AI checks whether you actually work before it recommends you.

A recommendation from an AI assistant almost never traces back to the brand's own website. It traces back to a Reddit thread, a YouTube walkthrough, a LinkedIn post from someone who actually used the product, because those sources answer the question a homepage cannot, which is whether the thing works for someone in the reader's situation. The assistant is not being lazy when it does this, it is doing what a careful buyer would do, which is to go find people who have already been through the decision and read what they said about it.

That behaviour sits at the heart of the third pillar of Total Graph Authority, the framework we use for earning a place in AI answers. The first pillar is being relevant through your own content, and the second is being trusted by the sources your audience already believes, but the third is being present in the conversations and how-tos your buyers are already reading and watching. The 2026 citation data shows that third pillar growing faster than either of the others, and it rewards a kind of usefulness that brand copy cannot fake.

What social utility actually earns you

Social content wins a specific kind of question, and it helps to be precise about which one. It rarely decides what your product is, since your own site and a solid third-party review can settle that, but it is often the only thing that answers whether the product works for a situation like the reader's and how to actually use it once they have it. Those eligibility and in-use questions get answered in transcripts, captions, on-screen text, comment threads, and community replies, all of which the models index and quote back into an answer. Picture a founder at a small logistics company asking an assistant whether a particular warehouse tool can run without a full-time admin, where the homepage will say it is easy to use because every homepage says that, so the model reaches instead for a Reddit reply from an operator who set it up alone over a weekend and a YouTube clip showing the actual onboarding screens. You earn a place in that answer by being useful where your buyers already gather, and the AI reflects that usefulness back rather than inventing it, because the room is full of real people helping each other and the model is standing in the doorway taking notes.

How fast social citations climbed

The shift toward social sources has happened quickly enough that most content teams have not adjusted to it yet. One study of 6.1 million citations across ten AI platforms found social references grew roughly fourfold between September and November 2025, climbing from about 2% to 7% of all citations, and a separate breakdown from Tinuiti put social's share at 6% in October and 9% by January 2026, with Reddit alone moving from around 2% to 5% and driving almost the entire gain. Google has leaned into this harder than anyone, to the point where one analysis found it cited Facebook in more than 19 million AI answers, treating an ordinary social post as a legitimate source for a question.

The base is still small, so this reads as an early-mover position rather than a settled one, and it lands unevenly enough across engines that a single approach will leave gaps. In that same 6.1-million-citation set, Google AI Overviews pulled about 5.0% of their citations from social, Gemini 2.9%, Perplexity 2.6%, and ChatGPT only 1.1%, which is a wide enough spread that where your buyers actually ask their questions changes what you should build.

Reddit and YouTube account for almost all of it

When an engine does reach for social, two platforms take almost all of what it pulls. In a study of more than 100 million citations, Reddit is the largest single social source at 46.4% of social citations and YouTube is second at 31.8%, and YouTube's lead over other video is wide enough to be almost a monopoly, cited around 200 times more often than any rival platform while Vimeo and TikTok each sit near 0.1%. On Google's AI Overviews it goes further still, where YouTube is the single most-cited domain of any kind at 29.5%, ahead of established reference sites, so a well-made video can outrank a polished article for the same question.

For most brands that narrows the work considerably, because a serious YouTube presence and genuine participation on the two or three subreddits where your buyers actually post will move more AI visibility than spreading effort thin across every network. A B2B software company, for instance, gets far more from a handful of honest setup walkthroughs and a real presence in its category's subreddit than from a daily posting cadence on a platform its buyers never open. Instagram and TikTok can wait unless your category already lives on them, which for a fashion or food brand it very well might.

The engines don't agree on which social source to trust

The costliest mistake in social advice right now is treating AI as one audience, when each engine has a different taste in social sources and optimising for the wrong one wastes the budget behind it. ChatGPT leans heavily on discussion and professional posts, so that of its social citations, Reddit accounts for about 53.6% and LinkedIn 25.9% while YouTube barely registers at 9.5%, and LinkedIn is influential enough there that it supplies roughly a third of ChatGPT's sources on how-to questions.

Perplexity and Google's AI surfaces pull in the opposite direction and reach for video first, with Perplexity taking about 22.7% of its social citations from YouTube and Google AI Overviews making it the top-cited domain overall. The same LinkedIn how-to content that ChatGPT loves shows up in only about 22% of Google's answers for those queries, which tells you the two engines are effectively reading different rooms. Because of that split, the right sequence depends entirely on who you are trying to reach, so a brand whose buyers research through ChatGPT gets more from an active Reddit presence and a steady stream of LinkedIn how-tos than from a video channel, while one whose customers ask Perplexity or lean on Google's AI answers should put YouTube at the front of the plan. Very few teams can fund all of it at once, and choosing by engine is how you avoid paying for citations that never surface where your buyers actually search.

What being present looks like across different businesses

The shape of this work changes a lot depending on what you sell, even though the underlying principle holds steady. A hotel earns its place in "is this good for families with young kids" answers through recent guest reviews and a couple of honest room-tour videos far more than through anything on its own booking page, because the model trusts a guest describing the pool at seven in the morning over a marketing line about being family friendly. A fintech tool aimed at small brokerages wins the "does this work for a three-person firm" question with a founder-led walkthrough and a candid answer in the subreddit where those brokers already compare notes.

Ecommerce sits at a different point in the journey, where Instagram earns most of its citations at the moment someone is close to buying and wants to see the product in use rather than in a catalogue photo. A B2B services firm, by contrast, lives on LinkedIn and Reddit, where a partner explaining how they actually ran a migration will out-cite a glossy case study every time. None of these require a bigger following, only content that answers the specific question a buyer in that category is asking, which is a very different brief from posting for reach.

How to write a video an AI will quote

Getting a video cited has less to do with production value than with answering a real question in a form a model can lift cleanly out of the transcript. The single change that helps most is to reframe the title as the query itself rather than a brand slogan, so that "How to cut cloud storage costs by 30%" will earn citations that a clever campaign name never will, a point both MediaPost's citation-hierarchy analysis and Search Engine Land's YouTube guidance arrive at independently. From there, write the description as a plain summary of who the video is for, the problem it solves, and the steps it covers, and say the answer out loud in the first minute so it actually lands in the transcript the model reads. None of this depends on being a big channel, since in the same 100-million-citation study 41% of the YouTube videos AI cited had fewer than 1,000 views, because the model is grading how well the content answers the question rather than how large the channel behind it is.

Timestamps are the one tactic that comes with a real caveat, because chaptered and timestamped citations show up almost entirely on Google, at 73% within AI Overviews and 27% within AI Mode, and essentially never on ChatGPT, Gemini, Copilot, or Perplexity. If Google's AI answers are where your buyers end up, chapter every video and label each section as the question it answers, but if your audience lives on ChatGPT or Perplexity, that same hour is better spent on a cleaner transcript and a sharper title. Take the warehouse-software company again, which might title a clip "Setting up inventory sync in 10 minutes", open by saying plainly that it is for teams without a dedicated admin, and chapter each configuration step, so that whether the buyer asks Google or Perplexity the video answers the exact question in the first line it shows.

Being present is the pillar you cannot shortcut with better copy, because it is earned in the places your buyers already trust, which is also why it keeps working as the engines change their sources from one quarter to the next. If you want to see where AI is pulling your category's answers from today, and which of the three pillars is holding you back, that is the work our AI search practice is built to do.