Summary
Before March 2026 a marketing-talent marketplace doing roughly $11.75 million in six months had no AI-search channel and no way to see one. Six months later the channel had billed $305,120, closed 16 deals worth $230,550, and grown from 8.1% to 13.5% of inbound form fills. This is how I built and measured it, and the two attribution errors it had to survive.
Nobody could see the channel, so nobody could grow it
By early 2026, buyers were asking ChatGPT and Perplexity where to hire a fractional CMO or a paid-media specialist. Some of those buyers were landing on our site and signing.
But the form that asked new customers how they heard about us had no option for it. A buyer sent by Claude ticked “Other” or “Google” and vanished into the general pile.
A channel nobody could count was a channel nobody could invest in or defend. So the work began not with content but with a single field.
Step one: a checkbox, and a path from form to dollars
On March 10, 2026, a new option appeared on the how-did-you-hear-about-us list: “LLM (ChatGPT, Claude, Gemini, Perplexity, etc.).” That date is the channel’s birthday. There is no history before it, which means every number that follows starts from a true zero.
The field alone was worthless without a path from it to money. I built the attribution chain in the company’s data warehouse: form fill, to contact, to deal, to billed monthly revenue. Not “influenced,” not “assisted.”
Billed. A customer either signed and paid or they didn’t.
Step two: be the answer buyers are already asking for
With measurement in place, the content work had a target. I pulled the questions buyers put to assistants (“best place to hire a fractional marketer,” “how much does a freelance CMO cost”) and built pages that answered them directly, in the first paragraph, in language an assistant could quote.
The pages went through my Article Factory: drafted against a scorecard, every link verified by code, edited by a human. Structured data and a plain-language file told AI crawlers who we were and what we sold.
Then I stood up the citation panel from my Optimization Loop: five engines, ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews, asked the same rotating set of buyer questions every week, every answer stored verbatim. That panel is what later showed us cited more often than Upwork or Toptal.
Step three: report it every week to someone who will push back
A channel survives on its reporting. The weekly audit in my Optimization Loop became the operating review the CEO reads: search performance on fixed seven- and 28-day windows, a split of brand versus buyer-intent versus off-target queries pulled at 25,000 rows because the default 5,000 cuts off exactly the part that changes the conclusion, and the AI-sourced customers named one by one with their billed revenue. One week’s edition listed eleven of them.
The review also carried last week’s open questions and closed them with corrected numbers. Which brings me to the two near-misses.
The $15,336 that was earned and not counted
The company’s attribution view kept one deal per contact. Sensible for most purposes; wrong for this one. When an AI-sourced customer signed a second time, hiring an SEO marketer in March and a social media manager in April, the second deal fell out of the AI channel’s total silently.
No error, no warning. The channel was being reported at $144,124 to date.
I rebuilt the roll-up to sum every deal per AI-sourced company and re-ran it against the named customer list. The true figure was $159,460. That is $15,336 of revenue the channel had already earned and wasn’t getting credit for.
I also documented the two data joins that must never be mixed, after a verification query without a distinct clause reported one month at $163,360 against a true $41,120. The number you report is exactly as good as the join behind it.
The $12 million that wasn’t ours to claim
The second near-miss ran the other way. A slide in a late-August review described roughly $12 million over six months, and in the meeting it was restated as revenue attributed to AI search. It wasn’t.
It was company-wide gross revenue, $11,754,420 for March through August. The AI channel’s share of that was $305,120, or 2.60%.
The $12 million framing came out of the deck the same week. A smaller number that survives scrutiny is worth more to a channel than a bigger one that doesn’t.
What the channel is worth
- $305,120 in AI-attributed billed revenue, March through August 2026. That is 2.60% of everything the company billed, from a channel that did not exist in February.
- 16 deals signed from the AI cohort since March, $230,550 in cumulative contract value.
- 582 form fills from the channel all-time. Its share of all inbound form fills rose from 8.1% to 13.5%.
- The first signed deal came 20 days after the field went live.
And when the company cut paid spend 55% through the summer and every other channel fell 35% to 43%, this one fell 5%. It doesn’t run on ads. A separate case study shows the regression that proves it.
At a glance
- The niche: Marketing-talent marketplace, roughly $11.75 million in six-month revenue.
- The problem: No visibility into AI-search demand: no attribution field, no path from lead to revenue, no way to invest in or defend the channel.
- What was built: An AI-search attribution field and a form-to-billed-revenue chain in the data warehouse; answer-shaped content against buyer questions; the five-engine citation panel and weekly audit from my Optimization Loop.
- Implementation: Field live March 10, 2026; attribution chain and panel inside the first month; weekly review from then on, with two attribution errors caught and corrected in flight.
- Results: $305,120 in AI-attributed billed revenue in six months (2.60% of company revenue), 16 deals, $230,550 in contract value, channel share of form fills from 8.1% to 13.5%.
- Built with: my Article Factory and Optimization Loop.
