A SaaS marketing lead recently pulled up her AEO vendor’s monthly report and saw a clean number: 340 conversions attributed to AI answer engines. When she asked how that number was built, the explanation involved a mix of referral traffic, self-reported survey data, and a proprietary model she could not inspect. She wanted AI citation tracking that could tie a specific Perplexity citation to a specific opportunity in her CRM. The vendor’s slide could not do that.
Key Takeaways
- Vendor attribution figures bundle referral traffic, survey data, and modeled estimates into one number, which hides which engine actually drove a deal.
- A useful dashboard joins raw citation events (engine, query, page) with raw revenue events (session, opportunity, close date), not summary metrics from either side.
- Share of Citation and Share of Voice measure different things, and a dashboard that treats them as the same will overstate how much AI actually influenced revenue.
Why a Vendor’s Attribution Number Should Make You Suspicious
Most AEO vendors report a single figure: AI-driven conversions, AI-assisted revenue, or something similar. The number looks precise. It rarely is. Behind it sits a mix of GA4 referral traffic from known AI domains, a survey question buried in a signup form, and a modeled estimate for the traffic that shows up with no referrer at all.
That last category is large. Many sessions that originate from a ChatGPT or Claude answer arrive at your site looking identical to direct traffic, because the click passed through a redirect or the user copied a link instead of clicking one. A vendor’s model has to guess at that gap. Your dashboard should not have to guess, because you control both sides of the join: the citation event and the CRM record.
What the Dashboard Actually Needs to Join
Two data sets have to sit next to each other, at the event level, not the summary level. On one side: GA4 or CRM data showing a session, its landing page, and the revenue or pipeline stage it eventually touched. On the other: citation-level data showing which engine cited your brand, for which query, linking to which page, and when.
The join happens on page and time, not on a vendor’s attribution logic. If your pricing page was cited by Perplexity for a comparison query on a Tuesday, and that same page saw an unusual spike in sessions from Perplexity’s domain on Wednesday, you have a defensible link. Do that consistently and you get a per-engine, per-page table instead of a single blended figure.
growth in AI brand presence for NewDay USA, tracked weekly across five engines over six months.See the case study
That kind of weekly tracking across five engines is what makes a page-level join possible in the first place. Without a consistent cadence, you cannot tell whether a session spike followed a citation or preceded it.
Building the Join Key: Referral Strings, UTM Gaps, and Time Windows
GA4 does catch some AI traffic on its own. Sessions from chat.openai.com, chatgpt.com, and perplexity.ai show up as referral sources if you set up a custom channel group for them. Gemini traffic is messier, since it often blends with organic Google referrals. Claude and Grok sessions frequently arrive with no referrer at all.
Given that gap, the join key cannot rely on referrer strings alone. Pair the landing page with a time window: if a citation event fires for a page and that page sees a session volume increase within the following 24 to 72 hours, log it as a probable AI-influenced session. It is a heuristic, not a certainty, but it is a heuristic you built and can audit, which is more than a bundled vendor number gives you.
Tag the CRM side the same way. Add a custom field on the opportunity record for “first touch landing page” and “session source category,” populated from GA4 at lead capture. That single field is what lets you eventually run a query like: how much closed revenue in Q3 touched a page that Perplexity cited in Q2.
Where This Breaks: Attribution Windows and Multi-Touch Reality
Last-click attribution fails here more often than it does with search. A prospect asks Claude a comparison question, does not click through, and instead opens a new tab and searches your brand name directly a week later. GA4 records that second visit as direct or organic. The AI answer that actually shaped the decision leaves no trace in your revenue data at all.
This is why the dashboard needs an assisted-conversion view alongside a last-touch view. Track brand-term search volume against citation volume for the same period. If citations rise and brand search rises two weeks later, that lag is a signal worth reporting even though no single session can be attributed to it directly. Report both numbers side by side and label them clearly. Do not merge them into one figure, which is exactly the vendor practice you are trying to move away from.
Reading the Dashboard: Share of Citation vs Share of Voice vs Revenue
Once the join is running, the dashboard should show three columns per engine: Share of Citation, referral or assisted sessions, and revenue touched. Share of Citation measures how often a model names your brand as a source. It is not the same as Share of Voice, which measures how often your brand is mentioned anywhere across the web. A brand can dominate Share of Voice through PR volume and still show near-zero Share of Citation because its content is not structured in a way models can retrieve and cite.
The dashboard’s job is not to prove AI works. It is to show which engine, which query, and which page actually moved a deal forward.
Keeping these numbers separate matters for budget conversations. A rising Share of Voice with flat revenue touched tells you PR is working but retrieval is not. A rising Share of Citation with flat sessions tells you the model is citing you but the click-through page is not converting. Those are two different fixes, and a blended vendor number hides the difference between them.
What to Do Next
Start with the parts you already control. In week one, set up a custom channel group in GA4 for known AI referral domains and add the CRM fields for landing page and session source category. In weeks two through four, stand up citation tracking per engine for your ten highest-intent queries, logging engine, query, cited page, and date for each event.
By month two, join the two tables in whatever BI tool your team already uses, whether that is Looker Studio, a spreadsheet, or a proper warehouse. Build the three-column view: Share of Citation, sessions, revenue touched, per engine, per page. Review it monthly for the first quarter, not weekly, since AI-influenced revenue moves slower than search-driven revenue and a shorter window will just show noise. Once the pattern holds for two consecutive quarters, you have a defensible internal number, one you built from raw events instead of a vendor’s summary slide.
FAQs
What Is AI Citation Tracking?
AI citation tracking records when a model such as ChatGPT, Perplexity, Gemini, Claude, or Grok names a brand as a source in its answer, along with the query, the engine, and the page cited. It is distinct from mention tracking, which counts brand references anywhere on the web regardless of whether a model cited them.
How Is Share of Citation Different From Share of Voice?
Share of Citation measures how often AI models pull a brand into an answer and name it as a source. Share of Voice measures how often a brand is mentioned across the web overall. A brand can have high Share of Voice from PR and press coverage while having near-zero Share of Citation because its content is not structured for models to retrieve.
Can GA4 Track AI Referral Traffic on Its Own?
Partly. GA4 can capture referral sessions from domains like chatgpt.com and perplexity.ai if you build a custom channel group for them. It cannot capture sessions that arrive with no referrer, which is common for Claude and Grok, so GA4 alone will undercount AI-influenced traffic.
Do We Still Need a Vendor if We Build This Ourselves?
Most teams still need a vendor for the citation-side data collection, since tracking prompts and responses across five engines at scale is not something an in-house team typically builds from scratch. The dashboard described here uses that vendor data as a raw input and joins it to your own GA4 and CRM records, rather than accepting the vendor’s bundled attribution conclusion as the final answer.
