References in AI prompts reduce ambiguity by giving the model something concrete to follow. A reference can be a document, example, image, style sample, dataset or previous approved output — but you should state what the model should learn from it.
Types of References
- Source reference: facts the AI may use.
- Style reference: tone, visual language or format.
- Example reference: shows a desired output pattern.
- Image reference: identity, pose, composition or product.
- Negative reference: example of what to avoid.
Tell the AI What the Reference Controls
Do not simply attach an example and say “make it like this.” Say whether to preserve structure, tone, identity, color palette, factual content or composition.
Reusable Reference Prompt Structure
Use [reference] only for [specific role]. Preserve [elements]. Do not copy [elements that should change]. Task: [task]. Constraints: [constraints]. Output format: [format]. If the reference does not contain enough information, flag the missing information instead of guessing.
Practical Examples
A previous email can define tone; a product photo can define geometry; a report can provide facts; an approved landing page can define hierarchy; a portrait can define identity.
For image-specific workflows, see How to Use a Photo as an AI Reference. For prompt context, see Why Context Matters in AI Prompts.
Common Mistakes
- Contradictory references.
- No explanation of what to copy.
- Assuming a reference is always factual.
- Using confidential material without permission.
- Mixing too many styles or examples.
Frequently Asked Questions
Are examples a form of context?
Yes. Examples provide concrete evidence of the pattern, style or structure you want the model to reproduce.
Can I use multiple references?
Yes, but give each reference a clear role so they do not compete.
