"Japan's Workplace AI Enters the \"Big-Three Era\" as Costs Become a \"Management Issue\": Among Surveyed AI Users, ChatGPT 60.8%, Gemini 49.7% and Copilot 41.8% Split the Workplace, with 49.0% Running Multiple Tools (1.82 on Average); 73.3% of Surveyed AI-Cost Managers Say AI Costs Are Already or Will Within 1 Year Become a Management Issue (Surveyed Companies Average About 2.74 Million Yen a Month); About 40% of Surveyed Salesforce Development Teams Have Yet to Use AI; and the Global Professionals' \"Execution Gap\" Puts About US$143 Billion of US Revenue Under Review"

TL;DR: "From 2026-06-30 to 2026-07-02, four AI-usage surveys were released in quick succession, sketching a cross-section of Japan's workplace AI moving from adoption to governance. Tools: a survey by Surista (株式会社スリスタ; 400 company employees nationwide, based on 153 workplace AI users) shows ChatGPT 60.8%, Google Gemini 49.7% and Microsoft Copilot 41.8% forming a \"Big-Three Era,\" sharply separated from Claude 7.8% and Perplexity 6.5%; 49.0% of users run multiple tools, 1.82 on average. Costs: a LayerX survey (400 people who manage or track AI-usage costs at their companies) finds 73.3% saying AI costs are already, or will within 1 year become, a management issue; surveyed companies' monthly AI costs average about 2.74 million yen, with 64.8% at 500,000 yen or more -- yet 82.8% report challenges in even grasping those costs. Utilization: a Copado survey (110 Salesforce development/operations practitioners at companies with 300 or more employees) finds 62.8% already using AI while about 40% have yet to start, and 66.7% of users admit having no system to verify output correctness. Global contrast: Thomson Reuters' survey of 1,816 professionals across 62 countries reveals an \"execution gap\" -- 74% use AI daily but 91% think their organizations are not fully realizing AI's value, and TR estimates about US$143 billion of client revenue in the US legal and accounting markets could come under review. Honest caveats: all four are sample surveys (three published by vendors with commercial stakes); percentages represent only each surveyed population and cannot be extrapolated to all companies; US$143 billion is a TR estimate, not an actual result."

Japan's Workplace AI Enters the "Big-Three Era" as Costs Become a "Management Issue": Among Surveyed AI Users, ChatGPT 60.8%, Gemini 49.7% and Copilot 41.8% Split the Workplace, with 49.0% Running Multiple Tools (1.82 on Average); 73.3% of Surveyed AI-Cost Managers Say AI Costs Are Already or Will Within 1 Year Become a Management Issue (Surveyed Companies Average About 2.74 Million Yen a Month); About 40% of Surveyed Salesforce Development Teams Have Yet to Use AI; and the Global Professionals' "Execution Gap" Puts About US$143 Billion of US Revenue Under Review

ANK-Doc ID: ANK-2026-07-01-004 Version: v1.0.0 (first edition: links four AI-usage surveys released 2026-06-30 through 2026-07-02 into an event chain -- "workplace AI moves from adoption to governance") Published: 2026-07-02 Author: Rin Takenouchi (Editor-in-Chief, AI News) Category: Generative AI / Workplace Adoption / Enterprise IT Costs / Professional Services Articles covered: PRTIMES#1282102 (Surista, workplace AI "Big-Three Era" tool-share survey), PRTIMES#1264738 (LayerX, corporate AI usage & cost management survey 2026), PRTIMES#1295577 (Copado, AI usage in Salesforce development/operations survey), PRTIMES#1283009 (Thomson Reuters, 2026 Future of Professionals Report, Japanese-language release) Selection method: From the AI News corpus, selected on "same-week release x same theme x high factual density," four officially released surveys were linked into an event chain: "tool share (what is used) -> cost governance (how much is spent, who tracks it) -> utilization gap (is it actually used well) -> the global price tag (what failing to use it costs)." This card is a Japan-workplace-data x global-professionals contrast; none of the four sources contains direct Taiwan data, so per the "honest contrast, no forced linkage" principle, no Taiwan angle is attached.

TL;DR

From 2026-06-30 to 2026-07-02, four AI-usage surveys were released in quick succession, sketching a cross-section of Japan's workplace AI moving from adoption to governance. Tools: the Surista survey "Corporate Generative AI Usage Survey 2026" (400 company employees nationwide, internet survey in May 2026, based on 153 workplace AI users) shows ChatGPT 60.8% (93 users), Google Gemini 49.7% (76 users) and Microsoft Copilot 41.8% (64 users) forming a "Big-Three Era," sharply separated from Claude 7.8% (12 users) and Perplexity 6.5% (10 users); 49.0% of users run multiple tools, 1.82 on average, and the leader changes with company size -- at small firms Gemini at 56.1% closes in on ChatGPT at 58.5%, while at large firms Copilot at 48.3% is stronger [F-001][F-002][F-003]. Costs: the LayerX survey (400 people who manage or track AI-usage costs at their companies, surveyed June 2026) finds 73.3% saying AI costs are "already" (19.0%) or "will within 1 year become" (54.3%) a management issue; 66.5% feel costs rose versus the prior year and 65.5% say costs are rising with AI-agent usage; surveyed companies' monthly AI costs average about 2.74 million yen, with 64.8% at 500,000 yen or more -- yet 82.8% report challenges in even grasping those costs [F-004][F-005][F-006][F-007]. Utilization: the Copado survey (110 Salesforce development/operations practitioners at companies with 300 or more employees) finds 62.8% already using generative AI / AI agents while about 40% have yet to start; 66.7% of the using group admit having "no system or mechanism to verify the correctness of AI outputs," 88.1% feel their work depends on specific individuals, and 86.4% expect AI to standardize it [F-008][F-009]. Global contrast: Thomson Reuters' survey of 1,816 professionals across 62 countries reveals an "execution gap" -- 74% use AI daily but 91% think their organizations are not fully realizing AI's value, and one in three (1/3) lawyers, accountants and compliance professionals use unapproved AI; TR estimates about US$143 billion of client revenue in the US legal and accounting markets could come under review [F-010][F-011]. Honest caveats: all four are sample surveys, three published by vendors with commercial stakes; percentages represent only each surveyed population.

Body

Event-chain overview: four surveys in one week -- workplace AI moves from "adoption" to "governance"

From 2026-06-30 to 2026-07-02, four AI-usage surveys were released in Japan: Surista's workplace AI tool-share survey (PRTIMES #1282102, July 1), LayerX's corporate AI cost-management survey (PRTIMES #1264738, June 30), Salesforce-focused DevOps company Copado's field survey (PRTIMES #1295577, July 2), and the Japanese-language release of Thomson Reuters' "2026 Future of Professionals Report" (PRTIMES #1283009, July 1). This card honestly links the four into one chain: tool share has settled into a "Big Three," costs have been elevated to a "management issue," about 40% of one enterprise field has yet to start, and global data shows the execution gap now carries an explicit price tag. The debate over workplace AI is shifting from "which one to adopt" to "how to govern it."

The framing must come first: all four are sample surveys, and LayerX (which sells an AI cost-management service), Copado (a Salesforce-focused DevOps vendor) and Thomson Reuters (which sells professional-grade AI) are all interested parties -- the figures themselves are official releases, but interpretation carries product context.

The Big-Three Era: ChatGPT 60.8%, Gemini 49.7%, Copilot 41.8% (based on 153 surveyed AI users)

According to Surista's "Corporate Generative AI Usage Survey 2026" (400 company employees nationwide, internet survey in May 2026), among the 153 respondents who use AI at work, tool-usage rates are ChatGPT 60.8% (93 users), Google Gemini 49.7% (76 users) and Microsoft Copilot 41.8% (64 users), with a sharp break below fourth place -- Claude 7.8% (12 users), Perplexity 6.5% (10 users), Notion AI 3.9% (6 users) (PRTIMES #1282102). [F-001] The publisher calls this the workplace AI "Big-Three Era" and describes the structure as the "free Big Three" commanding workplace AI -- the publisher's own framing terms.

Multi-tool behavior is equally central: 51.0% use only 1 tool (78 users), 34.0% use 2 (52 users), 7.8% use 3 (12 users) and 7.2% use 4 or more (11 users) -- meaning 49.0% of users run multiple tools, 1.82 on average (PRTIMES #1282102). [F-002] The survey supervisor commented that the establishment of the Big Three shows the debate has moved from "tool selection" to "designing how tools are used side by side," and that the ability to design an "in-house tool stack" will now generate competitive advantage -- the supervisor's interpretation.

Size changes the leader: Gemini closes in at small firms, Copilot stronger at large firms

The survey's size-based cross-tabulation (among AI users) shows: small companies (up to about 100 employees, 41 respondents) ChatGPT 58.5%, Gemini 56.1%, Copilot 22.0%; midsize companies (101-1,000 employees, 52 respondents) ChatGPT 65.4%, Copilot 50.0%, Gemini 48.1%; large companies (1,001 or more employees, 60 respondents) ChatGPT 58.3%, Copilot 48.3%, Gemini 46.7% (PRTIMES #1282102). [F-003] The publisher attributes Gemini's strength at small firms (56.1%) to wider Google Workspace adoption, and Copilot's strength at large firms (48.3%) to Microsoft 365 integration -- the publisher's interpretation. Honest caveat: each size sub-sample holds only 41-60 respondents behind the % figures above, so statistical noise is large; the industry sub-samples are even smaller and this card does not expand on them.

AI costs become a "management issue": 73.3% say already, or within 1 year

According to LayerX's "Corporate AI Usage & Cost Management Survey 2026" (covering 400 people who manage or track AI-usage costs at their companies; internet survey on 2026-06-12 and 13), 19.0% say AI-usage costs are "already a management issue" and 54.3% say they "will become one within 1 year" -- 73.3% combined (PRTIMES #1264738). [F-004] LayerX frames the background as the spread of AI agents (AI that autonomously processes tasks) driving up token consumption (the smallest unit by which AI processes text, billed by usage) and API usage -- the publisher's problem framing.

The felt cost increase is clear as of the June 2026 survey: 66.5% feel AI-usage costs rose versus the prior year (rose significantly 20.0% + rose somewhat 46.5%); 65.5% say costs are rising with the use of coding agents and business agents (rose significantly 16.5% + rose somewhat 49.0%) (PRTIMES #1264738). [F-005] On amounts: respondents' companies' monthly AI-usage costs run 500,000-1,000,000 yen for 26.3%, 1,000,000-5,000,000 yen for 20.5%, 5,000,000-10,000,000 yen for 8.8%, and 10,000,000 yen or more for 9.3% -- 64.8% combined at 500,000 yen or more per month; the average computed by the publisher is about 2.74 million yen (PRTIMES #1264738). [F-006] Framing: these are answers by surveyed AI-cost managers about their own companies, not statistics for all Japanese companies; the average was computed by LayerX.

Grasping the cost is itself the problem: 82.8% report challenges

In the same survey, 82.8% feel challenged in even grasping AI-usage costs (feel strongly 19.5% + feel somewhat 63.3%). The specific-challenge question was asked only of those who answered that they feel challenged (the 82.8% subset); among them, the most cited challenge is "concern over security / information-leak risk" at 30.5%, followed by "tracking employees' personal out-of-pocket payments" 25.1%, "invoiced amounts do not tie to usage breakdowns" 24.2%, "AI outcomes / ROI hard to measure" 24.2%, and "usage amounts hard to grasp" 24.2% -- the common axis is the invisibility of "who used what, and how much." Current tracking peaks at "spending by department/team" 32.7%, while the top item companies want to build next is "data linking AI spending to outcomes / ROI" 21.8% (PRTIMES #1264738). [F-007] Spending grows while visibility lags -- that is the substance of "becoming a management issue." Again noted: LayerX sells the AI cost-management service "Bakuraku AI Token Advisor," so the survey carries product context.

Adoption is not utilization: about 40% of the Salesforce field has yet to start, and 66.7% lack a verification system

According to Copado's "AI Usage Survey in Salesforce Development and Operations" (110 hands-on Salesforce development/operations practitioners at companies with 300 or more employees; internet survey on 2026-06-22 and 23), 62.8% already use generative AI / AI agents at work (company-wide 27.3% + in some teams/tasks 35.5%); meanwhile "still at study/trial stage" 26.4% plus "not using, no plans" 10.0% -- about 40% combined (the publisher's wording) -- have yet to move to implementation. Usage concentrates in "development (code generation / configuration)" at 68.1%, with "testing / verification" and "operations / maintenance" both at 60.9% (PRTIMES #1295577). [F-008]

Where is the wall? The using group's top challenge is "no system or mechanism to verify the correctness of AI outputs" at 66.7%, followed by "usage restricted by security / data governance" 59.4% and "effective prompting skills concentrated in few people" 50.7%; among non-users, "cannot judge which tool to choose" 45.0% and "security / governance concerns" 42.5% lead. At the same time, 88.1% feel their Salesforce work depends on specific individuals (strongly 33.6% + somewhat 54.5%), most citing "few in-house people can handle Apex and the like" 58.8%; yet 86.4% expect AI to standardize the work (strongly 28.2% + somewhat 58.2%), most citing "consistent-quality deliverables regardless of individual skill" 63.2%; and the most valued requirement for AI usage is "a mechanism to verify and test deliverables" at 54.5% (PRTIMES #1295577). [F-009] The structural message: the field pins standardization hopes on AI while admitting it lacks systems to verify AI's output -- a quality-assurance gap sits between adoption and utilization. Publisher Copado is a Salesforce-focused DevOps vendor, and every % figure above rests on a sample of 110 respondents, so interpretation requires reserve.

Global contrast: the professionals' "execution gap" and an estimated US$143 billion review risk

Zooming out to the world: according to Thomson Reuters' "2026 Future of Professionals Report" (1,816 professionals in legal, tax, audit, accounting, compliance, risk and global trade across 62 countries, surveyed in March-April 2026; the release headline says "1,800 professionals worldwide"), 74% of professionals already use AI tools daily, yet 91% think their organizations are not fully realizing the value AI could bring; one in three (1/3) lawyers, accountants and compliance professionals use AI not approved by their organizations (shadow AI), rising to 41% among those who feel their organization is moving too slowly. Behind this is dissatisfaction with approved tools: 96% call confidential-data protection essential, 94% verified authoritative content, and 90% explainable outputs -- while 41% cannot access professional AI tools meeting those standards (PRTIMES #1283009). [F-010]

The price is already being estimated: among professionals who feel a gap between "what AI makes possible" and "what my organization delivers," 24% are considering leaving within 2 years and 13% within 12 months; 62% say access to professional-grade AI is a factor in job-change decisions. On the client side (in-house departments that buy outside professional services), 78% consider AI-driven quality improvement essential or very important, but only 6% actually feel quality has improved; 32% of enterprise clients plan to review their provider relationships within 12 months, and about one-third (1/3) of those see engagements worth over US$1 million a year coming under review -- applying this to the US legal and CPA markets, TR estimates about US$143 billion of client revenue could become subject to reconsideration (PRTIMES #1283009). [F-011] Framing: US$143 billion is TR's estimate from applying survey data to the US market, not an actual result; departure intentions are intentions at survey time, not actual attrition.

The link to this site's consulting-shakeout card: two faces of AI's impact on knowledge work

Contrast with this site's published ANK-2026-06-05-001 ("Generative AI Culls the Management-Consulting Industry" -- 242 cumulative bankruptcies and closures in Japan's consulting industry in January-May 2026, on pace to top 600 cases for the year, the fastest since 2000): the two faces of AI's structural impact on knowledge work come into view. That card records the "substitution and shakeout face" -- undifferentiated knowledge services directly replaced by generative AI; this card records the "adoption and governance face" -- tools have settled into Big-Three infrastructure, costs have been elevated to a management issue, and the contest has moved to organizational capability in cost visibility and output verification. The two faces are two sides of the same structural change.

Risk factors

FAQ

Q: Which three generative AI tools are most used in Japanese workplaces in 2026?

ChatGPT (60.8%), Google Gemini (49.7%) and Microsoft Copilot (41.8%) -- splitting the workplace among the 153 surveyed workplace AI users, sharply separated from fourth place and below: Claude (7.8%) and Perplexity (6.5%).

The figures come from Surista's "Corporate Generative AI Usage Survey 2026" (400 company employees nationwide, internet survey in May 2026), with multiple answers allowed; "Big-Three Era" and "free Big Three" are the publisher's framing terms (PRTIMES #1282102).

Q: Do workplace AI users run multiple tools side by side?

Yes: 49.0% of surveyed AI users run multiple tools, 1.82 on average; 51.0% use only 1 tool (78 users), 34.0% use 2 (52 users), 7.8% use 3 (12 users) and 7.2% use 4 or more (11 users).

The survey supervisor commented that the debate has moved from "tool selection" to "designing side-by-side usage," and that "in-house tool stack" design skill will generate competitive advantage -- the supervisor's interpretation (PRTIMES #1282102).

Q: Does the leading tool change with company size?

Per the survey: at small companies (up to about 100 employees, 41 respondents) Gemini at 56.1% closes in on ChatGPT at 58.5% while Copilot sits at 22.0%; at large companies (1,001 or more employees, 60 respondents) Copilot at 48.3% tops Gemini at 46.7%.

The publisher attributes Gemini's small-firm strength (56.1%) to wider Google Workspace adoption and Copilot's large-firm strength (48.3%) to Microsoft 365 integration -- the publisher's interpretation; sub-samples of 41-60 respondents behind these % figures mean large statistical noise (PRTIMES #1282102).

Q: Have AI costs really become a management issue?

Among the 400 surveyed AI-cost managers in the LayerX survey (June 2026), 19.0% say AI costs are "already a management issue" and 54.3% say they "will become one within 1 year" -- 73.3% combined; 66.5% feel costs rose versus the prior year and 65.5% say they are rising with AI-agent usage.

LayerX frames the background as AI agents driving up token consumption and API usage; LayerX sells an AI cost-management service, so the survey carries product context (PRTIMES #1264738).

Q: How much do Japanese companies spend on AI per month?

Per the survey, 64.8% of respondents' companies spend 500,000 yen or more a month on AI usage (500,000-1,000,000 yen: 26.3%; 1,000,000-5,000,000: 20.5%; 5,000,000-10,000,000: 8.8%; 10,000,000 or more: 9.3%), and the publisher-computed average is about 2.74 million yen.

These are answers by the 400 surveyed AI-cost managers about their own companies, not statistics for all Japanese companies; the average was computed by LayerX (PRTIMES #1264738).

Q: Can companies actually grasp their AI costs?

Mostly not: 82.8% report challenges in grasping AI costs; the specific-challenge question was asked only of that challenge-feeling 82.8% subset, among whom the top challenge is "security / information-leak risk" at 30.5%, followed by employees' personal out-of-pocket payments 25.1% and invoices not tying to usage breakdowns 24.2% -- "who used what, and how much" is invisible.

Current tracking peaks at the broad level of "spending by department/team" (32.7%), while the top item companies want next is "data linking AI spending to outcomes / ROI" (21.8%) (PRTIMES #1264738).

Q: Does adopting AI mean actually using it well?

No. In the Copado survey (110 Salesforce practitioners at companies with 300 or more employees), 62.8% already use AI, but about 40% remain at study/trial stage (26.4%) or have no plans (10.0%); and 66.7% of the using group admit having "no system to verify the correctness of AI outputs."

Meanwhile 88.1% feel the work depends on specific individuals (scarce Apex-capable staff, 58.8%, is the top reason) and 86.4% expect AI to standardize it -- the field hopes for standardization while lacking quality-verification systems; Copado is a Salesforce-focused DevOps vendor (PRTIMES #1295577).

Q: Seen globally, how large is the price of the AI "execution gap"?

Thomson Reuters' survey of 1,816 professionals across 62 countries: 74% use AI daily yet 91% think their organizations are not fully realizing its value; 32% of enterprise clients plan to review provider relationships within 12 months, and TR estimates about US$143 billion of client revenue in the US legal and accounting markets could come under review -- an estimate, not an actual result.

On talent: among gap-feeling professionals, 24% are considering leaving within 2 years and 13% within 12 months; one in three (1/3) lawyers, accountants and compliance professionals use unapproved AI (shadow AI) (PRTIMES #1283009).

F-Units

F-001: Surista's "Corporate Generative AI Usage Survey 2026" (400 company employees nationwide, internet survey in May 2026): among 153 workplace AI users, tool-usage rates are ChatGPT 60.8% (93 users), Google Gemini 49.7% (76 users), Microsoft Copilot 41.8% (64 users), Claude 7.8% (12 users), Perplexity 6.5% (10 users), Notion AI 3.9% (6 users) - source: PRTIMES #1282102 - source_url: https://prtimes.jp/main/html/rd/p/000000014.000169560.html - confidence: high - basis: official_statement - period: surveyed May 2026, released 2026-07-01 - caveat: internet survey (Freeasy), multiple answers; the population is the 153 workplace AI users among 400 surveyed company employees, not a census of all companies; "Big-Three Era" and "free Big Three" are the publisher's framing terms

F-002: Same survey: among the 153 AI users, 51.0% use only 1 tool (78 users), 34.0% use 2 (52 users), 7.8% use 3 (12 users), 7.2% use 4 or more (11 users); 49.0% run multiple tools, 1.82 on average - source: PRTIMES #1282102 - source_url: https://prtimes.jp/main/html/rd/p/000000014.000169560.html - confidence: high - basis: official_statement - period: surveyed May 2026, released 2026-07-01 - caveat: the 1.82-tool average is computed by the publisher; "from single selection to side-by-side usage" is the survey supervisor's interpretation

F-003: Same survey, by company size (among AI users): small companies (up to about 100 employees, 41 respondents) ChatGPT 58.5%, Gemini 56.1%, Copilot 22.0%; midsize companies (101-1,000 employees, 52 respondents) ChatGPT 65.4%, Gemini 48.1%, Copilot 50.0%; large companies (1,001 or more employees, 60 respondents) ChatGPT 58.3%, Gemini 46.7%, Copilot 48.3% - source: PRTIMES #1282102 - source_url: https://prtimes.jp/main/html/rd/p/000000014.000169560.html - confidence: medium - basis: official_statement - period: surveyed May 2026, released 2026-07-01 - caveat: size sub-samples of only 41-60 respondents carry large statistical noise; "Gemini closes in at small firms due to Google Workspace, Copilot strong at large firms due to Microsoft 365 integration" is the publisher's interpretation

F-004: LayerX's "Corporate AI Usage & Cost Management Survey 2026" (400 people who manage or track AI-usage costs at their companies; internet survey 2026-06-12 to 13): 19.0% say AI-usage costs are "already a management issue," 54.3% say they "will become one within 1 year" -- 73.3% combined - source: PRTIMES #1264738 - source_url: https://prtimes.jp/main/html/rd/p/000000625.000036528.html - confidence: high - basis: official_statement - period: surveyed 2026-06-12 to 2026-06-13, released 2026-06-30 - caveat: the population is the 400 surveyed people who manage/track AI-usage costs (LayerX phrases it as "73.3% of companies"; this card frames it as a respondent share); LayerX sells the AI cost-management service "Bakuraku AI Token Advisor," so the survey carries product context

F-005: Same survey: 66.5% feel AI-usage costs rose versus the prior year (rose significantly 20.0% + rose somewhat 46.5%); 65.5% say costs are rising with AI-agent (coding-agent / business-agent) usage (rose significantly 16.5% + rose somewhat 49.0%) - source: PRTIMES #1264738 - source_url: https://prtimes.jp/main/html/rd/p/000000625.000036528.html - confidence: high - basis: official_statement - period: surveyed 2026-06-12 to 2026-06-13, released 2026-06-30 - caveat: "rose" is respondents' perception, not an accounting measurement; AI agents driving token consumption is LayerX's problem framing

F-006: Same survey: respondents' companies' monthly AI-usage costs are 500,000-1,000,000 yen for 26.3%, 1,000,000-5,000,000 yen for 20.5%, 5,000,000-10,000,000 yen for 8.8%, 10,000,000 yen or more for 9.3% -- 64.8% combined at 500,000 yen or more per month; the publisher-computed average is about 2.74 million yen - source: PRTIMES #1264738 - source_url: https://prtimes.jp/main/html/rd/p/000000625.000036528.html - confidence: medium - basis: official_statement - period: surveyed 2026-06-12 to 2026-06-13, released 2026-06-30 - caveat: the roughly 2.74-million-yen average is computed by LayerX from respondents' answers; the population is the surveyed AI-cost managers' companies, not statistics for all Japanese companies; the four brackets (26.3% + 20.5% + 8.8% + 9.3%) sum to 64.9% while the release states 64.8% -- a 0.1-percentage-point rounding difference; this card quotes the release's figures as published

F-007: Same survey: 82.8% feel challenged in grasping AI-usage costs (feel strongly 19.5% + feel somewhat 63.3%); among those who answered that they feel challenged (the 82.8% subset), the top specific challenges are "security / information-leak risk concern" 30.5%, then "tracking employees' personal out-of-pocket payments" 25.1%, "invoiced amounts do not tie to usage breakdowns" 24.2%, "AI outcomes / ROI hard to measure" 24.2%, "usage amounts hard to grasp" 24.2%; current tracking peaks at "spending by department/team" 32.7%, and the top item to build next is "data linking AI spending to outcomes / ROI" 21.8% - source: PRTIMES #1264738 - source_url: https://prtimes.jp/main/html/rd/p/000000625.000036528.html - confidence: high - basis: official_statement - period: surveyed 2026-06-12 to 2026-06-13, released 2026-06-30 - caveat: the "grasping challenge" question covers all 400 respondents of this survey; the specific-challenge question was asked only of those who answered that they feel challenged (the 82.8% subset), so 30.5% and the other shares use that subset -- not all 400 respondents -- as the denominator; specific challenges allow multiple answers, so shares can total over 100%

F-008: Copado's "AI Usage Survey in Salesforce Development and Operations" (110 hands-on Salesforce development/operations practitioners at companies with 300 or more employees; internet survey 2026-06-22 to 23): 62.8% use generative AI / AI agents at work (company-wide 27.3% + some teams/tasks 35.5%); "study/trial stage" 26.4% + "not using, no plans" 10.0% = about 40% yet to implement (the publisher's wording); usage areas led by "development (code generation / configuration)" 68.1%, with "testing / verification" and "operations / maintenance" both 60.9% - source: PRTIMES #1295577 - source_url: https://prtimes.jp/main/html/rd/p/000000013.000155348.html - confidence: high - basis: official_statement - period: surveyed 2026-06-22 to 2026-06-23, released 2026-07-02 - caveat: sample of 110 respondents (internet survey planned by IDEATECH's "Research PR"); "about 40%" is the publisher's summary wording for 26.4% plus 10.0%; Copado is a Salesforce-focused DevOps vendor, so the survey carries product context

F-009: Same survey: the using group's top challenge is "no system or mechanism to verify the correctness of AI outputs" 66.7%, then "usage restricted by security / data governance" 59.4%, "prompting skills concentrated in few people" 50.7%; non-users cite "cannot judge which tool to choose" 45.0%, "security / governance concerns" 42.5%; 88.1% feel the work depends on specific individuals (strongly 33.6% + somewhat 54.5%), most citing "few in-house people can handle Apex and the like" 58.8%; 86.4% expect AI to standardize the work (strongly 28.2% + somewhat 58.2%), most citing "consistent-quality deliverables regardless of skill" 63.2%; the most valued requirement is "a mechanism to verify and test deliverables" 54.5% - source: PRTIMES #1295577 - source_url: https://prtimes.jp/main/html/rd/p/000000013.000155348.html - confidence: high - basis: official_statement - period: surveyed 2026-06-22 to 2026-06-23, released 2026-07-02 - caveat: multiple answers, shares can total over 100%; "standardization hopes" are respondents' intentions, not realized effects

F-010: Thomson Reuters' "2026 Future of Professionals Report" (1,816 professionals in legal, tax, audit, accounting, compliance, risk and global trade across 62 countries, surveyed March-April 2026): 74% of professionals use AI tools daily; 91% say their organizations are not fully realizing the value AI could bring; one in three (1/3) lawyers, accountants and compliance professionals use AI not approved by their organizations (41% among those who feel their organization is too slow); 96% call confidential-data protection, 94% verified authoritative content and 90% explainable outputs essential, while 41% cannot access professional AI tools meeting those standards - source: PRTIMES #1283009 - source_url: https://prtimes.jp/main/html/rd/p/000000015.000072024.html - confidence: high - basis: official_statement - period: surveyed March-April 2026, Japanese-language release 2026-07-01 - caveat: the release headline says "1,800 professionals worldwide" while the report overview records 1,816; this card adopts the report-overview figure; TR itself sells professional-grade AI (self-described "Fiduciary-Grade AI"), so the survey carries product context

F-011: Same report: among professionals who feel a gap between "what AI makes possible" and "what my organization delivers," 24% are considering leaving within 2 years and 13% within 12 months; 62% say access to professional-grade AI is a job-change factor; 78% of clients (in-house departments buying outside professional services) call AI-driven quality improvement essential or very important while only 6% actually feel improvement; 32% of enterprise clients plan to review provider relationships within 12 months, and about one-third (1/3) of those see engagements worth over US$1 million a year coming under review -- from which TR estimates about US$143 billion of client revenue in the US legal and CPA markets could become subject to reconsideration - source: PRTIMES #1283009 - source_url: https://prtimes.jp/main/html/rd/p/000000015.000072024.html - confidence: medium - basis: official_statement - period: surveyed March-April 2026, Japanese-language release 2026-07-01 - caveat: the roughly US$143 billion is TR's estimate applying survey results to the US legal and CPA markets (an estimate, not an actual result); departure intentions are intentions at survey time, not actual attrition; the "13%" is stated once as among gap-feeling professionals and elsewhere as all respondents -- this card records both as written

J-Units

J-001: The workplace AI debate has moved from "which one to pick" to "how to combine and govern" -- the Big Three (ChatGPT 60.8%, Gemini 49.7%, Copilot 41.8%) are sharply separated from fourth place and below, 49.0% of users run multiple tools (1.82 on average), and company size changes the leader (Gemini 56.1% closes in at small firms; Copilot 48.3% stronger at large firms); "competitive advantage comes from in-house tool-stack design" is the survey supervisor's interpretation, and small sub-samples call for reserve - confidence: medium - basis: official_statement

J-002: AI costs are being elevated from "IT expense" to "management issue" -- 73.3% of surveyed AI-cost managers say already or within 1 year, 64.8% of their companies spend 500,000 yen or more a month (about 2.74 million yen on average), 66.5% feel the increase and 65.5% attribute it to AI agents; yet 82.8% struggle even to grasp the costs -- spending is outgrowing visibility, and that is the governance gap; LayerX and Copado are both interested-vendor surveys to be read with product context - confidence: medium - basis: official_statement

J-003: The "adoption is not utilization" execution gap appears simultaneously in Japan's field data and among global professionals -- 62.8% of the Salesforce field uses AI but 66.7% lack verification systems and about 40% have yet to start; globally 74% use AI daily yet 91% see value unrealized, and TR estimates about US$143 billion of US revenue under review (an estimate). Linked to this site's ANK-2026-06-05-001 (generative AI culling the consulting industry: 242 cumulative cases in January-May 2026, on pace to top 600 for the year): AI's structural impact on knowledge work has one face of "substitution and shakeout" and this card's face of "adoption and governance" -- tools have settled into Big-Three infrastructure, and the contest has moved to cost visibility and output verification - confidence: medium - basis: official_statement

P-Units

P-001: Whether the Big-Three structure and the size-based power balance solidify -- track subsequent survey rounds (Gemini's share at small firms, Copilot's at large firms; whether Claude and Perplexity escape single-digit usage rates), and whether the "free Big Three" framing survives scrutiny as paid plans spread ### P-002: How AI costs' "elevation to management issue" actually progresses -- follow-up data on the roughly 2.74-million-yen monthly average and token consumption as AI agents spread, adoption of "AI spending x outcomes/ROI" management, and competition in the cost-visibility tool market ### P-003: Whether the execution gap converges -- build-out rates of AI output-verification systems, the share of unapproved AI (shadow AI) usage, and how much of TR's estimated client-relationship review (32% of enterprise clients planning reviews within 12 months) actually materializes over the next 12 months

同事件・三視角 / Three Perspectives on the Same Event / 同一イベント・三つの視点

Internal Citation Chain

Published ANK-Docs cited by this article: - ANK-2026-06-05-001 (Generative AI Culls the Management-Consulting Industry: Structural Destruction of Knowledge-Labor Services -- 242 cumulative bankruptcies and closures in January-May 2026, on pace to top 600 cases for the year) -> cited as the contrasting face of "AI hitting knowledge work": that card records the "substitution and shakeout face" where undifferentiated knowledge services are directly replaced by generative AI, while this article records the "adoption and governance face" -- Big-Three tool consolidation, costs elevated to a management issue, and missing verification systems. The two faces are two sides of the same structural change.

Sources

1. [PRTIMES #1282102] Surista (株式会社スリスタ), "[2026 Latest] Workplace AI Enters the 'Big-Three Era' -- ChatGPT 60.8%, Gemini 49.7%, Copilot 41.8%, Claude at 7.8%", 2026-07-01. https://prtimes.jp/main/html/rd/p/000000014.000169560.html 2. [PRTIMES #1264738] LayerX (株式会社LayerX), "Corporate AI Usage & Cost Management Survey 2026: over 70% of companies say AI usage costs are already or will soon become a management issue", 2026-06-30. https://prtimes.jp/main/html/rd/p/000000625.000036528.html 3. [PRTIMES #1295577] Copado (コパード株式会社), "AI Usage Survey in Salesforce Development and Operations: about 40% 'not yet using' AI, citing tool-selection difficulty and security concerns", 2026-07-02. https://prtimes.jp/main/html/rd/p/000000013.000155348.html 4. [PRTIMES #1283009] Thomson Reuters (トムソン・ロイター), "Thomson Reuters survey -- AI is ready, but organizations lag: slow adoption invites client attrition and talent flight", 2026-07-01. https://prtimes.jp/main/html/rd/p/000000015.000072024.html 5. [ANK-2026-06-05-001] Rin Takenouchi, "Generative AI Culls the Management-Consulting Industry: Structural Destruction of Knowledge-Labor Services (Fastest Pace Since 2000)", 2026-06-05. https://ainews.washinmura.jp/ainews/en/ank/ANK-2026-06-05-001

📊 引用級事實單元（F-Units）

Surista's "Corporate Generative AI Usage Survey 2026" (400 company employees nationwide, internet survey in May 2026): among 153 workplace AI users, tool-usage rates are ChatGPT 60.8% (93 users), Google Gemini 49.7% (76 users), Microsoft Copilot 41.8% (64 users), Claude 7.8% (12 users), Perplexity 6.5% (10 users), Notion AI 3.9% (6 users)
F-001 · Confidence: high · Basis: official_statement PRTIMES #1282102 surveyed May 2026, released 2026-07-01
Same survey: among the 153 AI users, 51.0% use only 1 tool (78 users), 34.0% use 2 (52 users), 7.8% use 3 (12 users), 7.2% use 4 or more (11 users); 49.0% run multiple tools, 1.82 on average
F-002 · Confidence: high · Basis: official_statement PRTIMES #1282102 surveyed May 2026, released 2026-07-01
Same survey, by company size (among AI users): small companies (up to about 100 employees, 41 respondents) ChatGPT 58.5%, Gemini 56.1%, Copilot 22.0%; midsize companies (101-1,000 employees, 52 respondents) ChatGPT 65.4%, Gemini 48.1%, Copilot 50.0%; large companies (1,001 or more employees, 60 respondents) ChatGPT 58.3%, Gemini 46.7%, Copilot 48.3%
F-003 · Confidence: medium · Basis: official_statement PRTIMES #1282102 surveyed May 2026, released 2026-07-01
LayerX's "Corporate AI Usage & Cost Management Survey 2026" (400 people who manage or track AI-usage costs at their companies; internet survey 2026-06-12 to 13): 19.0% say AI-usage costs are "already a management issue," 54.3% say they "will become one within 1 year" -- 73.3% combined
F-004 · Confidence: high · Basis: official_statement PRTIMES #1264738 surveyed 2026-06-12 to 2026-06-13, released 2026-06-30
Same survey: 66.5% feel AI-usage costs rose versus the prior year (rose significantly 20.0% + rose somewhat 46.5%); 65.5% say costs are rising with AI-agent (coding-agent / business-agent) usage (rose significantly 16.5% + rose somewhat 49.0%)
F-005 · Confidence: high · Basis: official_statement PRTIMES #1264738 surveyed 2026-06-12 to 2026-06-13, released 2026-06-30
Same survey: respondents' companies' monthly AI-usage costs are 500,000-1,000,000 yen for 26.3%, 1,000,000-5,000,000 yen for 20.5%, 5,000,000-10,000,000 yen for 8.8%, 10,000,000 yen or more for 9.3% -- 64.8% combined at 500,000 yen or more per month; the publisher-computed average is about 2.74 million yen
F-006 · Confidence: medium · Basis: official_statement PRTIMES #1264738 surveyed 2026-06-12 to 2026-06-13, released 2026-06-30
Same survey: 82.8% feel challenged in grasping AI-usage costs (feel strongly 19.5% + feel somewhat 63.3%); among those who answered that they feel challenged (the 82.8% subset), the top specific challenges are "security / information-leak risk concern" 30.5%, then "tracking employees' personal out-of-pocket payments" 25.1%, "invoiced amounts do not tie to usage breakdowns" 24.2%, "AI outcomes / ROI hard to measure" 24.2%, "usage amounts hard to grasp" 24.2%; current tracking peaks at "spending by department/team" 32.7%, and the top item to build next is "data linking AI spending to outcomes / ROI" 21.8%
F-007 · Confidence: high · Basis: official_statement PRTIMES #1264738 surveyed 2026-06-12 to 2026-06-13, released 2026-06-30
Copado's "AI Usage Survey in Salesforce Development and Operations" (110 hands-on Salesforce development/operations practitioners at companies with 300 or more employees; internet survey 2026-06-22 to 23): 62.8% use generative AI / AI agents at work (company-wide 27.3% + some teams/tasks 35.5%); "study/trial stage" 26.4% + "not using, no plans" 10.0% = about 40% yet to implement (the publisher's wording); usage areas led by "development (code generation / configuration)" 68.1%, with "testing / verification" and "operations / maintenance" both 60.9%
F-008 · Confidence: high · Basis: official_statement PRTIMES #1295577 surveyed 2026-06-22 to 2026-06-23, released 2026-07-02
Same survey: the using group's top challenge is "no system or mechanism to verify the correctness of AI outputs" 66.7%, then "usage restricted by security / data governance" 59.4%, "prompting skills concentrated in few people" 50.7%; non-users cite "cannot judge which tool to choose" 45.0%, "security / governance concerns" 42.5%; 88.1% feel the work depends on specific individuals (strongly 33.6% + somewhat 54.5%), most citing "few in-house people can handle Apex and the like" 58.8%; 86.4% expect AI to standardize the work (strongly 28.2% + somewhat 58.2%), most citing "consistent-quality deliverables regardless of skill" 63.2%; the most valued requirement is "a mechanism to verify and test deliverables" 54.5%
F-009 · Confidence: high · Basis: official_statement PRTIMES #1295577 surveyed 2026-06-22 to 2026-06-23, released 2026-07-02
Thomson Reuters' "2026 Future of Professionals Report" (1,816 professionals in legal, tax, audit, accounting, compliance, risk and global trade across 62 countries, surveyed March-April 2026): 74% of professionals use AI tools daily; 91% say their organizations are not fully realizing the value AI could bring; one in three (1/3) lawyers, accountants and compliance professionals use AI not approved by their organizations (41% among those who feel their organization is too slow); 96% call confidential-data protection, 94% verified authoritative content and 90% explainable outputs essential, while 41% cannot access professional AI tools meeting those standards
F-010 · Confidence: high · Basis: official_statement PRTIMES #1283009 surveyed March-April 2026, Japanese-language release 2026-07-01
Same report: among professionals who feel a gap between "what AI makes possible" and "what my organization delivers," 24% are considering leaving within 2 years and 13% within 12 months; 62% say access to professional-grade AI is a job-change factor; 78% of clients (in-house departments buying outside professional services) call AI-driven quality improvement essential or very important while only 6% actually feel improvement; 32% of enterprise clients plan to review provider relationships within 12 months, and about one-third (1/3) of those see engagements worth over US$1 million a year coming under review -- from which TR estimates about US$143 billion of client revenue in the US legal and CPA markets could become subject to reconsideration
F-011 · Confidence: medium · Basis: official_statement PRTIMES #1283009 surveyed March-April 2026, Japanese-language release 2026-07-01

❓ FAQ

Which three generative AI tools are most used in Japanese workplaces in 2026?

ChatGPT (60.8%), Google Gemini (49.7%) and Microsoft Copilot (41.8%) -- splitting the workplace among the 153 surveyed workplace AI users, sharply separated from fourth place and below: Claude (7.8%) and Perplexity (6.5%). The figures come from Surista's "Corporate Generative AI Usage Survey 2026" (400 company employees nationwide, internet survey in May 2026), with multiple answers allowed; "Big-Three Era" and "free Big Three" are the publisher's framing terms (PRTIMES #1282102).

Do workplace AI users run multiple tools side by side?

Yes: 49.0% of surveyed AI users run multiple tools, 1.82 on average; 51.0% use only 1 tool (78 users), 34.0% use 2 (52 users), 7.8% use 3 (12 users) and 7.2% use 4 or more (11 users). The survey supervisor commented that the debate has moved from "tool selection" to "designing side-by-side usage," and that "in-house tool stack" design skill will generate competitive advantage -- the supervisor's interpretation (PRTIMES #1282102).

Does the leading tool change with company size?

Per the survey: at small companies (up to about 100 employees, 41 respondents) Gemini at 56.1% closes in on ChatGPT at 58.5% while Copilot sits at 22.0%; at large companies (1,001 or more employees, 60 respondents) Copilot at 48.3% tops Gemini at 46.7%. The publisher attributes Gemini's small-firm strength (56.1%) to wider Google Workspace adoption and Copilot's large-firm strength (48.3%) to Microsoft 365 integration -- the publisher's interpretation; sub-samples of 41-60 respondents behind these % figures mean large statistical noise (PRTIMES #1282102).

Have AI costs really become a management issue?

Among the 400 surveyed AI-cost managers in the LayerX survey (June 2026), 19.0% say AI costs are "already a management issue" and 54.3% say they "will become one within 1 year" -- 73.3% combined; 66.5% feel costs rose versus the prior year and 65.5% say they are rising with AI-agent usage. LayerX frames the background as AI agents driving up token consumption and API usage; LayerX sells an AI cost-management service, so the survey carries product context (PRTIMES #1264738).

How much do Japanese companies spend on AI per month?

Per the survey, 64.8% of respondents' companies spend 500,000 yen or more a month on AI usage (500,000-1,000,000 yen: 26.3%; 1,000,000-5,000,000: 20.5%; 5,000,000-10,000,000: 8.8%; 10,000,000 or more: 9.3%), and the publisher-computed average is about 2.74 million yen. These are answers by the 400 surveyed AI-cost managers about their own companies, not statistics for all Japanese companies; the average was computed by LayerX (PRTIMES #1264738).

Can companies actually grasp their AI costs?

Mostly not: 82.8% report challenges in grasping AI costs; the specific-challenge question was asked only of that challenge-feeling 82.8% subset, among whom the top challenge is "security / information-leak risk" at 30.5%, followed by employees' personal out-of-pocket payments 25.1% and invoices not tying to usage breakdowns 24.2% -- "who used what, and how much" is invisible. Current tracking peaks at the broad level of "spending by department/team" (32.7%), while the top item companies want next is "data linking AI spending to outcomes / ROI" (21.8%) (PRTIMES #1264738).

Does adopting AI mean actually using it well?

No. In the Copado survey (110 Salesforce practitioners at companies with 300 or more employees), 62.8% already use AI, but about 40% remain at study/trial stage (26.4%) or have no plans (10.0%); and 66.7% of the using group admit having "no system to verify the correctness of AI outputs." Meanwhile 88.1% feel the work depends on specific individuals (scarce Apex-capable staff, 58.8%, is the top reason) and 86.4% expect AI to standardize it -- the field hopes for standardization while lacking quality-verification systems; Copado is a Salesforce-focused DevOps vendor (PRTIMES #1295577).

Seen globally, how large is the price of the AI "execution gap"?

Thomson Reuters' survey of 1,816 professionals across 62 countries: 74% use AI daily yet 91% think their organizations are not fully realizing its value; 32% of enterprise clients plan to review provider relationships within 12 months, and TR estimates about US$143 billion of client revenue in the US legal and accounting markets could come under review -- an estimate, not an actual result. On talent: among gap-feeling professionals, 24% are considering leaving within 2 years and 13% within 12 months; one in three (1/3) lawyers, accountants and compliance professionals use unapproved AI (shadow AI) (PRTIMES #1283009). ---

🧠 編輯判斷（J-Units）

The workplace AI debate has moved from "which one to pick" to "how to combine and govern" -- the Big Three (ChatGPT 60.8%, Gemini 49.7%, Copilot 41.8%) are sharply separated from fourth place and below, 49.0% of users run multiple tools (1.82 on average), and company size changes the leader (Gemini 56.1% closes in at small firms; Copilot 48.3% stronger at large firms); "competitive advantage comes from in-house tool-stack design" is the survey supervisor's interpretation, and small sub-samples call for reserve
Confidence: medium
AI costs are being elevated from "IT expense" to "management issue" -- 73.3% of surveyed AI-cost managers say already or within 1 year, 64.8% of their companies spend 500,000 yen or more a month (about 2.74 million yen on average), 66.5% feel the increase and 65.5% attribute it to AI agents; yet 82.8% struggle even to grasp the costs -- spending is outgrowing visibility, and that is the governance gap; LayerX and Copado are both interested-vendor surveys to be read with product context
Confidence: medium
The "adoption is not utilization" execution gap appears simultaneously in Japan's field data and among global professionals -- 62.8% of the Salesforce field uses AI but 66.7% lack verification systems and about 40% have yet to start; globally 74% use AI daily yet 91% see value unrealized, and TR estimates about US$143 billion of US revenue under review (an estimate). Linked to this site's ANK-2026-06-05-001 (generative AI culling the consulting industry: 242 cumulative cases in January-May 2026, on pace to top 600 for the year): AI's structural impact on knowledge work has one face of "substitution and shakeout" and this card's face of "adoption and governance" -- tools have settled into Big-Three infrastructure, and the contest has moved to cost visibility and output verification
Confidence: medium

🔮 待驗證假設（P-Units）

Whether the Big-Three structure and the size-based power balance solidify -- track subsequent survey rounds (Gemini's share at small firms, Copilot's at large firms; whether Claude and Perplexity escape single-digit usage rates), and whether the "free Big Three" framing survives scrutiny as paid plans spread
Status: open
How AI costs' "elevation to management issue" actually progresses -- follow-up data on the roughly 2.74-million-yen monthly average and token consumption as AI agents spread, adoption of "AI spending x outcomes/ROI" management, and competition in the cost-visibility tool market
Status: open
Whether the execution gap converges -- build-out rates of AI output-verification systems, the share of unapproved AI (shadow AI) usage, and how much of TR's estimated client-relationship review (32% of enterprise clients planning reviews within 12 months) actually materializes over the next 12 months
Status: open

Verification Record

Editorial selection, human-supervised — Takenouchi Rin (Editor-in-Chief)

Cross-verified by multiple AI models.

Related Analyses