Your Analytics Cannot See the Thing You Now Care About
Google Analytics tells you what happened after someone arrived. Search Console tells you what happened on a results page. Neither one can tell you whether ChatGPT named your company when a buyer asked it for recommendations last Tuesday.
That is an increasingly large blind spot. Buyers are asking assistants for shortlists, and the assistant’s answer either includes you or it doesn’t. There is no impression to log, no session to attribute, and no line in any standard report. The measurement stack most businesses run was built for a search page with ten blue links, and it is quietly measuring a smaller and smaller share of what actually happens.
So we built instrumentation for it, and put it inside the client’s own site rather than in another tool nobody logs into.
The Four Signals That Are Actually Measurable
Nobody can give you a clean “AI referral” number, and anyone selling you one is overstating what’s knowable. What you can do is triangulate from four signals that are each individually imperfect and collectively pretty honest.
1. Branded search volume
This is the strongest proxy available, and it’s free. When an assistant recommends a company, the user’s next move is usually to search the company’s name. So a rising branded-search line while unbranded clicks stay flat is the signature of recommendations landing somewhere you can’t see.
The important part is separating branded from unbranded rigorously, including misspellings. Brand-adjacent typos are a surprisingly large share of the branded pattern, and if you leave them out you’ll underread the trend.
2. AI Overview and citation presence, per keyword
For every keyword we track, we record whether an AI Overview is present on that results page and whether the cited source is the client, a competitor, or neither. That turns an unanswerable question into a scoreboard: on the terms that matter to this business, who is the answer coming from?
This is the metric that changes strategy fastest. A keyword where you rank third organically but own the AI Overview citation is in far better shape than a keyword where you rank first and a competitor is cited above you.
3. Referral traffic from AI hosts
Assistants that link out produce actual referrers. The volume is small — usually a rounding error next to organic. The intent is extraordinary, because the visitor arrived already holding a recommendation. Tracked as a share and a trend rather than a total, this is a leading indicator worth watching.
4. The zero-click gap
Impressions climbing while clicks stay flat used to read as a problem. Increasingly it reads as consumption — your content being used to build an answer. We wrote about how to read that pattern, including the giveaway of full-sentence prompt-shaped queries appearing in Search Console. Tracking the gap deliberately, instead of being alarmed by it annually, is the difference between adapting and panicking.
Why It Lives Inside the Site
We build this as a password-gated dashboard inside the client’s own WordPress install, pulling GA4, Search Console, keyword and SERP data, Core Web Vitals, Google Business Profile metrics, form submissions with their traffic source attached, and phone-click events into one place.
Putting it in the site rather than in a separate product is a deliberate choice, for three reasons:
- It gets opened. A report emailed as a PDF is read once. A URL the team already has a login for gets checked.
- The context is right there. Seeing that a page is losing citations is more useful three clicks from the page itself.
- It isn’t another subscription. The data sources are ones the client already owns. Renting a fifth dashboard to look at your own data is a strange thing to pay for annually.
There’s also a plain-language layer on top that reads the underlying numbers and summarizes what moved and why — because a dashboard that requires an analyst to interpret it has just moved the bottleneck rather than removing it.
What We Do With What It Says
Measurement only matters if it changes the work. The loop looks like this:
- Find the terms where a competitor owns the citation. Those are the pages to restructure first — answer-first passages, facts in tables and lists, headings phrased as the question a person would actually ask.
- Find the terms where nobody owns the citation. These are the cheapest wins available, and they’re invisible to a rank-only report.
- Check whether the entity is unambiguous. If an assistant can’t confidently tell what a business is, where it operates, and what it sells, it won’t risk recommending it. That’s a structured data and consistency problem more than a content problem.
- Watch the branded line. It’s slower than everything else and it’s the one that indicates the work landed.
The Honest Caveats
Two things worth saying plainly, because this field is full of people who won’t.
Attribution here is directional, not precise. You cannot prove a given deal came from an assistant recommendation. What you can do is watch four correlated signals move together and act on the pattern. Anyone offering you a hard AI-sourced revenue figure is producing it, not measuring it.
The surfaces change under you. How AI Overviews render, which assistants cite sources, and what shows up in Search Console have all shifted more than once in the last two years and will shift again. Instrumentation built around one vendor’s current behavior ages badly. Built around the underlying signals, it survives.
If you want to know what your business currently looks like to an AI assistant — whether you’re being cited, who’s being cited instead, and what’s making you hard to recommend — that’s the work our AI search optimization engagements start with. Tell us the terms that matter and we’ll run them.
