Frequently Asked Questions

Gradebird is an AI-assisted workflow for handwritten exams. You give an in-person paper exam, scan the completed packets, and review rubric-aligned assessment candidates alongside each student's work. You can revise anything and release results only when you are satisfied.

Gradebird is designed especially for free-form handwritten exams where the work matters: calculations, derivations, explanations, multi-step reasoning, and diagrams. It is a natural fit for STEM courses, including questions that mix conceptual and quantitative reasoning. It can also handle traditional autograding approaches like multiple choice.

Legacy autograders often require narrowly structured questions or rubrics and rigid answer formats. Gradebird can assess free-form handwritten work holistically, considering the question, your rubric, and the student's full answer together. You can start with an exam you already use, refine the rubric as needed, and keep the flexibility to handle partial credit, alternate solution paths, and nuanced reasoning. Gradebird also produces editable, question-level feedback that you can return alongside each student's original work.

It just takes a quick formatting pass before you give the exam. Paste your questions into our exam builder, use the import tool, or just send us your source and we'll format it for you within 24 hours. The result is a set of unique, print-ready student packets with dedicated response spaces and trackable labels. Give the exam on these packets and then decide later whether to use Gradebird or just grade traditionally. No account is required until you decide to grade.

Gradebird does not ask you to trust a score in isolation. It places the student's work beside the proposed score and feedback, with supporting signals such as how closely different models agreed and a summary of any disagreements. Some instructors review every response in detail, while others use model agreement to focus attention where it is needed most. Some users even return the feedback unedited, making verification part of the students' responsibility. Gradebird is simple but flexible, adapting to the workflow you prefer.

You can change the score, rewrite the assessment, or reject the recommendation. AI output is treated as a candidate for review—not an unquestionable answer. When a response is ambiguous or models disagree, Gradebird surfaces that uncertainty so you know where to look more closely. In almost all cases, the resolution is more rubric clarity, rather than repeated human intervention.

Almost anything—from a few sentences of free-form guidance to a highly itemized, point-by-point rubric. You can write your own, ask AI for a draft or critique, and revise it before or during grading. Gradebird assesses responses using the rubric and course context you choose.

Yes. Gradebird keeps rubric revisions traceable. If the current rubric changes after assessments have already been approved, affected work is flagged so it can be either bulk promoted or re-reviewed against the new rubric, at the instructor's discretion.

Pretty much anything—just write the exam as if you were grading it yourself. Mix multiple-choice, short-answer, diagram, essay, or long-form calculation questions, with no need to fit into a legacy autograding format. Gradebird holistically assesses each response using whatever rubric you provide.

In our experience, AI models are actually better than humans at understanding messy handwriting or pale scans. But you always have the response image available for inspection, and you have full control to disregard the recommendation and grade the response yourself.

Yes. Gradebird can hide student names throughout the grading queue, allowing instructors and graders to focus on the work itself. Each packet gets a memorable bird alias—Kookaburra, Wren, or perhaps the occasional Dodo—so responses remain easy to navigate without revealing whose work you are reviewing. Authorized instructors can reveal the student identity whenever needed.

We have found that students really appreciate the detailed feedback on every question. If you prefer not to share feedback, then students do not even know that Gradebird is being used.

No. Students take the exam on paper and do not need to sign in to Gradebird. If you choose to share feedback, each student receives an individual protected link to their report. Institutional clients can request an LMS integration.

When instructors share detailed reports, most grading questions are answered before they become formal regrade requests. Students see their original work, score, and the reasoning and feedback behind it in one place; when a question does arise, everyone can start from the same evidence.

Absolutely. One of the most exciting benefits of AI-assisted grading is that rich, question-level feedback can turn every assessment into a teaching opportunity, showing students not only what they missed but how to improve. You can even use Gradebird for zero-point in-class work where the goal is practice, reflection, and feedback rather than a grade.

No specialized student devices are required. Gradebird generates printable exam packets, and completed exams can be uploaded as bulk scans. A normal campus or office scanner that produces clear PDF files is generally a natural fit for the workflow.

No—and that constraint is central to Gradebird's reliability. The QR-coded packets let Gradebird present the AI with a carefully scoped task: the exact question, the corresponding student work, the point value, and your rubric. With arbitrary pages, there is no consistently reliable, auditable way to determine which marks belong to which prompt, so Gradebird uses structured packets instead of asking the model to guess.

Gradebird uses the QR codes on each page to sort bulk scans by student, question, and response area—even when pages arrive out of order. It detects duplicate, unreadable, unmatched, and missing pages, flags anything that needs review, and lets you correct matches or upload rescans without re-sorting the stack. A truly lost page remains clearly marked for instructor resolution rather than being silently omitted.

Yes. Instructors can invite collaborators and assign graders to specific published questions. Permissions can separately control whether a grader may edit the rubric or see student identity, allowing access to match the role you intend.

Yes, but students still need to submit their homework on a unique Gradebird packet. Some instructors have used this as a low-stakes test of the system, but most agree that grading homework is unnecessary and sometimes counterproductive in the AI era. Gradebird is designed for in-class assessments.

Gradebird supports all common LMS platforms via roster import/export or direct integration. If yours is not yet supported, let us know and we'll add it.

Yes. Gradebird uses a defense-in-depth approach, including encryption in transit and at rest, role-based access controls, isolated document processing, and durable audit trails. Our security program is mapped to the SOC 2 Trust Services Criteria for Security, Availability, and Confidentiality, using CIS Controls v8.1 IG1 as reference points.

Gradebird works most smoothly when student names, email addresses, and institutional IDs are provided through CSV import or LMS integration, enabling seamless delivery of grades and feedback. However, Gradebird can also be used without sharing identifiable student data—just use numbered packets and keep the packet-to-student mapping outside Gradebird. Gradebird can assess the submitted work without knowing who wrote it, and you can return grades or feedback through your own process.

Gradebird sends AI providers the minimum data necessary for each usage of AI. Grading requests include only the question, rubric, and relevant context, together with the selected student response crop. Since no direct or indirect identifiers are intentionally included, this minimizes the possibility that the data contains sufficient information to identify the student under FERPA's reasonable-person standard. Furthermore, we only work with AI providers who contractually commit to absolutely no generalized model training on any data we send to them.

Yes. Institutional clients can use their own approved AI providers and provider accounts. Because the institution contracts with the provider directly, it can generally establish its own terms for privacy, retention, data use, and model training, subject to the provider's capabilities. Gradebird is also open to signing an institution-specific Data Protection Addendum (DPA) where appropriate.

No. In line with our strong commitment to privacy and FERPA's use-and-redisclosure protections, Gradebird will never sell Student Data or disclose it for advertising, marketing, data brokerage, or any unrelated commercial purpose. Any necessary disclosure is limited to providing and securing the service, following the institution's instructions, or complying with law.