Profiles Over Posts: How ChatGPT and Google AI Overviews Cite Social Platforms Differently

Prior analyses in this series measured which domains AI engines cite. This one goes a level below the domain to ask which asset inside a social platform earns the citation. The answer differs by engine in a way that domain-level measurement cannot detect.

We used BrightEdge AI Hyper Cube and AI Catalyst to analyze cited URLs and the prompts behind them across Facebook, Instagram, LinkedIn, Reddit and YouTube, on Google AI Overviews and ChatGPT, over 12 weekly observations. The finding is that a marketing team's posting output and a platform's profile infrastructure are being consumed by different engines.

Posted Content Accounts for Most Google Citations and a Minority of ChatGPT Citations

On Google AI Overviews, the content a brand publishes on a schedule accounts for the majority of what gets cited. On ChatGPT the same content types account for a substantially smaller share of that platform's citations.

The asset types are unchanged between engines. The weighting is not.

What We Analyzed

We measured, for each social platform and each engine, the distribution of cited URLs by asset type, and the distribution of cited prompts by question type. Asset types were classified from URL structure. Question types were classified from prompt text. All findings are reported as proportions within a platform and engine, so that differences between engines remain the unit of analysis rather than differences in the size of either prompt set.

Data Collected

Key Finding

On ChatGPT, profile and entity pages account for a larger share of a platform's citations than posted content does, on all three of the platforms where a brand maintains an owned presence.

The practical reading is that these two engines are performing different operations against the same platform. One is retrieving content. The other is retrieving identity.

Entity Questions Concentrate on LinkedIn

The prompt data supports the same conclusion from an independent direction. Entity questions, meaning prompts asking who a company is, who owns it or who makes a product, account for a substantially higher share of LinkedIn's cited prompts than of any other platform's.

Two independently classified datasets agree. Prompts show identity questions concentrating on LinkedIn, and URLs show company and jobs pages answering them.

Current Status Questions Concentrate on Instagram

A second question type separates in the same way. Prompts asking whether something is still available, still open or still happening account for the largest share of Instagram's cited prompts among the five platforms.

This has a maintenance implication rather than a publishing one. When an engine retrieves a profile to answer a current status question, outdated fields are returned as the current state. Discontinued products, closed locations, prior operating hours and dead links are not treated as absent information.

Citation Breadth at the Page Level Is Shallow

Within the LinkedIn company pages observed in the cited URL set, no individual page was cited for more than a single prompt. The pattern is consistent with a lookup rather than a ranking. Coverage across the range of questions a category generates appears to matter more than repeated citation of any one page.

Cited URLs also frequently resolved to regional or country versions of a company profile rather than the global profile. Auditing only the primary profile would in those cases leave the cited asset unreviewed.

The Weighting Varies by Industry

The pattern holds across the nine industries analyzed, but the volume attached to it does not. Entity questions account for approximately 32% of LinkedIn's cited prompts in travel and under 3% in healthcare. The corresponding investment in profile accuracy should be weighted to how often a category generates identity questions in the first place.

What Marketers Need to Know

Audit the assets that no one owns editorially. Company pages, jobs pages, bios, categories, hours and links are cited persistently and are typically outside the content calendar and its review cycle. On LinkedIn, jobs pages were cited at a rate comparable to profile summary content.

Treat profile accuracy as a maintenance function with an owner. The current status finding indicates that stale fields are returned as answers rather than skipped.

Extend the audit beyond the primary profile. Regional and subsidiary profiles appear in the cited URL set.

Do not reallocate away from posting. Posted content accounts for the majority of Google AI Overviews citations on the platforms measured. The profile work is additive to an existing publishing motion rather than a substitute for it.

Measure below the domain level. A report stating that a brand is cited on a given platform does not distinguish between these two behaviors, and the corrective action differs depending on which one is occurring.

Technical Methodology

Key Takeaways