Lesson 4: Prompting for Non-Visual AI Productivity (NVAIP)

Course: Foundations of Non-Visual AI Productivity (AI Basics)

Lesson content

Lesson Objective: By the end of this lesson, you will be able to write prompts that produce clear, structured outputs you can review efficiently using a screen reader or keyboard-first workflow.
What you will learn: how to write prompts that produce clear, structured outputs you can review efficiently using a screen reader or keyboard-only workflow. In NVAIP, good prompting is not “fancy wording.” It is a repeatable method that saves time and reduces mistakes.
NVAIP prompting rule (the course spine): In every prompt, request:
- Structure: headings, bullets, tables, or checklists.
- Constraints: length limit, tone, audience, and what not to change. Use negative constraints to state what the AI should not do.
- Verification: assumptions, uncertainty, and a short “what to verify” list.
The NVAIP workflow loop in prompting: Use the same loop you learned in Lessons 1 to 3.
- Orient: tell the AI your goal, audience, and constraints.
- Plan: ask for steps and checkpoints (especially for multi-step tasks).
- Execute: generate in small chunks (sections) rather than one long output.
- Verify: request assumptions, uncertain parts, and what to spot-check.
- Recover: ask “what changed?” and iterate with corrections.
How AI uses context: AI responds based on the information you provide in the current chat (including earlier messages) and any files or text you paste. If you want accurate output, provide the key details: names, dates, numbers, purpose, and the intended audience.
Core prompting ingredients (simple and reliable):
- Task: what you want the AI to do (summarize, draft, rewrite, extract, plan).
- Audience: who it is for (manager, client, student, public post).
- Constraints: length, tone, style, and what must not change.
- Negative constraints: what the AI should not do (for example: “Do not use flowery language.” “Do not change the technical terms.”).
- Format: request Markdown headings (for example: “Use H2 for sections”) so you can jump between them using your screen reader’s H key. Also request bullets, tables, or checklists when helpful.
- Verification: ask for assumptions, uncertainties, and a spot-check list.

Visualizing the prompt (prompt ingredients diagram):

Think of your prompt like ingredients that feed into the AI. The clearer the ingredients, the easier the output is to review.

Goal + Context + Task
+ Constraints + Negative constraints
+ Format (Markdown headings)
+ Verification request
            |
            v
           AI
            |
            v
Structured output you can navigate and verify

Example prompt (email rewrite) structure-first:
Rewrite the email below so it is clear and professional. Audience: my supervisor. Keep the meaning the same. Keep it under 150 words. Do not use flowery language. Do not change any technical terms. Format: Use Markdown headings (use H2) for “Summary” and “Action items,” then use bullet points for action items. After rewriting, list what you changed and why. Then give a 5-item checklist of what I should verify (names, dates, numbers, commitments).
Prompt patterns you can reuse (NVAIP signatures):
- Assumptions and uncertainty: “List assumptions and anything you are not sure about.”
- Negative constraints: “Do not use flowery language.” “Do not change technical terms.”
- Spot-check list: “Give 5 items I should verify before I send or share this.”
- Change control: “Show what changed and why (short list).”
- Structured output: “Use Markdown headings and bullets. No long paragraphs.”
- Versioning: “Provide Version A (short) and Version B (detailed).”
- Evidence when needed: “If you make factual claims, cite a source or say ‘no source provided.’”
Iterate, don’t accept the first draft: In NVAIP, the first output is a draft. Your job is to refine it using small corrections. Useful follow-ups:
- “Make it shorter and keep the meaning.”
- “Put the output into a table.”
- “Add missing steps and highlight risks.”
- “List any places where the answer might be wrong.”
- “Remove anything unnecessary. Keep technical terms unchanged.”
Mini exercise (5 minutes):
1. Choose a real paragraph or email you wrote recently.
2. Write an Orient line: “Goal is… Audience is… Constraints are…”
3. Add one negative constraint, for example: “Do not use flowery language” or “Do not change technical terms.”
4. Prompt AI to rewrite it using Markdown headings (use H2) plus bullets and a word limit.
5. Ask: “What changed and why?” and “Give a 5-item verification checklist.”
6. Verify 2 items (names, dates, numbers) against your original source, then finalize.
Reminder: response time and quality can vary by tool, model, and connection. When outputs are too long or hard to review, request a shorter structure (Markdown headings plus bullets) or ask the AI to produce the result in sections.

Continue

All lessons in this course

References

1. IBM, “What is artificial intelligence (AI)?” IBM Think. https://www.ibm.com/think/topics/artificial-intelligence
2. Microsoft, “Copilot prompt ideas” (Microsoft Support). https://support.microsoft.com/en-us/copilot
3. W3C, “Web Content Accessibility Guidelines (WCAG) 2.2”. https://www.w3.org/TR/WCAG22/