Quick answer
Keep originals, clean the text and number paragraphs. Ask AI to work only from those materials, identify supporting paragraphs and expose uncertainty. Check every summary claim against the source, especially numbers, negatives and conditions, before using it for revision.
Decide what you may submit
Choose material you have permission to use and send to the service. Remove names, student identifiers, accounts, unpublished research and unnecessary private details. Retention, training use, access permissions and regional conditions vary. Check the settings of your actual service instead of applying KitNelo’s local-processing statements to an external AI provider.
UNESCO’s guidance on generative AI in education and research emphasizes human-centered use and data privacy among other concerns. Our workflow is editorial advice, not an endorsement by UNESCO of this site or a model. Follow course restrictions on AI-assisted assignments.
Prepare text before passing recognition errors onward
- 1Record the material name and page, then give paragraphs stable labels such as P1 and P2. Keep the original images or files for checking.
- 2For screenshots, perform OCR and compare technical terms, numbers, units and reading order with the image. Extraction does not establish that the source itself is correct.
- 3Split long material by chapter or concept. Do not ask for complete coverage after truncating the input. Record which paragraphs were submitted and what remains outside the task.
- 4Start with a small, non-sensitive sample. Text statistics can help notice an incomplete paste or large length change. Character counts are not model token counts and do not measure summary accuracy.
An adaptable summary prompt
Example prompt: Use only the numbered material below. State the topic, key concepts and unresolved points. Identify the supporting paragraph for every claim. When evidence is absent, write that the material does not specify it, without adding outside knowledge. Preserve numbers, units, negatives and conditions. Add three self-test questions, followed separately by reference answers. Material: P1 … P2 … .
This is an editorial task template, not a guarantee that every model follows it. A paragraph label is not proof of a valid citation: open and compare the actual source. Request background information as a separate, clearly labeled task with accessible sources, then verify those independently rather than mixing additions into your notes.
Check each claim using a small example
Imagine an invented P1: The experiment was observed at room temperature for ten minutes; low-temperature conditions were not measured. A summary claiming stability at low temperatures adds an unsupported conclusion. Preserve room temperature, ten minutes and the unmeasured condition. This is a fictional teaching example, not a research result.
Split the summary into checkable statements. Find the original support for each and look for changes from possible to certain, some to all, or not demonstrated to demonstrated. Remove or flag unsupported claims. Repeated agreement from another generated answer is not independent evidence.
Check that self-test questions can be answered from the submitted material. Repair invented page references, questions needing missing information and answers contradicting the source. Save a separate verified copy and record its scope and corrections.
Use the summary to support learning
- Answer a question or explain a concept before looking at the reference answer, then compare with your checked summary and original. Use the active recall workflow to record your answer and omissions.
- Read the source for complex diagrams, derivations and procedures. A short summary does not guarantee every relationship is retained.
- Keep source locations, checking dates and unresolved points so a draft does not become confirmed knowledge later.
- KitNelo OCR and text statistics assist material preparation. These entries do not automatically invoke generative AI or validate model answers.
Common questions
Does asking the model not to make mistakes guarantee accuracy?
No. Instructions can constrain a task but do not replace source checks. Verify claims, numbers, conditions and references, and retain uncertainty when evidence is missing.
Why does a shorter summary omit important material?
Input scope, truncation and task requirements can all cause omissions. Check completeness first, then evaluate against the source’s purpose and conditions rather than length alone.
May I upload private class or work material?
Check material permissions, course or workplace rules and the service’s data settings first. If submission is not permitted or cannot be confirmed, use an allowed local preparation method.
Further reading & sources
Put the guide into practice
Open the relevant tool, preview the result and check your exported file.


