It’s the monthly marketing review. Slide four shows direct traffic, up 200% since summer.
The room relaxes. Finally. The AI work is paying off. People are hearing about us, remembering us, typing us in.
Then somebody asks a question nobody budgeted time for.
How many of those visitors scrolled?
The answer is fewer than last year.
Picture a company whose direct traffic doubled while the number of actual people reading it went down. That company has a chart that got promoted. And a problem that got hidden.
Hi, I’m Jeff Payne. You’re listening to The Jeff Payne Show, Episode #98: The Number That Got Promoted.
A few weeks ago on this show, I made the case that AI’s influence on your business falls into the column your analytics label “direct” or “unknown.” Somebody asks an AI tool a question, your name comes up, they don’t click — they remember.
Days later they type your name in search. No referrer. No trail. Just a visit that seems to have come out of nowhere.
I still believe the mechanism is real. But a belief is not a measurement, and this month the two came apart.
New research across a dozen company websites went looking for that hidden AI traffic. The researchers compared direct visits month over month, stripped out homepages and login pages — the places people type in anyway — and focused on the long, specific URLs nobody types from memory.
At first glance, it worked. 8 of 12 sites showed direct traffic growing. Some more than doubled.
And then they checked who showed up.
Here’s what they found, and I want you to hold two numbers side by side.
Since January, total direct traffic across those sites rose about 140%.
Engaged direct traffic — visits where a human actually stayed, scrolled, or clicked — was flat. Down 1%.
All of the growth, and none of the people.
One site logged more than 400,000 direct visits in a month. Fewer than 200 of them engaged. That’s not a traffic surge. That’s a turnstile counting wind.
The column every CEO has been told to watch for signs of AI influence is filling up with bots — automated traffic that arrives with no referrer, no engagement, and the same browser fingerprint, thousands of times over.
The column marked “unknown” didn’t get more meaningful this year. It got noisier. And noise that trends upward looks exactly like success on a slide.
Here’s the uncomfortable part. Visits that carried any identifiable trace of coming from an AI tool? Three thousandths of one percent of direct traffic. The signal may be in there. But it is buried under a pile that is growing faster than it is.
Now, the same researchers didn’t stop there — and the second half is the part I’d actually put in front of a board.
They took brands with rising AI visibility and asked this cleaner question: When a brand gets mentioned more by AI, does direct traffic or branded search go up a few weeks later?
And they found a match. One brand hit a statistical correlation coefficient of 0.72 between AI mentions and direct traffic, peaking three weeks later.
On paper, that is a marketer’s dream. A strong, clean, budget-ending number. The exact kind of chart an agency loves to frame.
Then the researchers ran the control. They scrambled the timeline—pairing AI visibility from one week with traffic from a completely different week, deliberately wrong—and gave every scrambled version the same seven chances to find a pattern.
Several of those accidental mismatches produced an equally strong, or even stronger, correlation.
Think about what that means. If you test hundreds of variables across multiple brands and time delays, math guarantees you will eventually stumble into a 0.72. Not because the cause-and-effect is real, but because you hunted for it long enough. Split the timeline in half, and that neat relationship lived in the second half for one brand, and the first half for another. It moved. Real causes don’t move.
So the research lands here, and I’ll be honest about it: across twelve sites, AI’s effect on direct traffic could not be detected in the analytics. An effect could still exist. It just can’t be isolated from the noise yet.
I told you a few weeks ago that a plan built on clean attribution would defund what’s working. That’s still true. What’s new is the other edge of the same blade: a plan built on reading the unknown column as AI will fund what isn’t.
So what do you do with a number you can’t trust in either direction?
You stop asking the dashboard, and you start asking the customer.
There is one form of attribution a bot cannot inflate and a lucky correlation cannot fake: a new customer telling you, in their own words, how they found you. Not a dropdown menu. Not a generic “search” or “referral” box. Their words. Which tool they asked, what they were trying to figure out, what made them type your name.
Companies that do this are finding something the analytics never showed them — a real share of new business that started in an AI conversation, confirmed by the only witness who was there. Not inferred from a column. Stated by the buyer.
That’s the move. The question “how did you hear about us?” is no longer a nicety on an intake form. It is the most reliable measurement instrument you own. Put it in the sales call. Put it in the onboarding email. Put the answers on the same monthly slide as the traffic numbers, and let them argue.
When direct traffic is up 200% and no customer mentions AI, you’ve learned something. When direct traffic is flat, and a third of new customers say “I asked ChatGPT” — you’ve learned something more important.
Three questions before your next marketing review.
Has your direct traffic grown this year? And if it has, did engaged direct traffic grow with it — or did the count go up while the people went down?
Has anyone shown you a chart where AI visibility and traffic rise together? Ask how many other versions of that chart they tried first.
And when was the last time someone in your company asked a new customer, in an open-ended way, how they found you — and wrote down the answer?
The number that got promoted didn’t earn it. The customer who tells you how they found you did.
I’m Jeff Payne. Thank you for listening. I will see you next time.
It’s the monthly marketing review. Slide four shows direct traffic up 200% since summer, and the room relaxes. Finally — the AI work is paying off. People are hearing about the company, remembering it, typing it in.
Then somebody asks the question nobody budgeted time for: how many of those visitors scrolled? And the answer is fewer than last year.
Picture a company whose direct traffic doubled while the number of actual people reading it went down. That company has a chart that got promoted and a problem that got hidden. This episode is about why that chart is on more CEO slides than ever this fall — and what to put next to it.
THE BELIEF THAT NEEDED TESTING
A few weeks ago on this show, I made the case that AI’s influence on a business lands in the column analytics tools label “direct” or “unknown.” The mechanism is simple: a buyer asks an AI tool a question, your name comes up, they don’t click — they remember. Days later they type you in. No referrer, no trail, just a visit that appears to come from nowhere.
I still believe the mechanism is real. But a belief is not a measurement, and this month the two came apart.
Kevin Indig at Growth Memo went looking for that hidden AI traffic across 12 company websites. He compared direct sessions month-over-month, stripped out homepages, login pages, pricing pages, and the other places people type in anyway, and focused on the long, specific URLs nobody enters from memory. At first glance it worked: 8 of 12 sites showed direct traffic growing, some by more than 100%.
Then he checked who showed up.
A belief is not a measurement. This month the two came apart.
ALL OF THE GROWTH, NONE OF THE PEOPLE
Hold two numbers side by side. Across the filtered January–September sample, total direct traffic rose 140.9%. Engaged direct traffic — sessions where a human actually stayed, scrolled, or clicked — moved -1.0%. Flat.
One property in the study recorded 409,000 direct sessions in a month. Only 198 of them counted as engaged. That isn’t a traffic surge. That’s a turnstile counting wind.
The column every CEO has been told to watch for signs of AI influence is filling with automated traffic — bots that arrive with no referrer, no engagement, and the same browser configuration thousands of times over. The unknown column didn’t get more meaningful this year. It got noisier. And noise that trends upward looks exactly like success on a slide.
The uncomfortable footnote: across the 12 sites, direct sessions carrying any identifiable AI trace averaged 0.003 percent of direct traffic. The signal may be in there. It is buried under a pile that is growing faster than it is.
The unknown column didn’t get more meaningful this year. It got noisier.
The 0.72 that set off a boulder
The research didn’t stop at the bot filter, and the second half is the part I’d put in front of a board.
Indig took four brands with tracked AI visibility and asked a cleaner question: when a brand is mentioned more by AI, does direct traffic or branded search rise in the weeks that follow? Across 632 tests — different brands, different AI metrics, different traffic measures, delays from zero to six weeks — one brand showed a 0.72 correlation between AI share of voice and direct traffic, peaking at a three-week lag. Its correlation with branded search impressions hit 0.59. A second brand showed 0.47.
That is a strong, clean, budget-ending number. The kind of chart an agency frames. Then came the control. He shuffled the dates — pairing AI visibility from one week with traffic from a different week, deliberately mismatched — and gave every wrong version the same seven delay options and let each keep its best result.
Several of the wrong alignments matched or beat the real one. For the first brand, 3 of 19 mismatched versions did. For the second, 2 of 38.
Then he split each timeline in half. The first brand’s relationship lived entirely in the second half; the second brand’s lived in the first. Neither held across both periods. It moved. Real causes don’t move.
Any chart tested hundreds of ways will produce a 0.72 somewhere — not because the effect is real, but because you looked hard enough.
This is one of the best written, most informative email newsletters I’ve ever read!
Thank you for sharing your expertise, and hope you’re doing well.
Beth Zimmerman
Founder, Executive Director
The other edge of the same blade
So the research lands across 12 sites of different sectors and sizes, AI’s effect on direct traffic could not be detected in the analytics. An effect could still exist. It can’t yet be isolated from the noise.
A few weeks ago I told you that a plan built on clean attribution would defund what’s working. That’s still true. What’s new is the other edge of the same blade: a plan built on reading the unknown column as AI will fund what isn’t.
Both mistakes come from the same instinct — asking a dashboard to settle a question the dashboard was never built to answer.
He didn’t win that by volume. He won it because every time someone recommended him, they said the right thing.
A plan built on clean attribution defunds what’s working. A plan built on reading the unknown column as AI funds what isn’t.
Stop asking the dashboard. Ask the customer.
There is one form of attribution a bot cannot inflate and a lucky correlation cannot fake: a new customer telling you, in their own words, how they found you. Not a dropdown. Not “search” or “referral.” Their words — which tool they asked, what they were trying to figure out, what made them type your name.
This is how the one company in the study’s orbit that did find AI-driven direct traffic found it. Growth leader Eric Wu described on the Siege Media Podcast how Ramp isolated direct visits to long blog URLs, filtered out email-campaign patterns, compared engagement against organic search visitors — and then confirmed it with self-reported attribution. The analytics suggested. The customers confirmed. That order matters.
The question “how did you hear about us?” is no longer a nicety on an intake form. It is the most reliable measurement instrument you own. Put it in the sales call. Put it in the onboarding email. Put the answers on the same monthly slide as the traffic numbers, and let the two argue.
When direct traffic is up 200%t and no customer mentions AI, you’ve learned something. When direct traffic is flat, and a third of new customers say “I asked ChatGPT,” you’ve learned something more important.
How did you hear about us?” is no longer a nicety on an intake form. It is the most reliable measurement instrument you own.
Three questions
The number that got promoted didn’t earn it. The customer who tells you how they found you did.
Frequently asked questions
Source: Research referenced in this episode: Kevin Indig, Growth Memo, “Does AI drive more direct traffic?” (12-site GA4 analysis and 632-test AI-visibility correlation study). Ramp’s self-reported attribution approach described by Eric Wu on the Siege Media Podcast.
COMPLETE THE FORM TO
BOOK A STRATEGY CALL
"*" indicates required fields
COMPLETE THE FORM TO
BOOK A STRATEGY CALL
"*" indicates required fields
Subscribe and Share – WE APPRECIATE YOUR SUPPORT
