a workforce program can show AI literacy, not just attendance, by scoring one real work task against a short rubric whose criteria map to the U.S. Department of Labor's five AI-literacy content areas. Have two people score each task independently, and report how often they agree. The DOL framework asks for "measurable AI skill development outcomes" but defines no test, so this is a method you can explain and defend. It is not an instrument any agency has approved.
This is general information about federal guidance, not legal or compliance advice. For how your outcomes count in WIOA reporting, ask your state workforce agency.
The question you are actually asking
Your funder, board or state contact has asked a short question: "How did you measure AI literacy?"
Most programs can answer only two ways. "Learners attended 12 hours" measures seat time. "Learners earned a certificate" measures course completion. Neither shows that a person can do anything differently at work. Think of a driving school that reports hours behind the wheel but never runs a road test. The hours are real, but they don't show anyone can drive.
You need a road test for AI use. It should take minutes to score, work for learners in different jobs, and hold up when someone asks, "Would a different scorer give the same result?"
What the federal framework says, and what it leaves out
The DOL issued its AI Literacy Framework as TEN 07-25 on 13 February 2026. It is voluntary. It lists five content areas:
- Understand AI Principles
- Explore AI Uses
- Direct AI Effectively
- Evaluate AI Outputs
- Use AI Responsibly
The framework wants results you can measure. It gives programs flexibility to choose the approach that "best drives measurable AI skill development outcomes," and asks them to "evaluate whether participants are gaining practical, transferable AI skills." It also admits that providers may be uncertain about "how to assess competency."
What it does not do is define a test, a score, a pass mark or a credential. That gap is the reason for this article.
Why a quiz is the wrong tool
A multiple-choice quiz is the obvious first idea. It's cheap and easy to grade. The catch is that it measures what a learner can recognize, not what they do when an AI tool hands them a confident wrong answer at work.
A related federal standard is useful here, even though it governs credentials rather than programs. TEGL 10-16, Change 3, the DOL's WIOA performance guidance, describes credentials issued "using a valid and reliable assessment of an individual's knowledge, skills and abilities." Borrow those two words:
- Valid means the task looks like the skill. Scoring real work against "checked the output against a source" is closer to the skill than a quiz question about what a hallucination is.
- Reliable means two scorers get the same answer. You can only claim that if you measure it.
Employers are asking for the same thing. A WGU survey of 3,128 U.S. hiring professionals, reported by COABE's AI Hub, found that 60% say AI makes candidates' real skills harder to evaluate. COABE suggests "pairing AI practice with work samples and explanations that show learners' own reasoning."
The crosswalk: five areas to scoreable evidence
Here is the core tool. For each DOL area, it names a behavior you can see on a page, the rubric criterion that scores it, and the evidence the learner must hand in. The four criteria are the ones in ed.cafe's 12-point work-sample rubric, each scored 0 to 3.
DOL AREA OBSERVABLE BEHAVIOR RUBRIC CRITERION EVIDENCE ON THE PAGE
-------------------- ----------------------------- --------------------- ------------------------------
Explore AI Uses Picks a real task from their 1. Real job task Task named, audience and
job where AI can help purpose stated
Direct AI Effectively Gives context and a clear 1. Real job task The prompt, as typed
instruction
Evaluate AI Outputs Checks output against a 2. Checked and fixed Named source, the error,
trusted source and fixes it the corrected text
Use AI Responsibly Keeps private data out; 3. Zero private data Invented data labeled as
owns the result invented
Understand AI Explains in plain words what 4. Plain explanation Two sentences: kept, changed,
Principles was kept, changed, and why and whyA sample passes at 8 of 12 points with no criterion at zero. The no-zeros rule matters. A learner who scores 3, 3, 0, 3 has done good work and also pasted private data into a tool. That should not pass.
Where the crosswalk is thin
Be honest about coverage when you report. Two DOL areas are only partly measured by one work sample.
Understand AI Principles is inferred from the explanation, not tested directly. A learner can write a good explanation without being able to say how a language model produces text. If your funder cares about that area, add three short-answer questions, scored right or wrong, and report them separately.
Explore AI Uses is scored through one chosen task. One task shows a learner can find one use. It does not show they can spot uses across their job. A second, shorter task in a different part of their work would strengthen this.
Saying this out loud makes your report more credible, not less.
A worked example with real-looking numbers
Illustrative example: a program finishes a cohort of 20 learners. Each hands in one work sample. Two staff members score all 20 independently, criterion by criterion. That is 80 criterion scores per scorer.
They compare results and record three numbers:
- Exact agreement: 56 of 80 criterion scores match exactly (70%).
- Within one point: 76 of 80 (95%).
- Same pass/fail verdict: 18 of 20 samples (90%).
The two samples where verdicts differ both hinge on criterion 3. One scorer counted a ticket number in the employer's real format as identifying; the other didn't. They write a rule ("a real-format account or ticket number counts as identifying") into a decision log and re-score both.
The report then reads: "14 of 20 learners passed the work-sample rubric (8 of 12, no zeros). Two scorers agreed on pass/fail for 18 of 20 samples before reconciliation." That sentence answers the funder's question.
These numbers are invented to show the arithmetic. ed.cafe has not yet scored a cohort, and the planned check on 20 real samples has not happened.
Reliability in practice: three habits
Most of the gain in agreement comes from three habits:
- Calibrate first. Before scoring, both scorers score the same three practice samples and talk through every gap of two points or more.
- Score criteria, not totals. Two scorers can both reach 8 for different reasons. Criterion-level comparison exposes that.
- Keep a decision log. Every rule you agree on goes in writing, so the next cohort's scorers inherit it.
If pass/fail agreement falls below what you would be comfortable defending, stop and recalibrate before you report anything.
Can this count as a WIOA measurable skill gain?
Possibly, but your state decides, not you. TEGL 10-16, Change 3 lists five types of measurable skill gain. Two are closest: a satisfactory report from a training provider on progress toward milestones set in advance, and passing an exam or showing progress on an occupational skill benchmark. Whether a rubric-scored work sample fits either one depends on your state's policy and how your program sets up its milestones.
For a line-by-line reading of how the Google AI Professional Certificate fares against these tests, see our TEGL 10-16 analysis. Never describe a rubric result as a credential.
Limits of this method
- It is a snapshot. One task, scored once. It does not show speed, volume or judgment over months.
- Scorers can't judge domain accuracy. A non-expert scorer can see that a source was named, not whether it was the best source for that trade. Ask for the source's name and date, and invite an employer to spot-check a few samples.
- Reliability must be shown each time. Last cohort's agreement doesn't carry over to new scorers.
- The DOL may issue more guidance. If it defines a measure, use theirs.
Decision checklist
- Does every learner hand in one real task with the prompt, the source, the fix and an explanation?
- Did two scorers calibrate, then score independently?
- Are agreement figures reported with their counts?
- Are certificate completion and rubric results reported as separate numbers?
- Have you told your funder which DOL areas the rubric covers only partly?
To watch a facilitator score one sample against these four criteria, open the rubric demo in the ed.cafe Staff Room.
Sources
- TEN 07-25: Artificial Intelligence Literacy Framework, U.S. Department of Labor ETA, 13 February 2026
- TEGL 10-16, Change 3: Performance Accountability Guidance for WIOA Title I, II, III and IV Core Programs, U.S. Department of Labor ETA, 11 June 2024
- AI Hub, COABE (Coalition on Adult Basic Education), accessed 15 October 2026
