AI Systems Can Cite Different Sources: What One Study Found

Review corrected social-citation figures from a commercial study, including its sample limits and implications for testing AI visibility.

Illustration for: AI Systems Can Cite Different Sources: What One Study Found

Do AI systems use the same sources?

Different AI systems can retrieve and cite different sources for similar questions. One commercial study tracked 848 prompts for 29 brands in eight industries from April 10 to September 16, 2026. Its headline corpus included 10.2 million citation events; those events are not a representative sample of all customer searches. Original study.

What did the study find?

The original study corrected its redirect handling on October 13, 2026. The figures below use that correction. Each percentage is a share of the named system’s citation volume in the engine-specific comparison—not a share of users, answers or recommendations.

Corrected citation shares in the study’s comparison dataset
Source categorySearch AI summariesConversational search modeCompared AI assistant
Video-sharing platform7.11%6.96%0.004%
Professional network1.53%1.85%0.294%
Discussion forum2.28%2.01%0.090%

The linked source identifies the particular platforms and systems. Its engine-specific denominators must not be substituted for the headline corpus. Study definitions and correction.

Does social media engagement get you cited?

The study estimated that 47.7% of cited videos had fewer than 1,000 views. Its engagement analysis sampled 1,800 of 9,941 unique cited videos and weighted the sample. This does not establish a 1,000-view eligibility threshold or prove that engagement causes citations.

Cited professional-network posts were a median of 35 days old when first observed being cited. First observed citation is not necessarily the first-ever citation. These findings may not generalize to your topics.

What should a local business do with this?

  1. Explain useful customer questions in formats suited to your audience.
  2. Make the underlying facts accurate and the evidence accessible.
  3. Test written and video material without assuming a format guarantees inclusion.
  4. Assess leads and outcomes, not only likes or citation counts.

Why are the AI platforms so different?

The study establishes differences within its dataset, not why those differences exist. It does not prove that a provider avoids a source for commercial reasons or that publishing on one site limits visibility to one assistant.

Frequently asked questions

Should I rely on one publishing channel?

Avoid dependence on a single channel. Choose useful channels for your audience and test results over time.

Is video useful for AI visibility?

Video sources appeared frequently in the studied search experiences. Its value for your business depends on your audience, topics and measured outcomes.

Do I need a large following to be cited?

A large following was not a prerequisite for every citation in this dataset. Engagement on a post and account audience size are different measures.

Sources and limitations

Reviewed October 29, 2026. External findings are not DigitalOS experiments or guarantees. Practical recommendations are DigitalOS guidance; research, products and policies can change.

  1. Original social-citation study · corrected October 13, 2026

    Commercial dataset: 848 prompts, 29 brands and eight industries, April 10–September 16, 2026. Corrected redirect handling changed the reported figures; not a representative census.

  2. Official AI-search eligibility guidance

    Living product documentation; no special AI schema is required for the search features described. Eligibility does not guarantee inclusion.

AI Visibility & AEO

Explore AI Visibility & AEO →

A quick, no-cost conversation about your goals, where AI can help first, and the next steps we'd recommend.