AI Scoring

If you’re already a customer, you can contact your account manager to request activation of Analytic rubrics and AI Scoring in your customer area. If you are not yet a Questionmark customer, please hit the ‘Talk to us’ button to organize a chat with us.

We are engineering on top of existing public large language models (LLMs). This allows us to optimize for grading use cases while still benefitting from the rapid pace of improvement in the generative AI industry. We are not tied to a specific LLM. We use prompting, ensembling (i.e. for evaluating different elements of a response, there may be more than one call to an LLM), and a variety of proprietary techniques to achieve our high accuracy.

The Rubric Editor lets you create and manage analytic rubrics, which define how responses are scored across multiple dimensions using a consistent scale. You can add rubric levels such as Very Good, Good, Poor, and Incomplete, and define the criteria for each level. Each dimension in the rubric automatically uses these levels, ensuring consistent scoring.

To get started, create or edit an Essay or File Upload question in the Standard Item Bank and set the Scoring style to Analytic rubric. A scoring instructions field, rubric levels, and dimensions will appear in the question editor. You can add new levels or dimensions as needed, and edit descriptions and scores to match your program’s requirements. The maximum score for a question is calculated automatically based on the highest rubric level and the number of dimensions.

Once your rubric is set up, it integrates with the Scoring Tool. Graders can use the rubric to review responses, apply consistent scoring, and, if AI Scoring is enabled, receive AI-generated suggestions that they can adjust before submitting final scores. This workflow saves time, ensures accuracy, and keeps graders in control.

No, your data won’t be used to train public models. Our AI service providers do not use any of the information provided to us or generated by us to train any model. 

No. AI Scoring ensures that your intellectual property remains protected. The data used during scoring is not stored or used to train any of the LLMs integrated in our solutions.

AI Scoring is built with data security and IP ownership in mind. Neither Questionmark nor the LLMs used claim copyright over source material or the AI-generated scores and feedback. Customers retain ownership of the scoring output in line with the applicable legal jurisdiction.

Customers retain full ownership of the scoring feedback and results produced via AI Scoring, subject to any applicable legal frameworks.

There is a limit of 25000 characters per response.

Yes, AI scores can be reviewed and modified using the Scoring Tool before submission.

When a score is edited, the AI highlight is removed, and the updated score appears in blue to indicate human modification.

Not by default. Spelling or typos will only impact scoring if the rubric explicitly includes them as a trait, such as a row for grammar, mechanics, or conventions. If no such criterion is defined, the AI model typically ignores minor errors and focuses on the overall quality and meaning of the response.

Ensure Analytic rubrics are enabled for your customer area. Confirm the assessment is using an Analytic rubric.