When a student submits a long-answer question, the auto-marker grades their response based on your mark scheme. In the new lesson experience, it also reads everything that comes before the question on the same page: the text, the images, the prior question stems.
So if your question says "Using the passage above, explain..." or "Refer to Figure 1," the auto-marker actually knows what the passage and the figure are.
Why this matters
Lessons built in the legacy experience grade each long-answer question in isolation. That works for self-contained questions but trips up on anything contextual. In the new experience, you can write the kind of question you'd write on a paper exam: anchored to a passage, a figure, a worked example.
Two scenarios that only work in the new experience
1. The long-scenario question
Drop a long scenario or case study above the question. Then write a mark scheme that requires the student to anchor their answer in the scenario.
Lesson page setup:
A paragraph (or several) describing the scenario: e.g. a 300-word case study about a school deciding whether to plant a vegetable garden on their roof, including the climate, the building's load-bearing limits, the school's budget, and the student demographic.
A Long Answer question: "Drawing on the scenario above, recommend whether the school should proceed. Justify each of your reasons with a specific detail from the scenario."
Mark scheme:
Award 1 mark per justified point. Each point must reference a specific detail from the scenario (load-bearing limits, climate, budget, demographic, etc.). Generic claims like "it's good for the environment" without a scenario reference get 0 marks. Maximum 4 marks.
In the legacy experience, the auto-marker would only see the student's answer and the mark scheme. It wouldn't know what the scenario said, so it couldn't tell whether a student was actually referencing it or making it up. In the new experience, it reads the scenario along with the answer, and your "must reference the scenario" criterion can actually be enforced.
2. The image-based maths question
Put a diagram, graph or photograph on the page, then ask the student to use the information in the image. The mark scheme can be as simple as the expected numeric answer.
Lesson page setup:
An image of a force diagram showing a 5 kg block on a 30° frictionless incline, with arrows for weight and normal force labelled.
A Long Answer question: "Using the diagram above, calculate the acceleration of the block down the incline. Show your working."
Mark scheme:
Correctly calculated result of 4.9 m/s² (accept 4.9 to 5.0). Award 1 mark for correct method (g sin θ or similar), 1 mark for correct numerical answer with units.
The mark scheme doesn't list the variables. They're only in the image. In the legacy experience, the auto-marker would have no idea what mass, angle or surface the question was about. In the new experience, it reads the image's caption and surrounding text, so it can grade the calculation in context.
What counts as "context"
The auto-marker picks up everything before the question on the same page:
Text paragraphs and headings
Images (with their captions)
Tables
Code blocks
Earlier question prompts (so multi-part questions on one page can build on each other)
Callouts
It does not pick up content from a previous page or the next page. If you want context to be available to the question, keep both on the same page.
What stays the same
Your mark scheme is still the primary signal. The auto-marker uses your mark scheme to decide marks, and the page context to interpret what the student wrote in light of the question.
Tools you already use still work:
Mark schemes (memo or instruction-based)
Manual remarking and feedback
The remark and resubmit flow
Insights that flag low-confidence grades
Tips
Put the context (passage, image, equation) directly above the question, not on a previous page.
Use clear, concrete language in your prompt. "Refer to the passage above" beats "Using your knowledge of the topic".
If the question depends on a worked example, leave the worked example on the page rather than hiding it in a separate document.
If the mark scheme depends on a calculation whose inputs are in an image, write the scheme as "correctly calculated result" or similar. Trust the auto-marker to read the image.