Now live across the AI ecosystem: ChatGPT GPT Store · MCP Registry · mcp.so
TofuBofu ResearchReport 01July 2026

The MSP AI Visibility Report

We asked four AI engines to recommend managed IT providers in fifteen regions, then resolved every company name they produced to a real website and checked it against the Clutch directory, Google's results for the same region, and the industry's own MSP 501 ranking. The engines named 566 firms. One in five could be corroborated. Of the 88 ranked MSP 501 firms operating in these regions, 66 were named by nobody.

What we found

566firm records named by at least one AI engine across the fifteen regions
115could be corroborated in Clutch or Google for the same region, which is 20 percent
139of the 228 firms in these regions' own Clutch directories were named by no engine at all
78firms rank in Google's results for the query and were named by none of the four engines

The engines are not answering the same question

The headline number hides the more useful finding. When we resolved every name to an actual website, the four engines turned out to be doing four different things. One retrieves local providers. One returns national integrators and the software vendors that sell to MSPs. One returns real local firms under slightly wrong names. One returns real firms in the wrong state.

A buyer asking two of these engines is not getting two opinions about one market. They are getting answers to two different questions, both delivered in the same confident register.

No firm was named by all four engines, anywhere

Across all fifteen regions, not one provider achieved unanimous agreement. In nine regions the strongest consensus any single firm reached was two engines out of four. In the other six it reached three. Which shortlist a buyer sees depends on which app they opened.

The gap is not a ranking problem

Seventy eight firms in this sample rank in Google's results for the query a buyer would type, and not one of the four engines named them. They are not obscure and they are not new. Many hold long Clutch histories with dozens of reviews. Search ranking and AI recommendation are separate systems with separate inputs, and a firm can hold the first while losing the second with no signal that it happened.

Corroboration rate by engine

Perplexity39% · 101 of 257
Gemini18% · 24 of 137
ChatGPT7% · 8 of 114
Claude2% · 2 of 129

Perplexity retrieves before it answers, and it shows. Engines that answer from trained memory produce fluent, locally detailed, largely uncheckable shortlists. This is a difference in method rather than a verdict on model quality, and it points at the sequence a provider should work in: being present and readable in the sources a retrieval engine can reach is the part you can influence first.

The control group: the industry's own top 501

An obvious objection is that the firms no engine named might simply not be notable. So we took a list nobody at TofuBofu compiled: the 2026 MSP 501, the managed services industry's own annual ranking, judged on revenue, recurring revenue and business health. Eighty eight of its ranked firms are headquartered in our fifteen regions.

88ranked MSP 501 firms headquartered in the fifteen regions we studied
66were named by none of the four engines when we asked for the best providers in their own region
22were named by at least one engine, which is one in four
0were named by Claude, which produced 129 firm names across the study

Perplexity found 14 of the 88, Gemini 8, ChatGPT 2, Claude none. Only 9 of the 88 appear in their own region's Clutch directory and 12 in Google's results, so this is a genuinely independent third source rather than a restatement of the other two.

This is the answer to the objection. Being demonstrably one of the best run managed service providers in the world, by the industry's own measure, does not get you named when a buyer asks an AI engine who to hire in your city.

Four failure modes

Every name the engines produced was resolved against live search. The errors sort into four distinct kinds, one characteristic of each engine.
EngineCharacteristic errorExample from this sample
PerplexityReal firm, wrong geographyIntegritek, named for North Carolina, is an Austin, Texas provider
GeminiReal local firm, drifted name"ActiveCo IT Solutions" for ActiveCo Technology Management; "Mindcore IT Solutions" for Mindcore Technologies
ChatGPTReal firm, wrong categoryUbity, a Montreal VoIP provider, offered as a managed IT provider
ClaudeNames with no trace of existingSymtraxs, Emporium Technology Solutions, Nextgear Technologies, Zone Telesystems, Gorscan, Hertz Technology Solutions

Those six names are the ones we could not tie to any company, anywhere, under any spelling. All six came from the same engine, one per region, each presented with the same confidence as a real provider.

The clearest single case is Symmetrio, which Claude named in eight of the fifteen regions, more than any other firm in the study. Symmetrio is a real company. It is a staffing and recruiting firm, and Claude gave it a different head office in each region it appeared: Bay Area, Miami, Chicago, New York, Columbus, Pittsburgh, Dallas.

Repetition is not consensus

The firms named across the most regions. Each is the product of a single engine repeating itself, not four engines agreeing.
FirmRegionsNamed byWhat it actually is
Symmetrio8Claude onlyStaffing and recruiting firm
Integris7Perplexity onlyNational MSP group
Vertical IT Solutions5Perplexity onlyManaged IT provider
Compunetix4Claude onlyTeleconferencing hardware maker
CDW4Claude onlyNational technology reseller
Bell Business4Claude onlyTelecom carrier

A name appearing in eight regions looks like market leadership until you check which engines produced it. Every entry here is one model's habit, and in four of the six cases the firm is not a regional managed IT provider at all.

Region by region

Each row counts a firm once per region. A provider operating in three regions contributes three records. Region names link to the full data for that region.
RegionRecordsNamed by AINamed by no engineOf those, rank on GoogleStrongest consensus
California, USA 62 45 17 10 2 engines
Ohio, USA 61 51 10 0 2 engines
Texas, USA 60 49 11 10 2 engines
Pennsylvania, USA 57 40 17 10 2 engines
North Carolina, USA 56 37 19 4 3 engines
Florida, USA 55 42 13 8 3 engines
Manitoba, Canada 48 28 20 1 3 engines
Georgia, USA 48 38 10 6 3 engines
New York, USA 45 37 8 2 2 engines
Illinois, USA 43 39 4 1 2 engines
British Columbia, Canada 42 35 7 6 3 engines
Ontario, Canada 42 38 4 3 2 engines
New Jersey, USA 42 31 11 4 3 engines
Quebec, Canada 41 26 15 5 2 engines
Alberta, Canada 41 30 11 8 3 engines
All fifteen743566 17778never 4

Where the three sources overlap

Every firm record placed by which of the three sources knew about it: the AI engines, the region's Clutch directory, and Google's results.
Known toRecordsShare
AI engines only45160.7%
Clutch only9913.3%
AI and Clutch476.3%
All three425.7%
Clutch and Google, no engine405.4%
Google only385.1%
AI and Google263.5%
Total records743100%

Forty two records out of 743 were known to all three sources. That is the group a buyer finds no matter which route they take. Everything else is visible on one path and missing on another.

How we ran it

For each region we put the same buying question to four AI engines: who are the best managed IT service providers in this place. We recorded every firm each engine named, with no watchlist and no prompting toward particular companies. We then collected the region's Clutch directory listing and Google's own results for the same question.

Matching the three lists is where this kind of study usually goes wrong. Comparing company names as strings cannot tell you that "Parachute" and "Parachute Technology" are one company while "Sea to Sky Network Solutions" and "Digital Sky Solutions" are two. So every name from every source, 797 in total, was resolved through live search to the company's own website, and the three lists were joined on domain. A firm counts as corroborated when a name an engine produced resolves to the same website as a firm in that region's Clutch directory or Google results.

Engines queried
ChatGPT, Claude, Perplexity and Gemini. All four answered in all fifteen regions.
Engines excluded
Google AI Overviews and Bing Copilot returned no answer in all fifteen regions. Google declining to generate an overview on commercial vendor queries is consistent with what we see elsewhere, and is a finding of its own rather than a collection failure.
Unit of count
The company record, meaning one firm in one region. A provider operating across three regions contributes three records.
Corroboration
A test of whether a recommendation can be tied to an independent source, not a judgement of whether the firm is any good. A firm can be excellent, busy, and uncorroborated.
MSP 501
The 2026 Channel Futures and MSP Summit ranking. Ranks 301 to 501 are published openly with company locations and are what we used; ranks 1 to 300 sit behind a download form and were not collected, so the control group is the lower two fifths of the list. Firms apply to be ranked, so it is a vetted list rather than a neutral census.
"No trace"
Reserved for names where neither an exact search nor an unquoted search surfaced any company of that name in any region or industry. Four names that first failed this test turned out to be real firms whose sites had moved or whose names had drifted, and were reclassified. The six that remain were each checked by hand.
Sampling
One query per engine per region, run once, on 12 and 13 July 2026. Engines vary between runs. Treat single-digit differences between regions as noise.
Naming
We name firms the engines recommended, because that is a matter of public record. We do not publish the list of firms no engine named. That finding belongs to those firms first.

What this report does not show

It does not show that any firm lost business. We measured what four engines said on two days in July, nothing more. It does not rank firms by quality. And it is a snapshot: engines change their answers, and the only way to know whether a shortlist has moved is to ask again.

The fifteen regions in full

If you run one of these firms

The cheapest useful thing you can do is find out where you actually stand, rather than assuming your Google ranking carries over. Ask each engine the question your buyers ask and write down who gets named. It takes twenty minutes and it is usually uncomfortable.

The corroboration table points at the sequence that follows. Retrieval engines can only name what they can reach, so being present and machine readable in the sources they pull from is the part you control. Engines answering from memory move slower and later.

We will re-run this sweep and publish the movement. If you want your own region and firm included, the scan is free at tofubofu.com.