What an Agent Should Do With LinkedIn and YouTube Citations

A year ago the offsite conversation was a footnote in an AI visibility program. It is not a footnote anymore. LinkedIn and YouTube are now real citation sources in AI answers, and the data shows they behave nothing alike. A tool that reports "you were mentioned on a social platform" is not enough. An agent has to know which page type on LinkedIn got cited, which video on YouTube got cited, and which engine did the citing, because the move is different in each case.

This post is the agent-action version of the LinkedIn and YouTube citation data. The evidence is two dated Promptwatch reports. The action is a named Promptwatch feature at the end.

The LinkedIn data: page type is the unit, not the platform

The LinkedIn citation page types report is the one to read first. Its measurement window runs from May 18 through June 17, 2026, and it covers LinkedIn citations found in ChatGPT, Google AI Mode, Google AI Overviews, and Perplexity. The headline is not "LinkedIn is cited." The headline is that the page type that wins changes by engine.

In the combined view across the four engines, Pulse articles accounted for 37.67 percent of LinkedIn citations, regular posts for 32.19 percent, and company pages for 13.35 percent. That combined number is a useful baseline, but it hides the split that matters. Google AI Mode cited Pulse articles for 44.85 percent of its LinkedIn citations, with posts a distant second at 27.37 percent. Google AI Overviews cited Pulse articles at 42.25 percent and gave meaningful weight to LinkedIn's own curated top content pages at 8.14 percent, a category AI Mode barely touched. Perplexity was the one engine where ordinary feed posts beat polished Pulse articles, 41.88 percent to 32.46 percent. Both Google surfaces cited the LinkedIn homepage 0 percent of the time.

That is four engines, four different winning formats. An agent that treats LinkedIn as one channel will optimize for the wrong format on at least three of them.

What an agent should do automatically with LinkedIn

The first move is to stop treating LinkedIn as a brand page problem. The data says the homepage and the company page are not where the citation comes from. The citation comes from Pulse articles and feed posts, and the engine decides which. An agent that manages a LinkedIn program should bias toward Pulse articles for the Google surfaces and toward fresh, specific feed posts for Perplexity, because that is what the data says each engine retrieves.

The second move is to track the page type on the prompt, not just the domain. A citation to linkedin.com is not actionable. A citation to a specific Pulse article on a specific prompt is. The agent should log which Pulse article got cited, on which prompt, by which engine, and flag the prompts where a Pulse article that is not yours won the slot. That is the gap a content program can close.

The third move is to connect the citation to a crawl. LinkedIn citations do not appear from nowhere. The engine retrieved the article. If the agent also reads crawler logs, it can see whether the LinkedIn surface was even in the retrieval path, which tells you whether the gap is a content gap or a crawl gap.

The YouTube data: small, focused, old

The YouTube picture is different. The YouTube citation data report covers February 1 through February 28, 2026, and it rests on more than 100 million tracked citations. In that window, YouTube accounted for 2.74 percent of citations in Perplexity, 2.25 percent in Google AI Overviews, and 0.05 percent in ChatGPT. The gap between Perplexity and ChatGPT is roughly 55x. A combined "AI citation rate" for YouTube would be a poor planning number, because the engine decides whether YouTube matters at all.

The shape of the cited videos is the part that surprised people. Nearly 80 percent of cited videos had fewer than 100,000 views, and 54 percent were more than two years old. The videos AI engines cite are not the viral ones. They are small, focused, and old. Engagement, views and subscribers, barely correlates with citation odds. Topical precision does.

What an agent should do automatically with YouTube

The first move is to check whether YouTube is even a surface for the client's engines before investing in it. If the client's buyers use ChatGPT, YouTube is close to a rounding error at 0.05 percent in the February data, and the budget is better spent elsewhere. If the buyers use Perplexity or Google AI Overviews, YouTube is a real surface, and the budget is justified. The agent should make this check per client, not per category.

The second move is to bias the YouTube program toward topical precision over production value. The data says the cited videos are small channels that consistently cover one topic area. An agent that manages a YouTube program for AI visibility should track which videos get cited on which prompts, and flag the topic areas where a cited video is not the client's. That is the gap a video brief can close.

The third move is to ignore recency as a signal. The cited videos are old. A program that deletes or unpublishes old videos because they are old is removing the exact asset the engines cite. The agent should flag old videos as assets, not liabilities, and protect them.

The wider social picture

LinkedIn and YouTube sit inside a larger social citation field. The social media citations by AI model report is a live page, published December 23, 2025, and its values move with the sample. A snapshot on August 30, 2026 put Reddit at 3.36 percent of citations in the combined view, YouTube at 2.94 percent, Facebook at 1.15 percent, LinkedIn at 0.74 percent, Instagram at 0.65 percent, TikTok at 0.13 percent, and X at 0.07 percent. Those are directional, not fixed. They should not be merged numerically with the fixed February YouTube study or the fixed May to June LinkedIn study, because the windows are different. Use the live page for the current direction, and use the dated studies for the hard numbers.

The direction is the part that matters for an agent. Reddit is the single biggest social source overall, but the per-engine split is what changes the move. ChatGPT leans Reddit. Perplexity leans YouTube. The Google surfaces lean LinkedIn Pulse. An agent that runs one social playbook across all engines is optimizing for the wrong channel on at least two of them.

The named feature: listYoutubeCitations, listRedditCitations, and citation trends

The measurement that turns the population report into a site-level action is the citation analytics layer in Promptwatch, exposed through the MCP tools listYoutubeCitations and listRedditCitations, and the citation trends view that sits behind them. listYoutubeCitations returns the YouTube citations for a project or prompt, with the video, the engine, and the prompt context. listRedditCitations does the same for Reddit. The citation trends view tracks how those citation sources and their volume move over time, so an agent can tell whether a channel is building, plateauing, or eroding, instead of reading a single count.

The acting half is the Content Agents layer. When listYoutubeCitations shows a competitor's video winning a prompt the client cares about, the content gap is the brief for the next video. When listRedditCitations shows a Reddit thread being cited on a buyer prompt, the action is to show up in that thread honestly, not to spam it. The Content Agents review inbox is the gate that keeps a human on the publish step.

The pattern is the one this site is built on. The tool reports the citation, the agent proposes the exact move, and a human ships it. LinkedIn and YouTube are the two offsite surfaces where that pattern matters most in 2026, because they are the two where the format, not the platform, is the unit of work.