A new benchmark says AI assistants ghost 80 of 100 small…

Modern Asset Lab ran 100 U.S. small businesses through blind buyer conversations with ChatGPT, Google AI Overviews, Perplexity and Gemini. The directional sample found most are readable, almost none are recommended.

Of 100 U.S. small businesses run through blind buyer conversations with ChatGPT, Google AI Overviews, Perplexity, and Gemini, 80 landed in the lowest tier. The average score was 50.9 out of 100. That’s the headline from a benchmark Modern Asset Lab published Tuesday, alongside its new five-layer framework, Match Protocol.

The Tulsa-based client-acquisition agency ran up to 32 conversations per business (average 27.8) across 10 industries and 11 metros. No industry cleared a 60 average. Plumbing led at 56.6, HVAC at 54.9. AI assistants never recommended 36 of the businesses and never named 31. Modern Asset Lab calls the n=100 sample “directional rather than nationally representative,” which is the honest framing, a vendor-produced study introducing a vendor-produced framework.

The structural finding survives the caveat. Businesses averaged 80 on the “Found” stage and 88 on “Trusted,” then collapsed to 18 on “Recommended,” the stage that carries 35% of the score. Being indexable isn’t being chosen.

“When your ideal customer asks ChatGPT or Google AI a question, the AI is a matchmaker scanning for the safest, most authoritative match. The benchmark confirms it: most businesses have never been introduced,” said Clint Rogers, founder of Modern Asset Lab.

The surrounding week makes the point harder to ignore. A Raptive survey of 3,083 U.S. consumers in September 2026 found trust in an AI answer rose from 42% with no source shown to 77.2% when the answer cited a trusted publisher. Citations are the currency. On October 15, MarTech’s roundup listed Somantra, Verndale, Vannor, and PostcardMania shipping AI-visibility products the same week. A category is forming in real time, which is usually what happens right after a benchmark gives it a number.

Zoom out and the stack is coherent. Meta and Sierra’s personal agent protocol defines how agents transact; Instinct’s $1B raise at a $10B valuation funds the consumer side; and the discovery layer, as this benchmark argues, is already deciding who gets named. Small B2C businesses getting locked out at the door is the symptom; the Match Protocol scorecard is one attempt to measure it.

For an owner-operator, the practical read is that a working website isn’t the finish line anymore. The content, third-party citations, and entity signals that make a business legible to an AI engine are the new top of funnel. That’s the work LemonLime prepares proactively, studying a company, its industry, and its competitors, then shipping the branded content and positioning that move a business from found to chosen.

The benchmark’s real contribution isn’t the 50.9. It’s the shape of the gap between 88 and 18.

A new benchmark says AI assistants ghost 80 of 100 small businesses

Sources