Your task
The work description you entered remains visible in the private report.
Data and evidence
The evidence behind each work-activity signal
See which source answers which question. We keep OpenAI technical exposure, observed AI use, modeled time saving, and occupational context separate. Only exact OpenAI E0/E1/E2 task classifications enter the public role exposure score.
Personal reports
A personal task uses public evidence only after you approve a task match
Your description is not treated as a new evidence source. The report suggests a comparable public task, shows both descriptions, and asks you to approve the match before using that task's estimate, explanation, and sources.
Matched public task
The public task you approve supplies its technical-exposure classification and source records.
AI-written text
OpenRouter may summarize the reviewed result, but it is not an evidence source and cannot alter any task classification or the role exposure score.
Evidence use
How each public AI source contributes
A source link alone is not enough. We also show what each source measures, how it is matched to a role or task, how it contributes to the analysis, and what it cannot establish.
| Source | Type of information | Where it applies | How it is matched | What it supports | What it does not show |
|---|---|---|---|---|---|
O*NET Database | Occupation, task, and task-rating backbone | Source backbone | O*NET-SOC Code + Task ID | Defines occupation-task keys and separately reports task importance, relevance, and frequency responses. | O*NET frequency is not a measure of time spent on a task and is never used as a work-time weight. |
OpenAI GPTs are GPTs | Primary task technical-exposure classification | Direct task evidence | O*NET-SOC Code + Task ID | Uses the same O*NET task key, preserves E0/E1/E2, and supports the paper's beta role score in which E2 contributes half and core tasks count twice supplemental tasks. | The result measures technical exposure under the study rubric. It is not a probability, observed adoption, work-time share, or automation forecast. |
EconEvals June 2026 | Supporting modeled time-saving research | Supporting task evidence | Exact normalized occupation title + task title | Provides a separate modeled chatbot time-saving band that remains inspectable on matched tasks. | Synthetic chatbot simulations are not observed workplace change and do not affect the headline task classification or role exposure score. |
Anthropic Economic Index | Observed AI use around task wording | Observed task-use evidence | Normalized exact task title | Useful signal that similar task wording appears in observed AI use. | Zero means no use was observed for that task in the public snapshot. It is not a zero-capability or zero-automation estimate, and Anthropic data does not change the role exposure score. |
Microsoft Working with AI | Occupation-level benchmark | Role or activity benchmark | Occupation or work activity mapping | Good comparison context for how generative AI applies across occupations. | Not task-level proof, so it should not be assigned to individual tasks. |
Anthropic Economic Index
Anthropic labor-market analysis used as an external real-world AI usage signal when occupation mappings are available.
Source homepageCedefop ESJS
European Skills and Jobs Survey signals used as an external interaction and task-behavior layer when occupation mappings are available.
Source homepageESCO
European Skills, Competences, Qualifications and Occupations taxonomy used for occupation matching and skill enrichment.
Source homepageEconEvals
Task-level modeled time-saving estimates for text-only chatbot use across O*NET tasks.
Source homepageMicrosoft Research Working with AI
Microsoft Research occupation-level AI usage or applicability signals used as an external usage layer when local extracts are available.
Source homepageO*NET
Primary United States occupation, task, skill, knowledge, ability, work activity, work context, and tools source.
Source homepageOECD PIAAC
Programme for the International Assessment of Adult Competencies public-use data used as an aggregate task-use context layer.
Source homepageOpenAI GPTs are GPTs
OpenAI occupation exposure mappings used as an external theoretical AI exposure layer when local extracts are available.
Source homepageVersions
Retrieved source versions
We keep track of exactly which version of each database we used and when it was retrieved.
| Source | Version | Retrieved | Status |
|---|---|---|---|
| EconEvals | econevals-june-2026-onet-30.2 | Aug 25, 2026 | ready |
| Microsoft Research Working with AI | microsoft-signals-2026-08-25 | Aug 25, 2026 | ready |
| OpenAI GPTs are GPTs | openai-signals-2026-08-25 | Aug 25, 2026 | ready |
| Anthropic Economic Index | anthropic-signals-2026-08-25 | Aug 25, 2026 | ready |
| Microsoft Research Working with AI | microsoft-signals-2026-08-25 | Aug 25, 2026 | ready |
| OpenAI GPTs are GPTs | openai-signals-2026-08-25 | Aug 25, 2026 | ready |
| Anthropic Economic Index | anthropic-signals-2026-08-25 | Aug 25, 2026 | ready |
| Anthropic Economic Index | anthropic-public-ai-evidence-2026-08-25 | Aug 25, 2026 | ready |
| OpenAI GPTs are GPTs | openai-public-ai-evidence-2026-08-25 | Aug 25, 2026 | ready |
| Anthropic Economic Index | anthropic-signals-2026-08-25 | Aug 25, 2026 | ready |
| Microsoft Research Working with AI | microsoft-signals-2026-08-25 | Aug 25, 2026 | ready |
| OpenAI GPTs are GPTs | openai-signals-2026-08-25 | Aug 25, 2026 | ready |
| Anthropic Economic Index | anthropic-public-ai-evidence-2026-08-25 | Aug 25, 2026 | ready |
| OpenAI GPTs are GPTs | openai-public-ai-evidence-2026-08-25 | Aug 25, 2026 | ready |
| OpenAI GPTs are GPTs | openai-signals-2026-04-24 | Apr 24, 2026 | ready |
| Microsoft Research Working with AI | microsoft-signals-2026-04-24 | Apr 24, 2026 | ready |
| Anthropic Economic Index | anthropic-signals-2026-04-24 | Apr 24, 2026 | ready |
| Anthropic Economic Index | anthropic-public-ai-evidence-2026-04-24 | Apr 24, 2026 | ready |
| OpenAI GPTs are GPTs | openai-public-ai-evidence-2026-04-24 | Apr 24, 2026 | ready |
| Microsoft Research Working with AI | microsoft-signals-2026-04-24 | Apr 24, 2026 | ready |
| OpenAI GPTs are GPTs | openai-signals-2026-04-24 | Apr 24, 2026 | ready |
| Anthropic Economic Index | anthropic-signals-2026-04-24 | Apr 24, 2026 | ready |
| Anthropic Economic Index | anthropic-public-ai-evidence-2026-04-24 | Apr 24, 2026 | ready |
| OpenAI GPTs are GPTs | openai-public-ai-evidence-2026-04-24 | Apr 24, 2026 | ready |
| Anthropic Economic Index | anthropic-signals-2026-04-24 | Apr 24, 2026 | ready |
| Microsoft Research Working with AI | microsoft-signals-2026-04-24 | Apr 24, 2026 | ready |
| OpenAI GPTs are GPTs | openai-signals-2026-04-24 | Apr 24, 2026 | ready |
| Anthropic Economic Index | anthropic-public-ai-evidence-2026-04-24 | Apr 24, 2026 | ready |
| OpenAI GPTs are GPTs | openai-public-ai-evidence-2026-04-24 | Apr 24, 2026 | ready |
| OpenAI GPTs are GPTs | openai-signals-2026-04-24 | Apr 24, 2026 | ready |
| Anthropic Economic Index | anthropic-signals-2026-04-24 | Apr 24, 2026 | ready |
| Microsoft Research Working with AI | microsoft-signals-2026-04-24 | Apr 24, 2026 | ready |
| Anthropic Economic Index | anthropic-public-ai-evidence-2026-04-24 | Apr 24, 2026 | ready |
| OpenAI GPTs are GPTs | openai-public-ai-evidence-2026-04-24 | Apr 24, 2026 | ready |
| ESCO | esco-api-2026-04-09 | Apr 9, 2026 | ready |
| O*NET | db_30_0_text | Apr 9, 2026 | ready |
Limitations
Source-specific caveats
Every data source has limitations. We show them so you can understand how each source should be interpreted.
ESCO · ESCO limitations
ESCO is a European taxonomy and may not map perfectly to O*NET occupations on a one-to-one basis.
EconEvals · EconEvals limitations
EconEvals models possible time savings for text-only chatbot use; it does not measure realized savings in workplaces.
O*NET · O*NET limitations
O*NET is rich and role-specific, but it is US-centered and should not be read as a direct measure of every local labor market.
OECD PIAAC · PIAAC limitations
PIAAC is used as aggregate context, not role-level ground truth.