AI Overviews Are Eating Your Website Traffic. Here's What to Measure Instead.

A founder reached out a few weeks ago, genuinely rattled. Her analytics showed website sessions well down year over year, and she had not changed a thing. Same retainer, same budget, same content. By every number she knew how to read, her marketing was quietly falling apart.

So I asked one question. Were the leads drying up too?

They were not. Demo requests were up. Her sales calendar was fuller than it had been in a year. Traffic was down and the business was growing at the same time, and nobody had told her those two things are allowed to be true at once.

They are, in a lot of companies right now. The cause is AI Overviews, and the real problem is that the number she trusted quietly stopped measuring what she thought it measured.

What Actually Changed

For fifteen years, website sessions were a fair proxy for interest. Someone searched, clicked your site, and that click logged as a session. More demand, more sessions. The proxy held, so we built every dashboard on it.

AI answers broke the proxy. Now someone searches a question and Google answers it right at the top, pulling from sites like yours without sending the click. The same thing plays out inside ChatGPT, Perplexity, and Gemini, where your site can be named as a source with no visit to show for it. A large and growing share of the searches that used to feed your funnel now end with no click to anyone. That is the zero-click reality, and it is not a rounding error.

A company good at answering its buyers' questions can watch its sessions fall precisely because it is good at it. You earned the trust and lost the click. Raw sessions cannot tell "fewer people want us" from "the AI is answering for us," and those two demand opposite responses. One says fix your marketing. The other says your marketing is working and your dashboard is lying.

Where Your Dashboard Goes Blind

GA4, the tool most companies still run on, was built for a web where a visit meant a click and a click carried a referrer. That world is coming apart, and it leaves three blind spots. GA4 cannot see an answer it was never sent, so the most common way people now meet your brand is invisible to it. It files the visitor who found you in ChatGPT, then typed your name into the address bar, as "direct." And its attribution leans on the last click, so the AI answer that started the journey gets no credit while the branded search at the end takes it all. You thank the wrong channel and cut the one that worked.

The Metrics That Still Tell the Truth

I spent fifteen years in software and web engineering before I ever ran a marketing team. For a long stretch I ran analytics across 27 websites pulling more than five million dollars, and before that I did federal data work where being wrong about a number had consequences. The discipline that carries over is blunt. When a metric stops tracking the outcome you actually want, you stop leading with it. Here is what I have operators watch now instead of raw sessions.

Revenue and qualified leads, at the source. This is the only number that was ever real. Deals, booked meetings, demo requests, qualified inbound. If you are not capturing form fills and tracking inbound calls with a system that logs their source, that is the first build. It turns "we feel busier" into a number you can defend to a board.

Branded versus non-branded search demand. Pull this from Google Search Console, not your analytics. Searches for your company name are demand AI Overviews cannot intercept, because there is no generic question to answer in your place. Branded impressions rising while total sessions fall is a strong business with a shrinking informational tail. If branded demand is falling too, you have a real problem, and now you know it is real.

Your share of AI answers. This one is new, and most companies are not looking. The question is no longer only "where do I rank," it is "when the AI answers my buyer's question, does it name me." Run the prompts your buyers would ask through ChatGPT, Perplexity, Gemini, and Google's overview, and record whether you appear and who gets cited instead. Citation share is becoming what page-one ranking was a decade ago.

AI-assisted and deep-direct traffic. Two GA4 segments surface this: one for referrals from chatgpt.com, perplexity.ai, and gemini.google.com, and one for direct traffic landing on deep interior pages instead of your homepage. Someone who arrives straight on a product or comparison page with no referrer often came from an AI answer that named it. Together they read the demand GA4 buries under "direct."

Assisted conversions, not just the last touch. Stop letting the last click take all the credit. Look at assisted paths, the channels that touched a buyer earlier even though something else closed the deal, because that is where AI answers and organic content quietly live. Do not gut the top of the funnel just because it does not get the closing credit.

The quality of the traffic you do get. When early-stage tire-kickers stop arriving because the AI handled their first questions, the visitors who remain are further along and readier to act. If sessions are down sharply and conversions hold flat, you are converting more per visit, which means the traffic you lost was low value. Measure the quality, not just the count.

How To Set This Up Without a Big Project

None of this needs a new platform, just the tools you already pay for pointed at the right things. In GA4, build three saved segments so you stop staring at the blended number: high-intent interior pages, AI-referral and deep-direct traffic, and converters. In Search Console, save a filtered view for branded queries. Add call tracking if you do not have it, because a company that cannot tell you where its calls came from is flying with half its instruments taped over. Then reorder your monthly review to read revenue and qualified leads first and raw sessions last.

That reordering is most of the fix. The data was never really the problem. The hierarchy of attention was.

What Good Looks Like Now

A strong company in 2026 can show falling total sessions, rising branded search, steady interior-page engagement, growing AI-referral traffic, a climbing conversion rate, and a fuller pipeline. Read top line down, that looks like decline. Read bottom line up, it is exactly what growth looks like when the search layer shifts underneath you.

The businesses that get hurt over the next year will not be the ones losing traffic. Every business is losing that kind of traffic. The ones that get hurt panic at the wrong number, cut the content and the spend that were quietly working, and dismantle a functioning marketing engine because the dashboard they inherited was built for a search engine that no longer exists. Measure the buyers, not the pageviews. The pageviews were always a stand-in for the buyers, and the stand-in stopped standing in.

If you want to audit your own measurement this week, do it in that order: revenue first, raw sessions last, and everything in between pointed at buyers instead of pageviews. Then check whether your revenue and your dashboard tell the same story. If they do not, trust the revenue and fix the dashboard. Doing that yourself gets you most of the way. A second set of eyes is what the Machine Score, our AI-marketing-operations audit, is for, but you do not need to hire anyone to start. Just reorder what you look at first.

William Hunt
Hunt and Machine
ryan@huntandmachine.com