Give useful context, sources, examples, and media

This course remembers completed sections, assessment point totals, and your last course page in this browser on this device. RayAI does not receive this progress or anything you write. Selected answers and written responses are not saved. Clearing browser data, using private browsing, or changing browser or device can remove or hide saved progress.

MODULE 3 | 35 minutes | FAST TRACK CORE + GO DEEPER

Give useful context, sources, examples, and media

Before uploading a document, image or recording, check for names, account details, faces, background information and hidden comments. Cropping or covering text may leave recoverable information. For practice, use a fictional sample or a small, checked public excerpt. Module 6 explains the fuller privacy and safety checks.

Core idea: context should change the answer

Useful context includes facts, definitions, source material, prior decisions, constraints, and examples that a capable person would need. Noise makes a prompt longer without improving the task. More context is not automatically better context.

• Tell the model whether it may use general knowledge, must use only supplied sources, or should search current authoritative sources if the tool supports it.

• Label instructions and source material clearly so they are not confused.

• State what to do when the answer is absent, ambiguous, or contradictory.

• Start a fresh conversation or restate the current brief when old context has become stale or conflicting.

• Check whether the tool is using saved memory, prior chats, connected files, or retrieved content that could influence the answer beyond the text you just entered.

SOURCE-GROUNDED TEMPLATE

The <source> and </source> tags mark the beginning and end of your source material. Replace [Authorized source material] with text you are allowed to share, leaving the tags in place. For example: <source>Budget ceiling: $600.</source>. Replace [Your question] with your question. The tags organize the text; they are not a security barrier.

Zero-shot and few-shot prompting

Zero-shot means giving instructions without an example. Try it first when the task is straightforward. Few-shot means adding one or more input-output examples. Use examples when you need a specific pattern, tone, classification rule, or edge-case behavior.

• Choose examples that match the real task and include meaningful variation.

• Show missing-value behavior explicitly, such as “Not provided” or “Needs confirmation”.

• Remove private, copyrighted, or restricted material you are not permitted to provide.

• If examples conflict with instructions, the output may become inconsistent. Reconcile them before testing.

Multimodal prompting

A prompt may include an image, screenshot, diagram, audio clip, video, or document. State what the model should inspect, what it should ignore, and how the answer will be used. Visual recognition and optical character recognition (OCR, turning text in an image into editable text) can be wrong, so verify names, numbers, labels, and small text.

SCREENSHOT TEMPLATE

See it: grounding the event plan

PLANNING NOTES

GROUNDED TASK

Spot it

Write one label beside each line below: Instruction (what the AI should do), Source (a fact to use), Example (a model to copy), or Noise (words that do not help).

1. Return a three-column table.

2. The budget ceiling is $600.

3. A sample confirmed row should read: Budget | $600 maximum | Planning note 2.

4. You are brilliant and always exceed expectations.

Try it: practice the skill

What to do

1. Choose the task suggested for this module, or use another simple, low-risk task.
2. Open ChatGPT or another AI assistant in a separate tab or window.
3. For this module, add the facts or source material the AI needs. Tell it clearly what to do when information is missing.
4. Enter your prompt and send it. Do not include private or sensitive information.
5. Read the AI answer.
6. Return to this course and complete the short notes below.
   If you cannot use an AI assistant now, read the worked example and write what you would try later. Skip the steps that ask you to open, send or rerun a prompt. Label your notes "supplied example" or "planned test". Describe only what the example shows; if no output is supplied, write what you would look for. Keep checks you performed separate from checks you plan. Do not invent an AI result.