Summary
In an August 2026 run of 200 answer cells across five AI engines, the marketing-talent marketplace I run search for was cited 74 times, ahead of Upwork (53), Toptal (34), Mayple (11), Fiverr (10) and Growth Collective (1). That is 40.4% of the 183 citations among the six tracked brands and presence in 37% of all cells. Six months earlier, nobody knew whether it appeared in AI answers at all.
The competitors were giants, and the referee was invisible
When a founder asks Perplexity “where should I hire a fractional marketing director,” the answer is a shortlist. If the shortlist says Upwork and Toptal, the deal is gone before anyone knew it existed. No ad can intercept that conversation.
In early 2026 the marketplace had no idea how often it made the shortlist. There was no panel, no baseline, and, most importantly, no denominator. Someone might type a question into ChatGPT, see the brand mentioned, and call it a win.
Someone else might not see it and call it a problem. Both were opinions. Neither was a measurement.
Building a referee
I built a citation panel that behaves like an instrument rather than an anecdote. It’s now the weekly GEO run inside my Optimization Loop.
Forty buyer questions, rotated so the set doesn’t go stale, run every week across five engines: ChatGPT, Perplexity, Gemini, Claude and Google AI Overviews. That’s 200 answer cells per run. Every answer is stored word for word, so a run in August can be compared to a run in June rather than to someone’s memory of June.
The panel records two different things for every cell: whether the brand is named, and whether one of its pages is cited as a source. Those are not the same, and the gap between them turned out to be the whole strategy.
A run costs $7.34 in API fees, which is what makes weekly repetition affordable.
Named everywhere, cited nowhere
The first thing the panel showed was uncomfortable. In an audit of 60 verbatim answers, every one of them named the marketplace.
Forty-eight of 48 assistant answers, twelve of twelve Google AI Mode answers. Awareness was not the problem.
Only 18 of the 48 cited one of its pages as a source. The assistants knew who we were and were building their answers from someone else’s content.
Being named put the company on the shortlist. Being cited would earn the click and make it the reference the assistant returned to.
So the content program had a precise target: become the source.
Becoming the source
The fix was less about volume and more about shape. Pages were rewritten so the direct answer to a buyer’s question sat in the first paragraph, self-contained, in language an assistant could lift without editing. Pricing pages got real numbers.
Comparison questions got honest comparisons. Structured data and a plain-language crawler file made the brand’s identity unambiguous. Every rewrite went through my Article Factory: drafted against a scorecard, every external link verified by code before publish, edited by a human.
Each week the panel showed which questions we’d become the source for and which we hadn’t, and the next week’s content went where the gaps were.
The August run
On August 27, 2026, across 200 cells with zero errors:
- The marketplace: 74 citations, 40.4% of the 183 awarded among the six tracked brands.
- Upwork: 53, 29.0%.
- Toptal: 34, 18.6%.
- Mayple: 11, 6.0%.
- Fiverr: 10, 5.5%.
- Growth Collective: 1, 0.5%.
Presence in 74 of 200 answer cells, or 37%.
The two numbers measure different things. The 40.4% is a share of citations among the six tracked brands, and it moves if the list changes.
The 37% is presence across all 200 answers, and it doesn’t. And because the question set rotates, this is a level rather than a trend: where the brand stands this week, not how fast it’s climbing.
What a citation is worth
The panel isn’t an end in itself. The buyers those citations send became the AI-search channel that billed $305,120 in six months and signed 16 deals.
Being the source an assistant quotes is the top of that funnel. Beating Upwork in a table is satisfying; the deals are the point.
At a glance
- The niche: Marketing-talent marketplace competing with Upwork, Toptal and Fiverr.
- The problem: No measurement of AI visibility. Named in every AI answer but cited as a source in only 18 of 48.
- What was built: The citation panel inside my Optimization Loop: 40 rotating buyer questions across five engines, 200 answer cells, every answer stored verbatim, naming and citing scored separately.
- Implementation: Weekly runs at $7.34 each; pages rewritten to be the quotable source; each run’s gaps set the next week’s content.
- Results: August 27, 2026 run: 74 citations of 183 among six tracked brands (40.4%), ahead of Upwork (53) and Toptal (34); presence in 37% of all answer cells.
- Built with: my Optimization Loop (citation panel) and Article Factory.
