Summary
Between April and August 2026 a marketing-talent marketplace cut paid spend 55%. Form fills fell 43%, Google-attributed leads fell 35%, and AI-assistant leads fell 5%. A 26-week regression explains why: $100,000 of weekly ad spend buys about 162 form fills and a statistically zero number of AI-search leads.
The thesis I had to test
On a data review call in late August, a senior colleague put the internal view plainly. The AI channel, he said, was “unattributable to ads, but it is ads.” The idea was reasonable: paid campaigns build brand awareness, awareness shows up in how AI assistants talk about you, and so cutting ads would eventually starve the AI channel too.
If he was right, the $305,120 the channel had billed was borrowed from the ad budget and would fall with it. The investment case for organic AI visibility would collapse. So I ran his own preferred test, a regression, and I ran it harder than it had been run before.
What a regression is, in one paragraph
A regression asks a simple question of two series of numbers: when one moves, does the other move with it, and by how much? Run on weekly ad spend against weekly leads, it produces a slope (how many leads an extra dollar buys) and a p-value (how likely it is that the slope is just noise). A p-value under 0.05 is the usual bar for “this is real.”
The internal argument had been made on six monthly data points. Six points can be made to say almost anything.
I used 26 closed weeks, March 9 through August 31, and I ran the same test on the Google channel first, to prove the method worked. It did: Google leads tracked spend at a correlation of +0.744.
What the data said
Over April to August, while paid spend fell 55%:
- All form fills: 1,240 to 706, a 43% drop.
- Leads naming Google: 186 to 120, a 35% drop.
- Leads naming an AI assistant: 100 to 95, a 5% drop.
- AI-search leads per $100,000 of ad spend: 17.7 to 37.7, up 113%.
The regression settled it. An extra $100,000 of weekly ad spend buys about 162 form fills into the funnel, with a p-value below 0.0001. Solid.
The same $100,000 buys 2.7 AI-search leads, with a p-value of 0.52 and an adjusted R-squared that is negative. In plain terms: the effect is indistinguishable from zero, and the model explains nothing.
The thesis was confirmed for every channel it named and refuted for the one under review.
What the data does and doesn’t say
The obvious headline would be that the AI channel is growing while everything else shrinks. The data doesn’t support it.
When I strip out the ad-spend effect and look at the trend in AI leads over time, I get +0.062 leads per week with a p-value of 0.68 and a confidence interval that straddles zero. That’s a null result.
The metrics that look like growth, leads per ad dollar and share of all form fills, have a flat numerator and a shrinking denominator. The channel isn’t growing. It’s holding still while the floor drops out from under everything around it.
So the finding is: AI-search leads are independent of ad spend. Not driven by it, not growing despite it. Independent, which is a less exciting word and a far more valuable property.
What independence meant for the budget
“Growing” would have invited the question of how fast and for how long. “Independent of ad spend” answered the question the finance team was asking: what happens to this revenue when the budget is cut. The answer, demonstrated across the largest spend cut in the company’s recent history, was that it stays.
Over that window, roughly $312,000 a month of paid spend went away. The AI channel kept billing.
The $305,120 it produced from March through August was earned while the company was spending less every month. That’s a floor a finance team can plan on, and floors are what budgets are built from.
Keeping the test alive
A one-time regression is a slide. The weekly panel in my Optimization Loop keeps running, so every closed week adds a point to the series and the test re-runs. If the channel ever starts tracking spend, it will show up within a month.
