Enterprise AI Questions Report. Edition 1.

AI is bought
as a business result

Our guides came up 10 million+ times as an answer to someone's question. Here is what they asked, and what it says about how AI is bought.

MEANWHILE, SOMEONE ASKED

our board is asking about ai decision authority structures now

The question names no model or vendor; it asks about a board's decisions and the structures behind them.

AI news is loud. Labs spend tens of billions a year,N1 and valuations sit at heights last seen in the dot-com years.N12 Each release is hailed as a breakthrough.N7

Amid this noise, our guides came up 10 million+ times in sixteen months as the answer to someone's question. Most of those questions came on a working day77.7% on a weekday, with a job to do.1

Their questions are about running a business. To them, AI is a way to solve a business problem, with a result they can measure. What they put next to AI, and in what order, matters to anyone who sells, builds or adopts it.

Try your own question at the end.

CHAPTER 01

The news is about models. The questions are not.

People ask how AI fits their business, not which AI to buy.

People use "AI" as an umbrella for the whole field, pointed at a problem.

WHAT WE ARE SOLDThe next model changes everything.S1
WHAT PEOPLE ASKHow can we integrate AI into our existing business processes?

News coverage tracks labs, models and the next release. People who ask seldom set one model against another.

Three general names make up most of that gap: agentic AI, artificial intelligence and generative AI.3 In a question, "AI" means the field. It is seldom a product on a shelf.

FIGURE 1

Around the word AI sit mostly business words: how the enterprise is organised, adoption, data and value.

Each dot is an idea asked beside AI. A bigger dot is asked across more of our guides. Colours: tool, practice, payback, role, other. Select a dot to see what it is asked with, and to open our guide.

Most put AI next to business words, led by how an enterprise is organised. A few weigh it against older tools, such as robotic process automation.

Questions that name AI and one other idea. Shares, not counts. Note 4 Note 2

Vendors pitch models and agents, and they sell them on benchmarks.

People do not shop for a model. They bring AI to a business problem.

CHAPTER 02

AI sits next to business words

The language around AI belongs in a board meeting rather than a lab.

How the enterprise is organised, adoption, data and value sit closest to AI.

WHAT WE ARE SOLDPick the smartest model and the rest follows.S2
WHAT PEOPLE ASKHow do CFOs communicate AI investment value to the board?

Look at what sits next to AI: how the enterprise is organised, adoption, digital transformation and data quality.4 Guardrails, risk management and getting real value from AI are there too.

It works the other way too. Start from a business idea, such as operational efficiency. AI turns up beside it more often than chance would give.5

ASKED

generative ai impact on performance management enterprise

ASKED

how does gen ai impact ethics and compliance?

In these questions, the business idea is the subject and AI is what might change it. On the full map of AI questions, governance and adoption are the big regions. The enterprise and its results sit close by.6

FIGURE 2

The map of what people ask about AI, as it grew.

Each dot is an idea. A bigger dot takes a bigger share of the last four weeks of questions. A line joins two ideas asked together. Colours: tool, practice, payback, role, other. Hollow rings are AI's general names. Rings mark the week's biggest risers. A white centre means the dot opens our guide. Press Play to watch the weeks.

Guides on this map

    The questions, week by week from March 2026. Note 6

    CHAPTER 03

    Governance comes first

    Nearly half of the AI questions are about governance.

    Governance leads in every month we can measure. The questions sound like people with a plan.

    WHAT WE ARE SOLDGovernance slows you down.S3
    WHAT PEOPLE ASKWhat AI governance controls should a CFO have in place before a board audit?

    Governance covers the rules and checks an organisation puts around AI. It is the largest subject in the AI questions we collected.7 Models themselves get about one in seven.

    Governance leads in all six months.8

    FIGURE 3

    Governance leads, every month.

    One square is one per cent of AI questions. Colours: governance, change, models, agents.

    All six months

    AI questions, March to August 2026. Shares, not counts. Note 7 Note 8

    There is little fear in the wording. These are people who plan to go ahead safely.

    ASKED

    which enterprise ai tools provide auditable outputs and full governance controls for regulated industries?

    ASKED

    what ai governance controls should a cfo have in place before a board audit?

    ASKED

    eu ai act high-risk systems requirements conformity assessments transparency human oversight

    A finance chief, a board audit, a regulator: governance is where AI meets the people who answer for the business. Boards are paying attention too. The share with no AI on the agenda fell from 45% to 31% in a year.9

    CHAPTER 04

    The questions run in one order

    AI changes the work, and the changed work pays back.

    Nine in ten questions put AI first, and almost every one puts the work before the result.

    WHAT WE ARE SOLDAdapt your organisation to the platform.S4
    WHAT PEOPLE ASKHow do business transformation leaders measure impact of AI on operational efficiency?

    Many questions carry a small story of cause and effect. A tool changes how the organisation works.

    ASKED

    how ai-powered workflow tools improve transparency in cross-functional document processes

    People ask what AI will change for them. They do not ask what they must change to suit AI. The next link is firmer still: when a question joins the work to a result, the work comes first.11

    Together, these links describe a sequence:

    AIa business problem solvedan impact you can measure
    FIGURE 4

    AI changes the work, and the work pays back.

    Three columns: tool, practice, payback. A wider line is asked more often. Select an idea to light its paths, or a line to read its questions.

    AI questions. Each example names no person or organisation. Note 11 Note 10

    Our guides reach the same order from the other direction.

    CHAPTER 05

    The pitch skips the middle step

    Most people put a change in the work between the tool and the result.

    The pitch goes straight from tool to result; most questions take a longer route.

    WHAT WE ARE SOLDBuy the tool, and the results follow.S5
    WHAT PEOPLE ASKWhy does enterprise AI deployment fail to deliver ROI?

    The order we are sold is shorter. Vendors often tell it with their own job cuts, and the change in the work drops out.19

    Some questions do follow the pitch. They go straight from buying a tool to getting a result.

    ASKED

    what ai agents are recommended for improving productivity in the workplace?

    ASKED

    how do leading enterprise ai platforms compare on measurable business outcomes?

    About one in three tool-first questions takes this short path; most put a change in the work between the two.12

    FIGURE 5

    What we are sold, and what people ask.

    The dashed grey arc is what we are sold: a jump from the tool to the payback. The solid line is what people ask: from tool through a change in the work to payback.

    AI questions. Shares of the paths that start from a tool or from a change. Note 12

    Our research finds that the longer path pays. A pilot that dazzles in a demo stalls when nothing around it changes.

    When a programme stalls, switching vendors is an easy move that seldom helps: the platform was seldom the problem.13

    Gains can come before a full redesign. A small, well-chosen quick win can pay first and earn trust. The larger results come when the work changes.

    CHAPTER 06

    The middle step, in their own words

    The middle step is the work, the people and the skills.

    People ask how AI fits the work, how staff will feel, and what skills they need.

    WHAT WE ARE SOLDGive everyone access to the chat tool.S6
    WHAT PEOPLE ASKIs the AI skills gap really a work design problem?

    Questions about this change keep coming back to how the work fits together. They also ask how people feel about the change and what skills the new work requires.

    The work.

    ASKED

    how can organizations integrate enterprise generative ai into enterprise workflows and operating models?

    ASKED

    what skills and organizational changes are required to operationalize ai for my business?

    Outside research agrees. Among the factors tested, workflow redesign has the strongest link to profit from AI.14

    The people.

    ASKED

    how are organizations managing employee trust and fear of displacement during ai transformation?

    ASKED

    how can leaders replace employee fear with excitement about ai?

    These leaders know that a tool nobody trusts goes unused. Our guide to the AI transformation leader says the leader reshapes how people and AI work together.

    The skills.

    Here the questions and our research agree on a surprise. Skills are not mainly a training problem.

    Required skills become clear once the organisation defines the changed work.

    CHAPTER 07

    The result they want back

    People want a result they can prove, and money is now the first measure.

    They name outcomes wider than money, and they want to prove each one.

    WHAT WE ARE SOLDHours saved per worker, per week.S7
    WHAT PEOPLE ASKHow are executives measuring AI success beyond productivity metrics?

    Questions about results reach past cost. The most common are business outcomes, transparency and operational efficiency.15 Competitive advantage comes next, and people want proof.

    ASKED

    how are enterprises measuring the business value of ai-driven digital transformation?

    ASKED

    how are executives measuring ai success beyond productivity metrics?

    ASKED

    how do you measure the business impact of ai?

    "Beyond productivity" is the phrase to notice. Time saved is easy to count, yet the people asking know it is not the whole story. The market is moving the same way: direct financial impact now leads as the measure of AI return.16

    A measure must exist before the tool goes live, or no one can say what changed.

    CHAPTER 08

    We have seen this before

    E-commerce changed how the store worked without replacing it.

    The noise of the dot-com years faded; the working questions stayed.

    In 2000, the dot-com index hit its peak. E-commerce was about 1% of US retail.17 After the crash, stores kept asking how to sell online, one working question at a time.

    By late 2020, e-commerce was 15% of retail, and the stores that did well used it to extend the store.18

    FIGURE 6

    The noise peaked and fell. The working share kept rising.

    Top: the Nasdaq Composite, a dotted line of year-end closes, then month-end closes in 2026, to 29 September 2026. Bottom: online sales as a share of US retail, one reading per quarter from the end of 1999; a thin dashed line marks the years before the Census measured it. The readings are joined by straight lines. Both panels share one time axis.

    Index: Nasdaq Composite (^IXIC), closing values from Yahoo Finance, checked against Nasdaq.com: year-end closes 1994 to 2025, month-end closes January to August 2026, and the close of 29 September 2026. US Census Bureau, Quarterly Retail E-Commerce Sales, e-commerce share of total retail, seasonally adjusted, Q4 1999 to Q2 2026 (revised 18 August 2026). The Census changed the series in April 2025. Note 17

    E-commerce took hold through those working questions, whatever the index did. AI looks the same from here. Under the noise, the calm questions about the work go on.

    What this means

    The market talks about AI as technology. The people asking treat it as a business result. Their questions start with AI, move to a problem solved and end with a result they can measure.

    If you sell AI

    • Give the problem and the result priority over the model.
    • Answer the governance question before the feature question.
    • Show how the work changes and who changes it.

    If you build AI

    • Build for the middle step: audit trails, oversight and fit with the work.
    • Make the business measure as visible as the model score.
    • Use a benchmark for initial screening; the decision requires more.

    If you adopt AI

    • Plan the change in the work before the tool arrives.
    • Set your baseline before you go live.
    • Bring your people in early.

    What people are asking

    Questions our research found people asking.

    ASKED

    Can't we just use standalone AI tools across the business?

    ASKED

    How do you measure the business impact of AI?

    ASKED

    How do CFOs communicate AI investment value to the board?

    ASKED

    Why can't enterprises use standard IT ROI methods for AI?

    ASKED

    Is the AI skills gap really a work design problem?

    ASKED

    Why are leaders, not employees, the barrier to scaling AI?

    ASKED

    How do you present AI ROI to a CFO?

    ASKED

    Do I have to replace RPA to adopt agentic AI?

    ASKED

    How do you avoid vanity metrics in a GenAI measurement program?

    ASKED

    How do you connect ML model metrics to business outcomes?

    ASKED

    What is the cost of bad data for AI?

    ASKED

    How do you evaluate an AI agent before buying?

    You need not win the noise. People with a job to do are already typing the questions that shape this conversation.

    TRY IT

    Try your own question

    Words that are ideas on our map get a wavy underline: tool, practice, payback, role.

    As you type, ideas from our map are offered. Use the arrow keys and Enter to pick one.As you type, ideas from our map are offered. Tap one to pick it.

    Or build one the way people ask it

    This box runs in your browser only. Nothing you type is sent or stored.

    The questions people asked, March to August 2026. Note 10 Note 11

    Notes

    The noise, sourced

    Every line the band at the top can show, with its source. Each is a verbatim fragment; "..." marks a cut. We checked each one at its source on 29 September 2026.

    1. approximately $85 billion — Alphabet, SEC 8-K, 2025-07-23
    2. $115-135 billion — Meta, SEC 8-K, 2026-01-28
    3. $125B in 2025 — Amazon CFO, via CNBC, 2025-10-30
    4. $37.5 billion — Microsoft, one quarter, via CNBC, 2026-01-28
    5. $40 billion at a $300 billion post-money valuation — OpenAI, 2025-03-31
    6. $13 billion ... $183 billion post-money — Anthropic, 2025-09-02
    7. nonstop one-upmanship — MIT Technology Review, 2025-12-15
    8. Peak of Inflated Expectations — Gartner Hype Cycle, 2025-08-05
    9. 50% ... more concerned than excited — Pew Research Center, 2025-09-17
    10. only 46% of people globally are willing to trust AI systems — KPMG and University of Melbourne, 2025-04-28
    11. highest level since the dot com bubble — Bank of England, 2025-10-24
    12. It was the internet then, it is AI now. — Pierre-Olivier Gourinchas, IMF, via Fortune, 2025-10-14
    13. inflated promises of AI vendors — Forrester, Predictions 2026, 2025-10
    14. we expect that this will reduce our total corporate workforce — Andy Jassy, Amazon CEO, 2025-06-17
    15. I need less heads — Marc Benioff, Salesforce, via CNBC, 2025-09-02
    16. elements of irrationality — Sundar Pichai, Alphabet, via BBC, 2025-11-18
    17. I think no company is going to be immune, including us — Sundar Pichai, Alphabet, via BBC, 2025-11-18
    18. expose the equivalent of 300 million full-time jobs to automation — Goldman Sachs, 2023-04-05
    19. AI’s $600B Question — David Cahn, Sequoia Capital, 2024-06-20
    20. when developers use AI tools, they take 19% longer — METR, 2025-07-10
    21. We are now confident we know how to build AGI — Sam Altman, OpenAI, 2025-01
    22. Developing superintelligence is now in sight. — Mark Zuckerberg, Meta, 2025-07-30
    23. Gen AI: too much spend, too little benefit? — Goldman Sachs, Top of Mind, 2024-06-27
    24. this spending has little to show for it so far — Goldman Sachs, Top of Mind, 2024-06-27
    25. The year the Frontier Firm is born — Microsoft Work Trend Index, 2025-04-23
    26. 170 million new jobs by 2030, while displacing 92 million others — World Economic Forum, 2025-01-08
    27. Are we in a phase where investors as a whole are overexcited about AI? — Sam Altman, OpenAI, via The Verge, 2025-08-15
    28. When bubbles happen, smart people get overexcited about a kernel of truth — Sam Altman, OpenAI, via The Verge, 2025-08-15
    29. only 25% of AI initiatives have delivered expected ROI — IBM CEO Study, 2025-05-06
    30. AI-Generated “Workslop” Is Destroying Productivity — Harvard Business Review, 2025-09-22
    31. gradually stop using contractors to do work that AI can handle — Luis von Ahn, Duolingo, via The Verge, 2025-04-28
    32. We are past the event horizon; the takeoff has started. — Sam Altman, OpenAI, 2025-06-10
    33. country of geniuses in a datacenter — Dario Amodei, Anthropic, 2024-10
    34. more than 80 percent of AI projects fail — RAND, 2024-08-13
    35. more than a quarter of all new code at Google is generated by AI — Sundar Pichai, Google, 2024-10-29
    36. a kind of industrial bubble — Jeff Bezos, via CNBC, 2025-10-03
    37. we will quickly lose even the social permission — Satya Nadella, Microsoft, via Tom’s Hardware, 2026-01-21
    38. going to replace literally half of all white-collar workers in the U.S. — Jim Farley, Ford, via The Autopian, 2025-06-30
    39. Take AI, there’s a lot of money going into it — Jamie Dimon, JPMorgan, via CNN, 2025-10-09
    40. AI is real, AI in total will pay off. — Jamie Dimon, JPMorgan, via CNN, 2025-10-09
    41. totally, totally gone — Sam Altman, OpenAI, via CX Today, 2025-07-30
    42. an AI that can effectively be a sort of midlevel engineer — Mark Zuckerberg, Meta, via Entrepreneur, 2025-01-20
    43. the heads of the top AI companies made promises they couldn’t keep — MIT Technology Review, 2025-12-15
    44. 2025 has been a year of reckoning. — MIT Technology Review, 2025-12-15
    45. invest $500 billion over the next four years building new AI infrastructure — Stargate Project, SoftBank press release, 2025-01-22
    46. create hundreds of thousands of American jobs — Stargate Project, SoftBank press release, 2025-01-22
    47. RPO is likely to exceed half-a-trillion dollars — Safra Catz, Oracle, 2025-09-09
    48. Q1 Remaining Performance Obligations $455 billion, up 359% — Oracle, press release, 2025-09-09
    49. Blackwell sales are off the charts, and cloud GPUs are sold out. — Jensen Huang, NVIDIA, 2025-11-19
    50. AI is going everywhere, doing everything, all at once. — Jensen Huang, NVIDIA, 2025-11-19
    51. The AI race is on, and Blackwell is the platform at its center. — Jensen Huang, NVIDIA, 2025-08-27
    52. NVIDIA intends to invest up to $100 billion in OpenAI — NVIDIA and OpenAI, press release, 2025-09-22
    53. Everything starts with compute. — Sam Altman, OpenAI, 2025-09-22
    54. the world's most ambitious AI buildout — Lisa Su, AMD, 2025-10-06
    55. artificial intelligence is the electricity of our age — Brad Smith, Microsoft, 2025-01-03
    56. on track to invest approximately $80 billion to build out AI-enabled datacenters — Brad Smith, Microsoft, 2025-01-03
    57. The rate of innovation and the speed of diffusion is unlike anything we've seen. — Satya Nadella, Microsoft, 2025-07-30
    58. We delivered our first-ever $100 billion quarter. — Sundar Pichai, Alphabet, 2025-10-29
    59. We are now processing over 1.3 quadrillion monthly tokens — Sundar Pichai, Alphabet, 2025-10-29
    60. Meta's vision is to bring personal superintelligence to everyone. — Mark Zuckerberg, Meta, 2025-07-30
    61. AI is a once-in-a-lifetime reinvention of everything we know — Andy Jassy, Amazon shareholder letter, 2025-04-10
    62. agentic enterprises, where humans and AI agents work side by side — Marc Benioff, Salesforce, 2025-09-03
    63. The risk of a sharp market correction has increased. — Bank of England FPC, via The Guardian, 2025-10-08
    64. it's going to be 10 times bigger than the Industrial Revolution, and maybe 10 times faster — Demis Hassabis, Google DeepMind, via The Guardian, 2025-08-04
    65. These models are being hyped up, and we're investing more than we should — Daron Acemoglu, MIT, via NPR, 2025-11-23
    66. True end demand is ridiculously small. — Michael Burry, via NPR, 2025-11-23
    67. OpenAI is very likely going to be the world's next multitrillion-dollar hyperscale company — Jensen Huang, Nvidia, via Fortune, 2025-09-29
    68. the AI boom that is now in the early stages of a bubble — Ray Dalio, Bridgewater, via Fortune, 2026-01-06
    69. Circular investment deals by major AI companies spark 'bubble' fears — Semafor, 2025-10-08
    70. AI is starting to get better than humans at almost all intellectual tasks — Dario Amodei, Anthropic, via CNN, 2025-05-30
    71. the biggest and most dangerous bubble the world has ever seen — Julien Garran, MacroStrategy Partnership, via CNN, 2025-10-18
    72. Oracle has become the poster child for fears of an AI bubble — Cory Johnson, analyst, via Yahoo Finance, 2025-12-26
    73. AI machines—in quite a literal sense—appear to be saving the US economy right now — George Saravelos, Deutsche Bank, via Al Jazeera, 2025-10-09
    74. The moment investors start demanding cash-flow returns ... some of these flywheels could seize — Matthew Tuttle, Tuttle Capital Management, via Al Jazeera, 2025-10-29
    75. AI Is the Bubble to Burst Them All — Brian Merchant, Wired, 2025-10-27
    76. This is a fifty-billion-dollar market, not a trillion-dollar market — Ed Zitron, via The New Yorker, 2025-08-12
    77. Brad, if you want to sell your shares, I'll find you a buyer — Sam Altman, OpenAI, via TechCrunch, 2025-11-02
    78. If we end up misspending a couple of hundred billion dollars — Mark Zuckerberg, Meta, via Business Insider, 2025-09-19
    79. AI Bubble Today Is Bigger Than the IT Bubble in the 1990s — Torsten Slok, Apollo, 2025-07-16
    80. The AI cycle is building at close to twice the pace of the housing boom — Torsten Slok, Apollo, 2026-08-06
    81. Two trillion dollars in annual revenue is what’s needed to fund computing power — Bain & Company, 2025-09-23
    82. broad adoption, real revenue, and productivity gains at scale, signaling a boom versus a bubble — Menlo Ventures, 2025-12-09
    83. the downside risk of current AI exuberance — Bank for International Settlements, 2026-06-28
    84. a genuine technological breakthrough that attracted capital in excess of what commercial returns could ultimately justify — Bank for International Settlements, 2026-06-28
    85. $3 trillion by the end of 2028 on just the data centers — Andrew Sheets, Morgan Stanley, 2025-07-25
    86. the companies that are so highly valued—actually have earnings and stuff like that — Jerome Powell, Federal Reserve, 2025-10-29
    87. AI infrastructure spending over the next five years could exceed five trillion US dollars — Bank of England FPC, 2025-12-02
    88. $1 trillion of AI-related investment around the globe in 2026 — Joseph Briggs, Goldman Sachs, 2026-08-07
    89. AI scaling is likely to continue through 2030, despite requiring unprecedented infrastructure — David Owen, Epoch AI, 2025-09-16
    90. 2026 will be the "Year of Delays" for data centers and AGI — David Cahn, Sequoia Capital, 2025-12-03
    91. AI linked to a fourfold increase in productivity growth and 56% wage premium — PwC, via PR Newswire, 2025-06-03

    Sold examples

    Notes

    1. When people ask

      77.7% of AI question activity falls on a weekday. ↩

      Where this comes from

      day_device.weekday

    2. How often the other idea is an AI tool

      13.3% of paired questions name a second AI tool; chance would give 26.3%. ↩

      on the .

      Where this comes from

      headline.F124

    3. Which names make the gap

      Agentic AI, artificial intelligence and generative AI make up 81.5% of the gap. ↩

      on the .

      Where this comes from

      headline.F124.top3_of_top10_gap

    4. What sits next to AI

      The ideas most often asked beside the general AI names. ↩

      Where this comes from

      umbrella.partners

    5. From business idea to AI

      Business ideas pair with AI 43% of the time; chance would give 39.9%. ↩

      on the .

      Where this comes from

      headline.F122

    6. The map of AI questions

      AI and mixed ideas, grown week by week over 27 weeks from 2026-03-02. ↩

      Where this comes from

      map.nodes, node.cum

    7. What the AI questions are about

      Each AI question sorted into one subject; governance is the largest. ↩

      Where this comes from

      topics.topics[], topics.strings

    8. Governance, month by month

      March to August 2026: governance takes between 39% and 55%, the largest subject each month. ↩

      Where this comes from

      explore/r96_time/r96_topics_by_period.json by_month

    9. Deloitte

      Deloitte, Governance of AI: a board imperative, 2nd edition, 2025-05-30. ↩

    10. Which way the change runs

      AI does the changing in 92.5% of questions that link it to a way of working; chance would give 50%. ↩

      on the .

      Where this comes from

      headline.F125

    11. Work before result

      The work comes before the result in 97.1% of such questions; chance would give 50%. ↩

      on the . On AI questions alone: 96.6%.

      Where this comes from

      headline.F126, F129

    12. The short path

      About one in three tool-first questions goes straight from tool to result. ↩

      Where this comes from

      flow_ai.family_counts

    13. Our guides

      Skills and upskilling strategy. ↩

    14. McKinsey

      McKinsey, The state of AI, March 2025. ↩

    15. The results people name

      The result ideas most often asked, ranked. ↩

      Where this comes from

      flow.nodes

    16. Futurum Group

      Futurum Group, 2026-02-17: direct financial impact 21.7%, nearly double; productivity down 5.8 points as the lead measure. ↩

    17. US Census Bureau

      US Census Bureau, Quarterly Retail E-Commerce Sales, e-commerce as a share of total retail, seasonally adjusted: 0.6% in Q4 1999 (the first reading), 1.0% in Q4 2000, 4.6% in Q4 2010, 15.0% in Q4 2020, 17.1% in Q2 2026 (preliminary; data revised 18 August 2026). Series break April 2025: from then on, the estimates cover only businesses with paid employees. ↩

    18. HBR

      HBR, study of 46,000 shoppers: 73% used more than one channel. ↩

    19. The job-cut story

      Salesforce cut support from about 9,000 to about 5,000 roles with AI agents (CNBC, 2025-09-02). Klarna said its AI assistant did the work of 700 agents (Klarna, 2024-02-27), then hired people again (CX Dive, 2025-05). Amazon linked 14,000 cuts to AI, then called them "not even really AI-driven" (Axios, 2025-10-31). ↩

    Appendix

    FIGURE A1

    The questions run forward along the chain; few run back.

    Each cell is the share of its row: where the questions from each kind go next. Rows and columns: tool, practice, payback.

    AI questions. Shares of each row. Note 12