Adoption · worldwide · May 2026

What AI is used for

1000 use cases across 27 enterprise domains and 9 of personal use, 869 of them carrying a measured share of worldwide AI usage. The two are measured separately and never added together: business usage through the API, personal usage on Claude.ai. The work is the subject here, not the model: which model does a job changes every few months, and the measurements behind this page record what was asked for, not what answered.

Executive view · May 2026

Business AI activity, compared

See where measured business use is concentrated, how the work is handled, and which domains combine current traction with the largest estimated task-time reduction.

Share of measured first-party API conversations assigned to use cases in this domain.

Measure01Software Engineering02Documents & Knowledge Work03Sales & Marketing04Data & Analytics05Finance06IT Operations07Media & Creative08Productivity & Coordination09Security10Education & Training
API activity share
15.3%
9.6%
8.8%
5.2%
5.0%
4.8%
4.4%
3.5%
2.3%
2.1%
Work intensity
90.1%
92.3%
98.2%
86.2%
93.1%
96.4%
90.5%
85.6%
97.1%
35.9%
Automation
92.6%
95.8%
95.9%
96.8%
95.9%
97.8%
98%
98%
98%
90.9%
Augmentation
7.4%
4.2%
4.1%
3.2%
4.1%
2.2%
2%
2%
2%
9.1%
Task-time reduction
92.9%
87.7%
90.8%
92.1%
91.1%
89.6%
88.7%
92.4%
92%
83.7%
Leading use casePrompt engineeringDocument summarisationLead qualificationScheduled reportingRegulatory compliance checksService desk ticket resolutionTranscript summarisationInbox organisationThreat detectionQuiz assistance

Darker cells are higher relative to the other domains in the same row. Select a measure to reorder the columns.

Method: the ten highest-activity enterprise domains with at least two use cases in the Economic Index's published top-100 API table. Behavioral averages are weighted by each included use case's share of first-party API traffic. Curated domains can overlap, so activity shares do not sum to 100%. These are Claude usage signals, not company adoption, market size, ROI or organizational maturity.

121 countries

Where AI is used

How much of worldwide Claude.ai usage comes from each country

Where AI is used
lessmore
  1. 01United States20.16%
  2. 02India7.12%
  3. 03France3.95%
  4. 04United Kingdom3.47%
  5. 05Brazil3.33%
  6. 06Japan3.28%
  7. 07Republic of Korea3.24%
  8. 08Germany2.98%
  9. 09Australia2.65%
  10. 10Canada2.63%
  11. 11Indonesia2.08%
  12. 12Spain2.05%

Only consumer usage on Claude.ai is reported by country; business traffic through the API isn’t, so this map describes where people use Claude, not where every AI product is used. “Per person” compares a country’s usage with its share of the working-age population, so 1.00× is the world average. Singapore, Bahrain, Malta, Mauritius report usage but are too small to draw at this scale: Singapore 0.62%, Bahrain 0.04%, Malta 0.04%, Mauritius 0.04%.

By domain

Use cases, domain by domain

Enterprise27

Personal9

Selected domain

Software Engineering

Writing the code, reviewing it, and finding out why it broke.

113 of 113 measured · 15.34% of business usage · 11.6% of consumer usage

Use cases in this domain

  1. Prompt engineeringWrite the instruction that gets the result reliably.2.37%0.09%
  2. Feature implementationBuild the change end to end, against the spec.1.58%0.88%
  3. Code reviewRead a change the way the next maintainer will.0.73%0.08%
  4. Agent architectureDesign how an agent plans, calls tools and stops.0.67%0.28%
  5. API debuggingFind out which side of the call is wrong.0.63%0.53%
  6. Algorithm implementationTurn the method into code that holds under load.0.62%0.50%
  7. Backend fixesPatch the service without taking it down.0.57%0.29%
  8. Structured output publishingEmit data in the shape the consumer agreed to.0.49%0.01%
  9. RefactoringChange the shape without changing the behaviour.0.43%0.39%
  10. ScriptingAutomate the thing being done by hand twice a week.0.43%0.05%
  11. Code comprehensionWork out what unfamiliar code actually does.0.38%0.07%
  12. UI componentsBuild the parts a screen is assembled from.0.33%0.29%
  13. StylingGet the layout to hold at every width.0.31%0.26%
  14. Test executionRun the suite and read what it is telling you.0.31%0.03%
  15. Mobile developmentShip to a device with its own rules.0.27%0.51%
  16. Web developmentBuild for the browser, and for the people in it.0.26%0.76%
  17. Implementation planningBreak the work down before starting it.0.24%0.16%
  18. Behaviour verificationProve the code does what it is claimed to do.0.24%0.01%
  19. LLM application developmentBuild the product around the model, not the demo.0.23%0.13%
  20. Test authoringWrite the test that would have caught it.0.23%0.08%
  21. Interface localisationTranslate an interface without breaking its layout.0.22%0.01%
  22. Machine learningTrain, evaluate and decide whether it is good enough.0.21%0.19%
  23. Visual verificationCheck the rendered screen against what was intended.0.19%0.03%
  24. Build pipelinesMake the build repeatable, and fast enough to trust.0.16%0.09%
  25. Repository inspectionSurvey an unfamiliar codebase before touching it.0.16%0.01%
  26. Algorithm correctnessProve the method is right, not merely that it ran.0.15%0.04%
  27. Shell scriptingGlue the pieces together at the command line.0.14%0.13%
  28. Dependency and licence reviewKnow what is in the build and what it obliges.0.13%0.25%
  29. DebuggingNarrow it down until only one explanation is left.0.13%0.07%
  30. API queryingCall someone else's service and handle what comes back.0.12%0.03%
  31. AI coding assistantsUse, evaluate and set up the tools that write code.0.11%0.14%
  32. Version controlBranch, merge and recover when it goes wrong.0.10%0.07%
  33. Pull request lifecycleMove a change from opened to merged.0.10%0.01%
  34. Feature specificationWrite down what finished means before building.0.09%0.21%
  35. Web scrapingCollect from pages that were not built to be collected from.0.09%0.11%
  36. API test validationCheck a contract holds before the caller finds out.0.08%0.00%
  37. Knowledge graphsModel how things relate, not just what they are.0.08%0.00%
  38. API designDesign the interface other teams will have to live with.0.07%0.04%
  39. Module designDecide what belongs together and what does not.0.07%0.04%
  40. Interface copyWrite the words inside the product.0.07%0.04%
  41. Assistant featuresDesign what an in-product assistant should do.0.07%0.04%
  42. Systems programmingWork close to the metal, where mistakes are expensive.0.07%0.03%
  43. Compilers and parsersTurn one language into another, correctly.0.07%0.01%
  44. Project scaffoldingStand a new project up with the decisions already made.0.06%0.19%
  45. Configuration debuggingFind the setting that is quietly wrong.0.06%0.08%
  46. Screenshot analysisRead an interface from a picture of it.0.06%0.01%
  47. Virtual environmentsIsolate dependencies so projects stop fighting.0.06%0.00%
  48. Environment setupGet a machine to the state the project expects.0.05%0.17%
  49. Database faultsWork out whether it is the query, the data or the engine.0.05%0.06%
  50. Application architectureDecide the boundaries before they harden.0.04%0.15%
  51. Configuration authoringWrite config that is readable a year later.0.04%0.04%
  52. ConcurrencyGet parallel work right, where bugs hide best.0.04%0.01%
  53. Type systemsMake the compiler catch what review would miss.0.04%0.01%
  54. Embedded systemsWrite for hardware with no room to be wasteful.0.03%0.17%
  55. Backend frameworksRouting, persistence and the shape of a service.0.03%0.11%
  56. ContainerisationPackage the service so it runs the same everywhere.0.03%0.08%
  57. Game engine toolingBuild the editor and pipeline a team makes a game with.0.03%0.06%
  58. Automation scriptingScript the repetitive thing, once.0.03%0.06%
  59. Filesystem workMove, watch and clean up files safely.0.03%0.04%
  60. Developer platform administrationRun the accounts, keys and access a team builds on.0.03%0.02%
  61. Data access layersStand between the code and the database, carefully.0.03%0.02%
  62. Reinforcement learningTrain against a reward, and check what it really rewards.0.03%0.02%
  63. Conversational applicationsBuild something that holds a conversation and a task.0.03%0.01%
  64. Regular expressionsWrite the pattern, then prove it does not overmatch.0.03%0.01%
  65. Release notesSay what changed in terms the reader cares about.0.03%0.01%
  66. Coverage analysisFind the code no test has ever run.0.03%0.00%
  67. Desktop applicationsBuild for a window rather than a tab.0.02%0.21%
  68. Computer science fundamentalsThe complexity and correctness questions under the code.0.02%0.21%
  69. CMS and pluginsExtend the platform a site is published on.0.02%0.15%
  70. Persistent knowledge systemsGive a system memory it can be trusted with.0.02%0.09%
  71. Desktop application debuggingDebug where there is no browser console to lean on.0.02%0.08%
  72. Computer visionGet a machine to read an image reliably.0.02%0.06%
  73. Enterprise architectureFit a new system into an estate that already exists.0.02%0.05%
  74. Runtime environmentsKnow what the code will find when it runs.0.02%0.04%
  75. Desktop automationDrive the applications that have no API.0.02%0.03%
  76. Low-level scriptingWork with binaries and bytes when nothing else will do.0.02%0.02%
  77. Distributed systemsReason about failure when the parts are far apart.0.02%0.02%
  78. Reverse engineeringWork out how something behaves without its source.0.02%0.02%
  79. Software processHow a team decides, reviews and ships.0.02%0.02%
  80. Structured content authoringAuthor content in a schema a product can render.0.02%—
  81. Browser APIsUse what the platform gives you before adding a library.0.01%0.04%
  82. Multiplayer and anti-cheatKeep a shared system fair under adversarial load.0.01%0.04%
  83. Editor and tooling setupConfigure the environment people work in all day.0.01%0.03%
  84. Object-oriented designModel the domain in the shapes the language offers.0.01%0.03%
  85. Repository setupLay out a project so contributions are easy.0.01%0.03%
  86. Product designDecide what the software should do before how.0.01%0.03%
  87. Community toolingBuild what a player community runs itself on.0.01%0.03%
  88. DevOps toolingWire the tools a delivery pipeline is assembled from.0.01%0.02%
  89. Embedded hardwareWhere the firmware meets the board.0.01%0.02%
  90. Data structuresChoose the structure the access pattern wants.0.01%0.01%
  91. Graph algorithmsModel the problem as a graph, then solve it.0.01%0.01%
  92. Memory managementKnow what is allocated and when it is freed.0.01%0.01%
  93. OptimisationMeasure first, then make it faster.0.01%0.01%
  94. Numerical methodsCompute with floats without being lied to.0.01%0.01%
  95. Offline-first applicationsWork when the connection does not.0.01%0.01%
  96. App store publishingGet through review and out to users.0.01%0.01%
  97. Network protocolsRead the wire when the library is not enough.0.01%0.00%
  98. Computer architectureReason about what the hardware will actually do.0.00%0.06%
  99. IoT and sensorsRead the physical world into a system.0.00%0.03%
  100. Automata and formal methodsProve a machine can only do what it should.0.00%0.02%
  101. Control flowGet the branching right before optimising it.0.00%0.01%
  102. JavaScriptThe language the browser gives you, used well.0.00%0.01%
  103. Open-source ecosystemsDepend on, contribute to and license other people's code.0.00%0.01%
  104. App prototypingGet something running before the design is settled.not measured0.85%
  105. AI product developmentDecide what the model should and should not be asked.not measured0.28%
  106. UI behaviourMake the interface respond the way people expect.not measured0.24%
  107. Model comparisonChoose between models on the job in hand.not measured0.14%
  108. Frontend frameworksComponent work in the framework the team chose.not measured0.13%
  109. Template replicationReproduce a known-good structure across projects.not measured0.05%
  110. Simulation faultsDebug systems whose output is meant to be uncertain.not measured0.04%
  111. Regression fixingRepair what used to work, without breaking the fix.not measured0.03%
  112. Vector databasesStore and search by meaning rather than by key.not measured0.03%
  113. Natural language processingTokenise, embed and evaluate text properly.not measured0.01%

Shares are of all measured usage in May 2026: business usage through the Anthropic API, consumer usage on Claude.ai. A use case with no matching topic in the index is listed without a number rather than estimated.

Where this is a bigger share of AI use

17 countries

Of everything each country was measured using AI for, how much of it was this.

  1. 01Benin4.03%
  2. 02Cambodia3.24%
  3. 03Thailand2.81%
  4. 04Rwanda2.59%
  5. 05Togo2.33%
  6. 06Angola2.22%
  7. 07Kenya2.04%
  8. 08Republic of the Congo2.03%
  9. 09Albania1.92%
  10. 10Ghana1.81%
  11. 11Tanzania1.81%
  12. 12Brazil1.58%

So a country ranks high here when this work is a large part of what it uses AI for — not when it does the most of it. A smaller market whose AI use is concentrated on a few things will outrank a larger one that spreads across many, which is why the top of this list is rarely the countries you would name first. The index publishes only a handful of top use cases per country, so this ranks where the work was prominent enough to reach that list, not everywhere it happens. Consumer usage on Claude.ai only: API traffic isn’t broken down by country.

United Arab Emirates

Zero Government Bureaucracy

The programme asks federal entities to cancel at least 2,000 procedures, halve the time the rest take, and remove the steps that need not exist at all. Those are targets counted in procedures and minutes — and minutes are measured. Below are 15 kinds of work a government procedure is actually made of, with how long each takes a person and how long the same work takes with AI.

u.ae
The step stops existingThe resident is never asked for what the government can already read, or already holds.
The step takes minutesSomeone still has to decide — but not after hours of reading first.
The step happens in their languageA service a resident cannot read is a step they cannot complete.
Document generation7h 20m95%
Document summarization2h 20m92%
Forms and documents7h 53m92%
Regulatory compliance7h 50m92%
Data record lookup31m91%
Document analysis3h 48m91%
Translation17m91%
Document review5h 8m90%
Named entity recognition20m89%
Transcript summarization2h 27m89%
Legal documents40m89%
Structured data extraction29m88%
Invoice extraction19m86%
Document extraction17m85%
Document metadata14m84%

The programme asks for half. Every one of these cuts more.

Which tasks belong to a procedure is our selection, not a measurement: the index records what people worldwide ask AI to do and has no notion of a government procedure. The two times are the index's own per-conversation estimates, and they describe the work rather than any particular entity's process — a procedure also carries queueing, routing and approval that no model touches. Read this as which work is worth pointing at first, not as a forecast of any one service.

Worldwide

The most-used use cases

The 100 largest use cases by measured usage, across every kind of work rather than one domain. Switch between business usage through the API and everyday use on Claude.ai, and between work and all usage.

  1. 13.27%Used for work 96%18m → 2m
  2. 22.58%Used for work 97%21m → 3m
  3. 32.03%Used for work 98%18m → 3m
  4. 42.00%Used for work 84%20m → 2m
  5. 51.94%Used for work 100%24m → 3m
  6. 61.58%Used for work 78%44m → 2m
  7. 71.34%Used for work 85%5.3h → 12m
  8. 81.24%Used for work 99%29m → 3m
  9. 91.11%Used for work 91%1.1h → 2m
  10. 101.06%Used for work 68%18m → 4m
  11. 110.99%Used for work 99%50m → 5m
  12. 120.96%Used for work 87%22m → 3m
  13. 130.93%Used for work 98%44m → 4m
  14. 140.88%Used for work 98%34m → 3m
  15. 150.85%Used for work 90%20m → 2m
  16. 160.84%Used for work 98%22m → 3m
  17. 170.76%Used for work 96%1.1h → 5m
  18. 180.75%Used for work 97%26m → 3m
  19. 190.75%Used for work 99%14m → 2m
  20. 200.73%Used for work 99%49m → 4m
  21. 210.73%Used for work 100%1.0h → 4m
  22. 220.66%Used for work 100%1.4h → 4m
  23. 230.64%Used for work 96%20m → 2m
  24. 240.63%Used for work 95%1.7h → 6m
  25. 250.63%Used for work 97%17m → 3m
  26. 260.59%Used for work 94%1.5h → 9m
  27. 270.59%Used for work 98%16m → 2m
  28. 280.58%Used for work 100%36m → 2m
  29. 290.56%Used for work 93%1.4h → 5m
  30. 300.56%Used for work 98%2.4h → 9m
  31. 310.56%Used for work 99%1.3h → 6m
  32. 320.52%Used for work 83%2.4h → 5m
  33. 330.51%Used for work 93%49m → 5m
  34. 340.48%Used for work 98%22m → 2m
  35. 350.48%Used for work 97%1.1h → 5m
  36. 360.46%Used for work 98%1.6h → 5m
  37. 370.46%Used for work 83%1.3h → 5m
  38. 380.45%Used for work 99%20m → 3m
  39. 390.41%Used for work 97%3.5h → 10m
  40. 400.41%Used for work 80%20m → 3m
  41. 410.41%Used for work 95%29m → 3m
  42. 420.41%Used for work 97%40m → 4m
  43. 430.40%Used for work 98%19m → 3m
  44. 440.40%Used for work 99%1.5h → 5m
  45. 450.39%Used for work 100%32m → 3m
  46. 460.38%Used for work 95%1.8h → 4m
  47. 470.38%Used for work 97%17m → 2m
  48. 480.37%Used for work 98%40m → 3m
  49. 490.37%Used for work 97%1.2h → 5m
  50. 500.37%Used for work 100%37m → 3m
  51. 510.37%Used for work 85%29m → 2m
  52. 520.36%Used for work 99%33m → 3m
  53. 530.35%Used for work 98%17m → 3m
  54. 540.35%Used for work 95%43m → 3m
  55. 550.34%Used for work 100%28m → 4m
  56. 560.33%Used for work 96%21m → 3m
  57. 570.32%Used for work 100%1.3h → 3m
  58. 580.31%Used for work 99%1.3h → 5m
  59. 590.30%Used for work 92%2.6h → 9m
  60. 600.30%Used for work 91%26m → 2m
  61. 610.30%Used for work 88%2.1h → 5m
  62. 620.29%Used for work 80%26m → 3m
  63. 630.29%Used for work 99%56m → 6m
  64. 640.28%Used for work 97%1.1h → 4m
  65. 650.28%Used for work 100%17m → 2m
  66. 660.28%Used for work 90%1.8h → 8m
  67. 670.28%Used for work 98%19m → 2m
  68. 680.27%Used for work 97%33m → 3m
  69. 690.27%Used for work 97%34m → 6m
  70. 700.27%Used for work 93%16m → 2m
  71. 710.27%Used for work 99%40m → 4m
  72. 720.26%Used for work 100%17m → 2m
  73. 730.25%Used for work 91%29m → 5m
  74. 740.24%Used for work 97%52m → 4m
  75. 750.24%Used for work 78%17m → 2m
  76. 760.24%Used for work 89%51m → 3m
  77. 770.24%Used for work 100%2.3h → 4m
  78. 780.24%Used for work 91%37m → 4m
  79. 790.23%Used for work 98%2.1h → 9m
  80. 800.23%Used for work 100%17m → 2m
  81. 810.23%Used for work 88%8.0h → 13m
  82. 820.23%Used for work 95%24m → 2m
  83. 830.23%Used for work 69%14m → 2m
  84. 840.23%Used for work 81%28m → 3m
  85. 850.23%Used for work 99%2.5h → 8m
  86. 860.23%Used for work 95%2.1h → 6m
  87. 870.23%Used for work 90%1.7h → 8m
  88. 880.22%Used for work 82%4.8h → 13m
  89. 890.22%Used for work 100%1.4h → 4m
  90. 900.22%Used for work 100%28m → 3m
  91. 910.21%Used for work 94%54m → 5m
  92. 920.21%Used for work 100%23m → 3m
  93. 930.21%Used for work 78%31m → 4m
  94. 940.21%Used for work 71%31m → 2m
  95. 950.20%Used for work 92%31m → 3m
  96. 960.20%Used for work 93%4.1h → 8m
  97. 970.20%Used for work 93%1.8h → 7m
  98. 980.19%Used for work 89%2.5h → 8m
  99. 990.18%Used for work 92%58m → 5m
  100. 1000.18%Used for work 96%1.3h → 4m

About the data. Shares come from the Anthropic Economic Index (CC-BY 4.0), which classifies millions of real conversations. It measures usage of Claude, so it shows what AI is asked to do and in what proportion — not a survey of which companies have adopted what. Time estimates are Anthropic’s, per conversation.

What a number means. A use case’s share is its portion of all measured usage in that source. Business and consumer usage are measured separately and never added together. A use case with no matching topic in the index is listed without a number rather than estimated. Release 2026-06-26 · dataset · retrieved Sep 26, 2026.