Imagine this scenario: A student submits a clear, well-organized essay. The argument flows. The thesis statement is clean and coherent. No grammar issues. The conclusion lands.
What does that tell us about what the student learned?
When a tool can produce, reorganize, or substantially rewrite an assignment, the gap between the quality of the work and the understanding behind it becomes harder to ignore. The work may look better, but that doesn’t automatically mean the student understands more. It also doesn’t automatically mean they learned less. The responsibility is to look more closely.
A stronger product is not the same as stronger understanding
An improved assignment can allow for many things: deeper understanding, useful feedback, language support, or a tool doing work the student can’t yet do independently.
Each of those is a different instructional situation. A finished product alone may not reveal why it improved.
In July 2026, Stanford and ETS published recommendations arguing that assessments focused primarily on final products may no longer capture the knowledge and skills they were designed to measure. Their proposed direction includes richer evidence gathered through student portfolios, conversations, formative and constructive feedback, and demonstrations of competency.
So, for schools, instead of focusing on the final product, the recommendation is to look more closely at what the student’s participation reveals.
What remains the student’s work?
In our pilot at New Directions Secondary School, students documented AI use, reflected on their authorship, participated in structured peer review, and developed a classroom AI policy. These activities gave students opportunities to discuss choices that a finished essay wouldn’t necessarily make visible: Why did you ask for help here? What did you keep, reject, or change? What did you decide not to outsource and why?
The pilot was small and conducted at one school, with no comparison group (yet). It doesn’t establish that these practices caused stronger learning or that their effects lasted well beyond the unit. But it offers a practical starting point: make students’ decisions available for discussion so you have the whole story behind the finished product.
Disclosure and transparency matter, but they are not, by themselves, evidence of understanding. A student can document every prompt and still struggle to explain the resulting argument. Another can use substantial language support while demonstrating a strong grasp of the underlying ideas. The evidence has to match what was intended to be taught.
Start with the learning goal
If the goal is argumentation, ask the student to explain why their evidence supports their claim.
If the goal is source evaluation, ask what they trusted, what they questioned, and why.
If the goal is independent writing, include an opportunity to write independently with questions that would allow the student to reflect on their authorship and voice.
If the goal is revision, examine the decisions made between drafts.
Not every assignment needs every check. Nor should every student have to defend their work as though they are under suspicion. There’s a level of trust that needs to be established both ways for this to be successful (more on that later).
The point is to build proportionate, routine opportunities to see understanding: a brief conversation, an annotated revision, a new problem, or an explanation of choice.
NYC’s pause makes this question more urgent
NYC’s 2026-2027 AI policy restricts student-facing generative AI in younger grades and limits high school use to approved settings. Its Technology in Schools Coalition is expected to deliver recommendations in April 2027.
As schools examine those experiences, they should not treat better-looking work as sufficient evidence of success. We should also ask what students can explain, apply in a new context, and do without the same assistance. That is a more demanding standard than adoption, and it’s more useful than treating every AI-supported assignment as a failure.
At FutureSkills, we see human AI literacy as the capacity to make (and take responsibility for) decisions about AI’s role in our thinking. Assessments should help make that capacity visible.