Measure AI search visibility in 4 separate layers, from what AI answers say about you to the revenue that follows. An AI visibility checker covers the first layer. Analytics, the sales process and a website audit cover the other 3. Keeping them apart stops a mention in an AI answer being reported as revenue it can't prove. Here's what Dataflos tracks in each layer, and how to start in 30 days.
- 01Answer Visibility
- 02Acquisition
- 03Commercial Evidence
- 04Agent Readiness
Why One Number Can't Measure AI Search Visibility
AI search visibility spans events in different places, so no single number can carry it. A brand mention and a citation happen inside the AI answer. A visit and an enquiry happen on channels the business owns.
Ranking and last click reports flatten those events into one funnel they can't prove. The 4 layers keep them apart. Together, they show where a business appears and what people do next. They also show which commercial outcomes have defensible evidence and whether the website supports automated journeys.
Layer 1: Answer Visibility
Answer visibility records whether a brand appears for a fixed panel of real buyer questions, which competitors share the answers and which pages they cite. An AI visibility checker measures it on a sample, by hand or with a tool.
What a checker can prove, how to judge one and the scorecard to report are in our guide:
Layer 2: Acquisition
Acquisition measures movement from an AI or generated search experience into channels the business can observe directly. Start with the signals you can observe, and keep possible exposure in a separate view:
| Signal | Treat As |
|---|---|
| Identified ChatGPT referrals | Observed acquisition |
| Other AI platform referrals, where the browser preserves the referrer | Observed acquisition |
| Click data from each platform, where available | Observed acquisition |
| Growth in branded search | Possible exposure: separate view |
| Direct visits to the named service or product | Possible exposure: separate view |
| Visits to pages frequently cited in generated answers | Possible exposure: separate view |
A buyer may meet a brand in an AI answer, return later through branded search and enquire on a direct visit. Last click models credit the final touchpoint, so use them to see what closed the conversion, not as proof of the whole journey.
Adobe's AI Traffic Trends Report for August 2026 shows why the observed view matters in US retail: visitors referred by AI converted at a 60% higher rate than other traffic in July 2026. Treat it as directional for a UK business.
Create the strongest available evidence trail:
Keep The Referrer
Retain the original referrer and campaign parameters.
Store Both Sources
Store the first and the most recent known source.
Ask The Buyer
Ask a short "how did you hear about us?" question.
Compare Over Time
Compare branded demand and enquiry language with changes in answer visibility.
Layer 3: Commercial Evidence
Commercial measurement answers whether discovery activity is reaching the right market and helping create revenue.
At minimum, describe each enquiry by service interest, sector and company size or operating context. Then follow it through the sales process and record the time to decision:
- 01Enquiry
- 02Qualified
- 03Opportunity
- 04Pipeline Value
- 05Outcome
More AI referrals aren't automatically better. Visits from 10 people evaluating a relevant service can be worth more than 1,000 broad informational visits.
For a lean business, the monthly commercial view can stay simple:
| Measure | Monthly Question |
|---|---|
| Qualified enquiries | Did direct or declared AI discovery reach the right buyers? |
| Opportunity rate | Did those enquiries become real sales conversations? |
| Evidenced pipeline | What value was associated with direct or declared AI touchpoints? |
| Won revenue | Which closed work retained direct or declared discovery evidence? |
| Sales cycle movement | Did useful content help buyers progress faster? |
Document the rule behind every label, using the 3 evidence levels above. Keep engagement with cited pages, branded demand and other correlated changes in a separate view.
Layer 4: Agent Readiness
Agent readiness asks whether an automated system can understand the business and complete an intended action reliably. Search is moving from answers towards actions: Google said in May 2026 that Search would call some businesses on a user's behalf and link through to finish bookings, starting in the US that summer. Treat announced capability as a direction of travel, not universal UK availability.
An agent readiness audit should check:
OpenAI recommends allowing OAI-SearchBot for a site to appear in ChatGPT search results. Google's guidance on agent friendly websites says agents can read a site through screenshots, the raw HTML and the accessibility tree, so semantic HTML and labelled controls serve agents and people alike.
Dataflos applies the same principle behind the website. For an agency client, we shipped a read only Model Context Protocol (MCP) server over one marketing performance table: it runs read only queries, each checked by a warehouse dry run against a 1 GB limit. That's evidence of agent ready delivery, not yet of adoption or time saved.
The Minimum Data Model
The first version doesn't need a complicated warehouse. It needs consistent definitions, agreed before anyone draws a chart. Retain these fields:
| Data Set | Group | Fields |
|---|---|---|
| Prompt observations | Context | Observation date and time; platform and model or experience where known; location or account context where relevant |
| Prompt observations | Prompt | Prompt ID and prompt category; run version |
| Prompt observations | Result | Brand mentioned (yes or no); recommendation status; cited URL and source domain; description accuracy; response notes or stored evidence |
| Owned channel activity | Visitor | Session and visitor identifier within the consent model |
| Owned channel activity | Journey | First and latest source; landing page; content or tool interactions |
| Owned channel activity | Commercial | Enquiry ID; company or account; opportunity ID; pipeline and revenue outcome |
Joining data across platforms will rarely be complete. Document what's directly observed, self-reported and inferred.
How To Start In 30 Days
Start with one baseline for each layer, built in 4 steps over a month:
Define The Decision Space
Choose 30 to 50 high value questions. Group them by problem, buyer stage and service. Set the baseline across the main platforms, by hand or with an AI visibility checker. Record citations, competitors and accuracy.
Fix Acquisition Tracking
Verify AI referrers and campaign parameters. Preserve original source data. Add a short question on how buyers found you to relevant enquiry forms, and keep activity on cited pages separate from referrals.
Connect Commercial Evidence
Add the minimum source fields to the sales process. Define direct, declared and correlated evidence. Build a monthly view of qualified enquiries, opportunities, pipeline and outcomes.
Audit Agent Readiness
Test the pages and actions most likely to follow an AI recommendation. Check crawl access, factual consistency and accessibility. Then test forms, confirmation states and safe failures.
The Management View
The monthly conversation is short, 5 questions across the 4 layers:
That's enough to decide what to publish and improve. It also shows which sources to cultivate and where the website is losing demand.
Frequently Asked Questions
How do you measure AI search visibility?
Keep 4 layers separate: answer visibility from a fixed prompt panel, acquisition linked to AI in analytics, commercial evidence in the sales process, and the agent readiness of the website. Agree the definitions first, record a baseline for each layer, and report correlation as correlation.
Does zero-click mean zero value?
No. In Pew Research Center's study of US browsing in March 2025, people clicked a traditional result after 8% of searches with an AI summary, against 15% without one. Fewer clicks don't show a source had no influence: a buyer can read an answer, then come back later through branded search or a direct visit, with no AI referrer recorded. Measure exposure and visits in separate views.
Is my website ready for AI agents?
Test it against the layer 4 audit: important pages crawlable, facts consistent across the site and its listings, structured data matching what's visible, forms with clear labels and confirmation states, and booking or enquiry routes that fail safely.
Trust The Evidence. Keep The Limits Visible.
If your reporting stops at rankings or starts at the contact form, Dataflos can build the measurement layer between answers, citations and buyer movement.
Show Us The Gap