform?
Yes. A CRM with consistent source fields and linked evidence is enough to begin. The constraint is whether visits, buyer answers, and opportunities can be connected reliably. Add software when it solves a specific reporting problem you have identified.
Should all Direct enquiries count as AI-influenced?
No. Direct doesn’t identify the source. Include an enquiry in the AI-evidence group only when a recorded referral or a buyer’s account supports that decision. Retain unknown enquiries as a separate group.
What happens when a buyer names several discovery sources?
Preserve all the sources they describe and distinguish first discovery from later research. Keep the opportunity once in any combined total. Don’t allocate arbitrary percentages of its value to each source unless you have explicitly chosen and documented such a model.
Can you turn pipeline into an AI search ROI figure?
Open pipeline can’t establish realized ROI. A financial comparison needs a defined outcome period, costs, and an agreed revenue or profit basis. Even when a won deal has AI evidence, that association alone doesn’t establish that the entire deal was incremental.
