Read the footnote

What the 95% measured

In August 2025 a statistic went round the business press: 95% of companies were getting zero return from their spending on generative AI. It came from a report by MIT's NANDA project. The report does say "95% of organizations are getting zero return". Two sentences later it says "Just 5% of integrated AI pilots are extracting millions in value."11MIT NANDA, "The GenAI Divide: State of AI in Business 2025", as served by MIT on 18 Aug 2025 -- "95% of organizations are getting zero return"; "Just 5% of integrated AI pilots are extracting millions in value"; "no measurable P&L impact"; "ROI impact measured 6 months post-pilot"; "Preliminary Findings". https://web.archive.org/web/20250818115520/https://nanda.media.mit.edu/ai_report_2025.pdf (T3 -- research report, preliminary) Those are different claims. One counts organizations; the other counts pilot projects.

Read further and the number narrows again. MIT measured "ROI impact ... 6 months post-pilot". A return meant a "measurable P&L impact", as described by users or executives, not figures from company accounts. The 5% that succeeded were task-specific tools, not AI in general. The report calls itself "Preliminary Findings".11MIT NANDA, "The GenAI Divide: State of AI in Business 2025", as served by MIT on 18 Aug 2025 -- "95% of organizations are getting zero return"; "Just 5% of integrated AI pilots are extracting millions in value"; "no measurable P&L impact"; "ROI impact measured 6 months post-pilot"; "Preliminary Findings". https://web.archive.org/web/20250818115520/https://nanda.media.mit.edu/ai_report_2025.pdf (T3 -- research report, preliminary)

None of this was hidden; it was all in the report. It did not travel with the number. Axios wrote that "95% of organizations studied get zero return on their AI investment";22Axios, 21 Aug 2025 -- "95% of organizations studied get zero return on their AI investment". https://www.axios.com/2025/08/21/ai-wall-street-big-tech (T4 -- news coverage) HBR, that "95% of investments in gen AI have produced zero returns".33Harvard Business Review, 29 Aug 2025 -- "95% of investments in gen AI have produced zero returns". https://hbr.org/2025/08/beware-the-ai-experimentation-trap (T4 -- commentary) Each took one of the report's own phrasings and left the rest behind.

We traced three more widely shared numbers about AI's payoff, from EY, PwC and McKinsey, back to the surveys behind them. Each had lost a detail that changes what it means.

Four numbers, traced

Three things stood out.

In three of the four, the organization that ran the survey cut the detail first. EY's press release said "half (50%) of CEOs surveyed" identified "AI technologies as the single largest contributor to productivity gains."55EY, press release, 1 Oct 2026 -- "half (50%) of CEOs surveyed identifying AI technologies as the single largest contributor to productivity gains"; "more than half (56%) of respondents report at least some reliable visibility into AI costs and returns". https://www.ey.com/en_gl/newsroom/2026/10/ai-drives-productivity-but-ceos-struggle-to-turn-gains-into-grow (T3 -- publisher press release) The chart it summarizes says CEOs ranked all factors, and "Percentages represent the top three rankings combined."66EY-Parthenon CEO Outlook, October 2026, chart "AI delivers the most value when paired with other levers of transformation", note -- "Percentages represent the top three rankings combined." https://www.ey.com/en_gl/ceo/ceo-outlook-global-report (T3 -- industry survey report) Half the CEOs put AI somewhere in their top three; the survey does not say how many put it first. PwC's report says 56% saw neither higher revenues nor lower costs from AI over 12 months;88PwC, 29th Global CEO Survey, 2026, p. 6 -- "More than half (56%) say their company has seen neither higher revenues nor lower costs from AI, while only one in eight (12%) report both of these positive impacts." Question (p. 7): "In the last 12 months, what impact did AI have on the following at your company?" 4,454 CEOs, fielded 30 Sep - 10 Nov 2025 (p. 23). https://www.pwc.com/gx/en/ceo-survey/2026/pwc-ceo-survey-2026.pdf (T3 -- industry survey report) its press release says "no significant financial benefit to date".77PwC, press release, 19 Jan 2026 -- "56% say they have seen no significant financial benefit to date". https://www.pwc.com/gx/en/news-room/press-releases/2026/pwc-2026-global-ceo-survey.html (T3 -- publisher press release) MIT's loose line sits in the report itself.

A shortened number can understate as easily as overstate. EY's release says "more than half (56%)" have "at least some reliable visibility" into AI costs and returns.55EY, press release, 1 Oct 2026 -- "half (50%) of CEOs surveyed identifying AI technologies as the single largest contributor to productivity gains"; "more than half (56%) of respondents report at least some reliable visibility into AI costs and returns". https://www.ey.com/en_gl/newsroom/2026/10/ai-drives-productivity-but-ceos-struggle-to-turn-gains-into-grow (T3 -- publisher press release) On the chart, 56% chose that option and a further 16% chose a stronger one, so 72% have at least reliable visibility.1414EY-Parthenon CEO Outlook, October 2026, chart "Most CEOs can estimate AI cost and returns, but only a few have a complete view" -- 56% "We have reliable visibility into most AI costs and can assess ROI with reasonable confidence."; 16% "We have comprehensive, real-time visibility into AI costs and can assess ROI with great confidence." https://www.ey.com/en_gl/ceo/ceo-outlook-global-report (T3 -- industry survey report) The release undersells its own data.

Some numbers are made up. Tech Times headlined McKinsey's 2026 survey "Record AI Spending Can't Move Earnings Needle for 94% of Enterprises".1515Tech Times, 26 Aug 2026 -- headline "Record AI Spending Can't Move Earnings Needle for 94% of Enterprises, McKinsey Finds". https://www.techtimes.com/articles/325590/20260826/record-ai-spending-cant-move-earnings-needle-94-enterprises-mckinsey-finds.htm (T4 -- news coverage) No 94% appears in McKinsey's text. It may be 100 minus the 6% McKinsey classes as "high performers", a stricter bar than "any EBIT impact";1111McKinsey, "The state of AI in 2026", Exhibit 6 note -- "The 2026 survey measured impact at the business-function level rather than aggregating from individual use cases within each function." https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai (T3 -- industry survey report) that derivation is our inference.

For MIT, PwC and McKinsey, at least one outlet kept the details in its article; none of the headlines quoted here did.

Five things to check before you repeat a number

None of this is new; it is basic survey reading. Do it in the report's methods section and chart footnotes, not its headline sentence, because MIT's headline sentence was the problem.

1. What exactly was asked? "Neither higher revenues nor lower costs" is not "no financial benefit". PwC's 56% includes the 22% of CEOs who said AI raised their costs.1616PwC, 29th Global CEO Survey, 2026, p. 6 -- "26% of CEOs say costs have decreased due to AI, while 22% report an increase." https://www.pwc.com/gx/en/ceo-survey/2026/pwc-ceo-survey-2026.pdf (T3 -- industry survey report) Higher costs and no change are different findings sharing one number.
2. How could people answer? Picking one, picking up to two, and ranking everything produce different numbers. Where people could pick several answers, the column adds to more than 100%, so no single figure is a share of the whole. EY's 50% is a top-three share.66EY-Parthenon CEO Outlook, October 2026, chart "AI delivers the most value when paired with other levers of transformation", note -- "Percentages represent the top three rankings combined." https://www.ey.com/en_gl/ceo/ceo-outlook-global-report (T3 -- industry survey report)
3. Who was asked? EY asked its question about AI in deal-making only of the roughly 91% already piloting or using AI there.1717EY-Parthenon CEO Outlook, October 2026, chart "AI's transaction advantage begins with information, not execution", note -- "The respondents whose organizations are piloting or using AI-powered technology for transaction evaluation were asked to answer this question". Piloting 46% + multiple activities 33% + embedded 12% = 91%. https://www.ey.com/en_gl/ceo/ceo-outlook-global-report (T3 -- industry survey report) McKinsey asked its 2025 measures question only of respondents whose organizations regularly use AI.1818McKinsey, "The state of AI in 2025: Agents, innovation, and transformation", Exhibit 6 note -- "Asked only of respondents who said their organizations regularly use AI in at least 1 business function". https://www.mckinsey.com/~/media/mckinsey/business%20functions/quantumblack/our%20insights/the%20state%20of%20ai/november%202025/the-state-of-ai-2025-agents-innovation_cmyk-v1.pdf (T3 -- industry survey report) A share of a subgroup is not a share of everyone.
4. Over what period, and measured how? MIT measured "ROI impact ... 6 months post-pilot", as reported by the people involved.11MIT NANDA, "The GenAI Divide: State of AI in Business 2025", as served by MIT on 18 Aug 2025 -- "95% of organizations are getting zero return"; "Just 5% of integrated AI pilots are extracting millions in value"; "no measurable P&L impact"; "ROI impact measured 6 months post-pilot"; "Preliminary Findings". https://web.archive.org/web/20250818115520/https://nanda.media.mit.edu/ai_report_2025.pdf (T3 -- research report, preliminary) PwC asked about the last 12 months.88PwC, 29th Global CEO Survey, 2026, p. 6 -- "More than half (56%) say their company has seen neither higher revenues nor lower costs from AI, while only one in eight (12%) report both of these positive impacts." Question (p. 7): "In the last 12 months, what impact did AI have on the following at your company?" 4,454 CEOs, fielded 30 Sep - 10 Nov 2025 (p. 23). https://www.pwc.com/gx/en/ceo-survey/2026/pwc-ceo-survey-2026.pdf (T3 -- industry survey report) A number with no period attached has lost one.
5. Do two numbers come from the same question? EY's May 2026 survey found 11% of CEOs link AI impact to financial reporting.1919EY-Parthenon CEO Outlook Survey, May 2026, printed p. 8 -- 11% "AI impact is linked to financial reporting and reviewed regularly by senior management". https://www.ey.com/content/dam/ey-unified-site/ey-com/en-gl/campaigns/ceo/documents/ey-gl-ceo-outlook-survey-05-2026.pdf (T3 -- industry survey report) Its October survey found 16% have real-time visibility into AI costs and returns.1414EY-Parthenon CEO Outlook, October 2026, chart "Most CEOs can estimate AI cost and returns, but only a few have a complete view" -- 56% "We have reliable visibility into most AI costs and can assess ROI with reasonable confidence."; 16% "We have comprehensive, real-time visibility into AI costs and can assess ROI with great confidence." https://www.ey.com/en_gl/ceo/ceo-outlook-global-report (T3 -- industry survey report) Those are different questions, so a trend drawn between them would be invented. McKinsey says it changed how it measured impact in 2026, so its "about the same share as last year" compares two methods.1111McKinsey, "The state of AI in 2026", Exhibit 6 note -- "The 2026 survey measured impact at the business-function level rather than aggregating from individual use cases within each function." https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai (T3 -- industry survey report)

If a report does not answer these questions, that is the finding. EY publishes no margin of error, weighting, response rate or regional sample size;2020EY-Parthenon CEO Outlook, October 2026, "About CEO Outlook" -- an anonymous online survey of 1,200 CEOs, fielded August-September 2026; no margin of error, weighting, response rate or regional sample size is stated. https://www.ey.com/en_gl/ceo/ceo-outlook-global-report (T3 -- industry survey report) we found no derivation of MIT's 95% in its report.11MIT NANDA, "The GenAI Divide: State of AI in Business 2025", as served by MIT on 18 Aug 2025 -- "95% of organizations are getting zero return"; "Just 5% of integrated AI pilots are extracting millions in value"; "no measurable P&L impact"; "ROI impact measured 6 months post-pilot"; "Preliminary Findings". https://web.archive.org/web/20250818115520/https://nanda.media.mit.edu/ai_report_2025.pdf (T3 -- research report, preliminary) Such a number can be described, but not repeated as fact.

What the four surveys say about AI's payoff, details included

• EY (October 2026, 1,200 CEOs): 16% have "comprehensive, real-time visibility" into AI costs and can assess returns "with great confidence".1414EY-Parthenon CEO Outlook, October 2026, chart "Most CEOs can estimate AI cost and returns, but only a few have a complete view" -- 56% "We have reliable visibility into most AI costs and can assess ROI with reasonable confidence."; 16% "We have comprehensive, real-time visibility into AI costs and can assess ROI with great confidence." https://www.ey.com/en_gl/ceo/ceo-outlook-global-report (T3 -- industry survey report)
• PwC (fielded September to November 2025, 4,454 CEOs): 56% saw neither higher revenues nor lower costs from AI in the previous 12 months; 12% saw both.88PwC, 29th Global CEO Survey, 2026, p. 6 -- "More than half (56%) say their company has seen neither higher revenues nor lower costs from AI, while only one in eight (12%) report both of these positive impacts." Question (p. 7): "In the last 12 months, what impact did AI have on the following at your company?" 4,454 CEOs, fielded 30 Sep - 10 Nov 2025 (p. 23). https://www.pwc.com/gx/en/ceo-survey/2026/pwc-ceo-survey-2026.pdf (T3 -- industry survey report)
• McKinsey (fielded May to June 2026, 1,719 respondents): 37% attribute at least some EBIT impact to AI; about 6% attribute at least 5% of EBIT to it and call the impact "significant".1010McKinsey, "The state of AI in 2026: On the road to ROI" -- "Thirty-seven percent of respondents attribute at least some EBIT impact to AI use (about the same share as last year). And the proportion of AI high performers (those who attribute at least 5 percent of EBIT to their use of AI and describe the technology's impact as "significant") has remained flat at about 6 percent of all respondents." 1,719 participants, fielded 4 May - 8 Jun 2026. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai (T3 -- industry survey report)
• MIT NANDA (January to June 2025, 52 organizations interviewed, 153 leaders surveyed): 5% of task-specific enterprise AI tools reached production with a reported productivity or P&L impact.11MIT NANDA, "The GenAI Divide: State of AI in Business 2025", as served by MIT on 18 Aug 2025 -- "95% of organizations are getting zero return"; "Just 5% of integrated AI pilots are extracting millions in value"; "no measurable P&L impact"; "ROI impact measured 6 months post-pilot"; "Preliminary Findings". https://web.archive.org/web/20250818115520/https://nanda.media.mit.edu/ai_report_2025.pdf (T3 -- research report, preliminary)

They measure different things, so they cannot be added up. Read in full, they agree on one point: by each survey's own measure, only a minority can measure or report a financial return from AI.

Appendix: method and limits

EY. We downloaded the May 2026 report PDF and read its tables. The October 2026 edition exists only as a web page; its charts are Flourish embeds, which carry their plotted values, titles and footnotes. We captured 15 embeds (12 of survey data, 280 cells by region; one of monthly M&A data; two containers) and rebuilt every table from the captures. All 20 figures quoted in the report's chart descriptions match the extracted values; single-choice questions total 95-100% per region. A newsletter paired the "one in ten" figure attributed to EY's Dan Diasio with the May survey on 1 October 2026; the figure rests on secondary accounts of a 24 September conference session reported by The Information, whose article is paywalled.2121The Information, "Businesses Still Don't See AI Returns, Consulting Exec Says", 24 Sep 2026 (paywalled; public teaser read). https://www.theinformation.com/newsletters/applied-ai/businesses-still-see-ai-returns-consulting-exec-says (T4 -- news coverage)

MIT NANDA. MIT's URL for the report now redirects. We read the copy MIT served at nanda.media.mit.edu/ai_report_2025.pdf, as archived by the Wayback Machine on 18 August 2025 (26 pages, cover-dated July 2025, sha256 dd5c5f53...8b8a55b68f). Two other copies we checked, the widely linked mlq.ai download and a third-party mirror, are byte-identical to it.

PwC. We read the report PDF and press release. We read some Forbes pieces from search excerpts, not full text.

McKinsey. We read the live 2026 report page. We did not reach the exact EBIT question wording or learn whether the 37% is a share of all respondents or of AI users.

Selection. We chose these four statistics because they are prominent; we did not sample. The cases show that details can go missing and where; they say nothing about how often.

Dataset

The EY October 2026 survey as published, extracted from the report's chart data and attributed to EY: questions.csv (12 questions, exact wording and answer rules), responses.csv (280 cells, by region), mna_timeseries.csv (80 months of the M&A series the report charts), prose_stats.csv (every numeric sentence from the report page and press release) and SOURCE.txt.

Footnotes

1. MIT NANDA, "The GenAI Divide: State of AI in Business 2025", as served by MIT on 18 Aug 2025 -- "95% of organizations are getting zero return"; "Just 5% of integrated AI pilots are extracting millions in value"; "no measurable P&L impact"; "ROI impact measured 6 months post-pilot"; "Preliminary Findings". https://web.archive.org/web/20250818115520/https://nanda.media.mit.edu/ai_report_2025.pdf (T3 -- research report, preliminary) ↩ ↩² ↩³ ↩⁴ ↩⁵ ↩⁶ ↩⁷

2. Axios, 21 Aug 2025 -- "95% of organizations studied get zero return on their AI investment". https://www.axios.com/2025/08/21/ai-wall-street-big-tech (T4 -- news coverage) ↩

3. Harvard Business Review, 29 Aug 2025 -- "95% of investments in gen AI have produced zero returns". https://hbr.org/2025/08/beware-the-ai-experimentation-trap (T4 -- commentary) ↩

4. Forbes (Martin Gutmann), 8 Oct 2025 -- "lacks the kind of rigor required to support a definitive claim". https://www.forbes.com/sites/martingutmann/2025/10/08/why-that-viral-ai-statistic-might-be-more-hype-than-science/ (T4 -- commentary) ↩

5. EY, press release, 1 Oct 2026 -- "half (50%) of CEOs surveyed identifying AI technologies as the single largest contributor to productivity gains"; "more than half (56%) of respondents report at least some reliable visibility into AI costs and returns". https://www.ey.com/en_gl/newsroom/2026/10/ai-drives-productivity-but-ceos-struggle-to-turn-gains-into-grow (T3 -- publisher press release) ↩ ↩² ↩³

6. EY-Parthenon CEO Outlook, October 2026, chart "AI delivers the most value when paired with other levers of transformation", note -- "Percentages represent the top three rankings combined." https://www.ey.com/en_gl/ceo/ceo-outlook-global-report (T3 -- industry survey report) ↩ ↩² ↩³

7. PwC, press release, 19 Jan 2026 -- "56% say they have seen no significant financial benefit to date". https://www.pwc.com/gx/en/news-room/press-releases/2026/pwc-2026-global-ceo-survey.html (T3 -- publisher press release) ↩ ↩²

8. PwC, 29th Global CEO Survey, 2026, p. 6 -- "More than half (56%) say their company has seen neither higher revenues nor lower costs from AI, while only one in eight (12%) report both of these positive impacts." Question (p. 7): "In the last 12 months, what impact did AI have on the following at your company?" 4,454 CEOs, fielded 30 Sep - 10 Nov 2025 (p. 23). https://www.pwc.com/gx/en/ceo-survey/2026/pwc-ceo-survey-2026.pdf (T3 -- industry survey report) ↩ ↩² ↩³ ↩⁴

9. Computerworld, 21 Jan 2026 -- "More than half (56%) of CEOs have not realized either revenue or cost benefits from AI". https://www.computerworld.com/article/4119630/enterprise-ai-investments-are-forging-ahead-despite-elusive-roi-2.html (T4 -- news coverage) ↩

10. McKinsey, "The state of AI in 2026: On the road to ROI" -- "Thirty-seven percent of respondents attribute at least some EBIT impact to AI use (about the same share as last year). And the proportion of AI high performers (those who attribute at least 5 percent of EBIT to their use of AI and describe the technology's impact as "significant") has remained flat at about 6 percent of all respondents." 1,719 participants, fielded 4 May - 8 Jun 2026. https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai (T3 -- industry survey report) ↩ ↩²

11. McKinsey, "The state of AI in 2026", Exhibit 6 note -- "The 2026 survey measured impact at the business-function level rather than aggregating from individual use cases within each function." https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai (T3 -- industry survey report) ↩ ↩² ↩³

12. Forbes (Robert Kramer), 17 Jul 2026 -- "nearly nine in ten organizations now use AI in at least one function, yet most report no significant effect on enterprise-wide EBIT". https://www.forbes.com/sites/robertkramer/2026/07/17/why-ai-requires-a-new-enterprise-operating-model/ (T4 -- commentary) ↩

13. The Register, 25 Aug 2026 -- "37 percent of respondents 'attribute at least some EBIT [earnings before interest and taxes] impact to AI use'". https://www.theregister.com/ai-and-ml/2026/08/25/mckinsey-says-enterprise-ai-is-finally-on-the-road-to-roi/5292388 (T4 -- news coverage) ↩

14. EY-Parthenon CEO Outlook, October 2026, chart "Most CEOs can estimate AI cost and returns, but only a few have a complete view" -- 56% "We have reliable visibility into most AI costs and can assess ROI with reasonable confidence."; 16% "We have comprehensive, real-time visibility into AI costs and can assess ROI with great confidence." https://www.ey.com/en_gl/ceo/ceo-outlook-global-report (T3 -- industry survey report) ↩ ↩² ↩³

15. Tech Times, 26 Aug 2026 -- headline "Record AI Spending Can't Move Earnings Needle for 94% of Enterprises, McKinsey Finds". https://www.techtimes.com/articles/325590/20260826/record-ai-spending-cant-move-earnings-needle-94-enterprises-mckinsey-finds.htm (T4 -- news coverage) ↩

16. PwC, 29th Global CEO Survey, 2026, p. 6 -- "26% of CEOs say costs have decreased due to AI, while 22% report an increase." https://www.pwc.com/gx/en/ceo-survey/2026/pwc-ceo-survey-2026.pdf (T3 -- industry survey report) ↩

17. EY-Parthenon CEO Outlook, October 2026, chart "AI's transaction advantage begins with information, not execution", note -- "The respondents whose organizations are piloting or using AI-powered technology for transaction evaluation were asked to answer this question". Piloting 46% + multiple activities 33% + embedded 12% = 91%. https://www.ey.com/en_gl/ceo/ceo-outlook-global-report (T3 -- industry survey report) ↩

18. McKinsey, "The state of AI in 2025: Agents, innovation, and transformation", Exhibit 6 note -- "Asked only of respondents who said their organizations regularly use AI in at least 1 business function". https://www.mckinsey.com/~/media/mckinsey/business%20functions/quantumblack/our%20insights/the%20state%20of%20ai/november%202025/the-state-of-ai-2025-agents-innovation_cmyk-v1.pdf (T3 -- industry survey report) ↩

19. EY-Parthenon CEO Outlook Survey, May 2026, printed p. 8 -- 11% "AI impact is linked to financial reporting and reviewed regularly by senior management". https://www.ey.com/content/dam/ey-unified-site/ey-com/en-gl/campaigns/ceo/documents/ey-gl-ceo-outlook-survey-05-2026.pdf (T3 -- industry survey report) ↩

20. EY-Parthenon CEO Outlook, October 2026, "About CEO Outlook" -- an anonymous online survey of 1,200 CEOs, fielded August-September 2026; no margin of error, weighting, response rate or regional sample size is stated. https://www.ey.com/en_gl/ceo/ceo-outlook-global-report (T3 -- industry survey report) ↩

21. The Information, "Businesses Still Don't See AI Returns, Consulting Exec Says", 24 Sep 2026 (paywalled; public teaser read). https://www.theinformation.com/newsletters/applied-ai/businesses-still-see-ai-returns-consulting-exec-says (T4 -- news coverage) ↩