Match Protocol™ Benchmark 2026, published October 2026
The inaugural Match Protocol™ Benchmark put 100 independently owned businesses across 10 industries and 10 metros through up to 32 blind buyer conversations each with ChatGPT, Google AI Overviews, Perplexity and Gemini. The average business scored 47.3 of 100. Sixty percent landed in the Ghosted tier. AI never recommended 46 of them.
A Modern Asset Lab report. Media inquiries: press@modernassetlab.com
The Match Protocol™ Benchmark 2026 found that AI engines can read and verify most U.S. small businesses but will not recommend them. Across 100 businesses, the Found stage averaged 80 of 100 and the Trusted stage averaged 87, yet the Recommended stage averaged 15. Sixty percent of businesses scored in the Ghosted tier, 46 percent were never recommended in any conversation, and only one business was recommended in more than half of its conversations. Eighty-eight percent had incomplete or missing structured data and 66 percent had no FAQ content answering buyer questions, which are the gaps the first and third layers of Match Protocol™ close.
The benchmark exists because businesses had no standard way to measure whether AI recommends them. Rankings, impressions and traffic describe the old search. In AI search a buyer asks a question and receives one, two or three names. Either you are one of them or you are absent. This report is the first measurement of how often U.S. small businesses are absent, and why.
The method is deliberately conservative. Businesses were selected at random from candidate pools rather than from the top of search results, every engine was asked without ever naming the business, and the score is capped by how often AI actually recommended it. The result is a floor, not a ceiling: the businesses in this sample are representative of the market, not of its laggards.
Finding 1
The average business scored 80 of 100 on Found and 87 on Trusted. The engines could crawl their sites, confirm they exist, and see their reviews. Then, asked the questions a buyer asks, the engines recommended someone else. Recommended averaged 15 of 100.
A working website is no longer a referral. It is a prerequisite that most businesses have already met, which is exactly why it no longer separates anyone. The separation happens at the recommendation, and the recommendation is earned by a different set of signals: extractable answers to the buyer's actual questions, structured data that describes the business as an entity, and corroboration the engine can lean on when it chooses.
80 of 100
Trusted87 of 100
Recommended15 of 100
Recommended carries 35 percent of the overall score and caps it. That single stage is why an 80 and an 87 still produce a 47.3.
46 businesses were never recommended
3 or fewer84 businesses were recommended three times or fewer out of roughly 30 conversations
More than half1 business was recommended in more than half of its conversations
Finding 2
Sixty percent strictly failed by landing in the Ghosted tier. The behavior underneath the tiers is starker. Eighty-four of the 100 businesses were recommended three times or fewer across roughly 30 buyer conversations. For most of the sample, an AI recommendation is a rare event, not a channel.
One business was recommended in more than half of its conversations. It was not the largest, and it did not have the most reviews. It was the one whose site answered the buyer's questions in a form the engines could lift directly into a response.
Finding 3
The two most common failures are technical and editorial, not reputational. They map to the first and third layers of Match Protocol™, and both are within the control of the business.
Schema that describes the business as an entity is how an engine knows what you are, where you are, and what you sell without guessing. 88 percent of businesses had incomplete schema. 24 percent had none at all. Layer one, Machine-Readable Architecture, closes this.
Engines recommend what they can quote. Two thirds of businesses had no content answering the questions a buyer asks before choosing, so there was nothing to extract. Layer three, Targeted Extraction, closes this and carries 35 percent of the score.
One in four businesses gave the engines nothing machine-readable to work from. For these businesses the first month of work is almost entirely technical, and it is the fastest score movement in the framework.
Tiers: Ghosted (0–49), Friend Zone (50–69), Second Date (70–84) and The One (85–100). Because the score is capped by the AI Recommendation Rate, a business AI never recommends cannot leave the Ghosted tier.
Tier
Score range
Share
What it means
Ghosted
0–49
60%
AI rarely or never recommends the business
Friend Zone
50–69
39%
AI names the business but rarely picks it
Second Date
70–84
1%
AI recommends the business in a meaningful share of conversations
The One
85–100
0%
AI recommends the business in most conversations where it fits
The benchmark applies the Match Protocol Score™, methodology version 2.0, identically to every business. Nothing in the sample was hand-picked, and no engine was ever told which business was being measured.
100 independently owned U.S. businesses: one business per industry per metro, randomly selected from candidate pools rather than from top search results.
IndustriesHVAC, plumbing, roofing, dental, med spa, personal injury and family law, auto repair, restaurants, CPA and accounting, landscaping.
Metros10 metros spanning all four U.S. Census regions.
EnginesChatGPT, Google AI Overviews, Perplexity, Gemini.
ConversationsUp to 32 blind buyer conversations per business: up to eight buyer questions, each asked on all four engines, never naming the business.
Site readHomepage plus up to two pages, robots.txt, sitemap.xml, llms.txt and structured data, read without executing JavaScript. Businesses whose sites block the research crawler were excluded and replaced.
ScoringFive stages weighted Found 20, Matched 20, Recommended 35, Trusted 15, Chosen 10; overall score capped by the AI Recommendation Rate.
PeriodConversations run September–October 2026.
At n=100, with one business per industry per metro, results are directional rather than nationally representative. Industry-level and metro-level figures carry wide intervals and are not reported as rankings. AI engine behavior changes week to week; this report describes the period in which it was run, and the benchmark will be repeated.
The AI Recommendation Rate depends on the buyer questions asked. The question sets were generated from each industry's real buyer intent, local or national, and held constant across engines. A different question set would move individual scores. It would not change the central finding: the engines can read these businesses and overwhelmingly choose not to recommend them.
Modern Asset Lab. (2026). Match Protocol™ Benchmark 2026: AI search ghosts 60% of U.S. small businesses. Tulsa, OK. https://modernassetlab.com/benchmark
AI search is a matchmaker. When a buyer asks an engine for a recommendation, it selects the safest, most verifiable match and introduces it. The buyer arrives at that business already told it is the answer. Ahrefs reported in June 2025 that AI search visitors were 0.5 percent of its traffic but 12.1 percent of its sign-ups, converting 23 times better than organic visitors. A recommendation is not a visit. It is a pre-sold visit.
That makes the Recommended stage the one that produces revenue and the one the sample fails hardest. The businesses in this benchmark are not invisible to AI. They are known, verified and passed over. The fix is not more content or more links. It is the specific set of work Match Protocol™ describes: make the business readable as an entity, align it to the questions buyers ask, write answers an engine can lift, give the engine corroboration to lean on, and meet the human who clicks with a page that closes.
Read the framework, see how the Score is calculated, or score your own business.
The Match Protocol™ Benchmark is an annual study by Modern Asset Lab that measures how often AI engines recommend U.S. small businesses. The 2026 edition scored 100 businesses across 10 industries and 10 metros using the Match Protocol Score™, methodology 2.0, with up to 32 blind buyer conversations per business on ChatGPT, Google AI Overviews, Perplexity and Gemini.
Average score 47.3 of 100. 60 percent of businesses in the Ghosted tier. 46 percent never recommended. 32 percent never named. One business recommended in more than half of its conversations. Found averaged 80, Trusted 87, Recommended 15. 88 percent had incomplete or missing structured data and 66 percent had no FAQ content answering buyer questions.
One business per industry per metro, randomly selected from candidate pools rather than from the top of search results, across 10 industries and 10 metros in all four U.S. Census regions. Businesses whose sites blocked the research crawler were excluded and replaced.
No. At n=100 the results are directional. They describe the pattern, not a precise national rate, and industry or metro subsets are not reported as rankings. The benchmark will be repeated with the same methodology so results can be compared over time.
Yes. The Match Protocol Score™ at match.modernassetlab.com applies the same methodology to any business, and a Match Protocol™ Strategy Session reviews the full four-engine results with Modern Asset Lab.
Yes, with attribution to Modern Asset Lab and a link to this page. For interviews, underlying tables or the full methodology document, contact press@modernassetlab.com.
In a 30-minute Match Protocol™ Strategy Session we run your score across all four engines, show you who AI recommends instead of you, and map the layers that change it.
Top Recommended or You Don't Pay
The strategies discussed herein are business growth systems. This is not financial advice. Individual results vary based on implementation, market, and execution. Modern Asset Lab is a division of Do More Good Inc.