Editorial starter: this is a proposed research method, not a completed experiment or a claim about platform behavior.
Introduction
When an AI answer includes an image, the visible result raises several questions: where did the image come from, which page was credited, and what information was available on that page?
This research note outlines a repeatable way to record those observations. It does not assume that any single metadata field determines selection.
Methodology
Define a fixed set of brand and product questions. For each question, record the platform, date, exact prompt, displayed image, and any visible source link. Repeat the same questions across sessions to identify variation.
Keep a useful observation log
- Save the exact prompt and timestamp.
- Record whether an image appeared at all.
- Preserve the source URL and visible attribution.
- Separate observed behavior from explanations you have not tested.
Test setup
Inspect the source page for image metadata. This browser-side snippet can inspect a supplied HTML string; it does not reproduce an AI platform’s selection process.
function getOgImage(html) {
const document = new DOMParser()
.parseFromString(html, 'text/html');
return document.querySelector(
'meta[property="og:image"]'
)?.getAttribute('content') ?? null;
}What to measure
Compare image presence, source attribution, and consistency across repeated prompts. Record missing results as well as successful ones. A changed image does not by itself establish why the system selected it.
Limitations
A small prompt set cannot represent all users or all answers. Platform updates, session context, and location may affect observations. Metadata and selection can be correlated without a causal relationship.
Next steps
Run the protocol, retain the raw records, and publish the sample size and observation window alongside the findings. Until then, this note remains a research plan.