Why ‘Reddit is 40% of AI citations’ tells you nothing about your brand

Here is the most quoted AI citation statistic of the last year: Reddit is 40.1% of AI citations, per a Semrush analysis of 150,000 citations from June 2025. Wikipedia is 26.3%. YouTube is 23.5%. I have seen that line in roughly every AI search deck I have been sent, and it is a real number. It is also useless for the one query you actually have, which is whether your brand shows up when your buyer asks ChatGPT or Perplexity to recommend something in your category.

That gap is the whole problem with measuring AI visibility from category stats. An aggregate domain share is the average across every brand, every query, every industry on the internet. Reddit being a huge slice of citations and your brand being absent from AI answers are both true at once. The threads get cited. You are just not in them.

This post is about why brand AI visibility cannot be read off a leaderboard, why the published studies disagree with each other by an order of magnitude, and how to get the one number that matters by hand in an afternoon. If you want the background on the discipline itself, start with what generative engine optimization is.

The category stat is the weather, not the forecast for your house

A national weather report tells you it rained somewhere today. It does not tell you whether you need an umbrella to walk to your car. Both things are true, and only one of them changes what you do this morning.

AI citation statistics work the same way. “Reddit is 40% of citations” is the climate of AI search. It is genuinely interesting, and it points at a real shift in where answer engines pull from. But your brand does not live in the aggregate. It lives in a few dozen specific queries your buyers type, answered by specific engines, on a specific day. The category stat is the weather. Your own check is the forecast for your house.

The math underneath is simple. A domain share counts how often a source shows up across millions of answers. It says nothing about which brands those answers name. Reddit can supply 40% of the citations behind a million AI responses, and your brand can appear in none of them. The two numbers never touch.

Four studies, four different Reddit shares

If a category stat were a hard fact about your situation, the studies measuring it would agree. They do not. Here is Reddit’s reported share of AI citations from four credible sources, all published in roughly the same window:

Three percent. Twelve percent. Forty percent. Sixty percent. None of these is a mistake. They differ because they measured different engines, different time windows, different denominators, and counted “citation” in slightly different ways. Semrush looked at three engines in one window. Profound aggregated more than four billion citations across every engine including Copilot, where Reddit ranks 31st. The 5W figure stitches together nine separate datasets.

The most telling number is in that last row. The same publisher, Semrush, watched Reddit’s share of ChatGPT responses move from close to 60% to around 10% in about two weeks around September 2025. Their own research notes it “doesn’t tell us exactly why these shifts happened.” If a single source can swing 6x in a fortnight, no static “Reddit is X%” headline survives the next data release. The number you screenshotted last quarter is already stale.

This is the same reason a single check of any AI answer is a vanity metric. The answers are non-deterministic, so you measure a panel of prompts and report a range, not one screenshot. I get into that argument properly in GEO vs SEO, but the short version: the instability you see across these four studies is the same instability you get when you run one prompt twice in a row.

It cuts both ways: a tiny share can be your kingmaker

The disagreement is not just noise. It is the signal. A source that looks irrelevant in the aggregate can be the exact one AI pulls for your niche, on the engine your buyers actually use.

The clearest evidence I have seen comes from the Tinuiti Q1 2026 AI Citations Trends Report, which tracked nine commercial categories across seven AI platforms. Look at how far Reddit’s share spreads depending on which engine you ask:

• 44% of Google AI Overviews’ social citations
• 24% of all Perplexity citations in January 2026
• over 5% of ChatGPT citations
• 0.1% of Google Gemini citations

That is one source ranging from nearly half of an engine’s social citations to a rounding error on another. The same report found Reddit’s citation share grew at least 73% across all tracked categories from October 2025 to January 2026, and more than doubled in some industries. So “Reddit is 40%” is not even stable as a category fact. It moves with your engine, your vertical, and the month.

Flip it around and the practical lesson lands. Say a niche directory or one comparison thread supplies 0.4% of all citations on the internet. In your category, on the engine your buyer uses, that one page might be where the answer comes from every single time. The aggregate buries it. Your own check surfaces it. That specific page, the one AI keeps citing to recommend a competitor instead of you, is what we call a kingmaker source. Finding yours is the entire point of citation readiness work.

Why no published study has your number

People keep waiting for the study that hands them their answer. It is not coming, and the reason is structural, not lazy. Your number sits at the intersection of three things no aggregate can hold:

• Your brand. Whether AI names you, specifically, in the answer text. A domain leaderboard counts sources, not the brands those sources push AI to recommend.
• Each engine on its own. ChatGPT, Perplexity, Gemini, and Copilot diverge for a mechanical reason. Google’s surfaces are grounded on Google’s index, so classic SEO carries over. ChatGPT leans more on training-data weight plus a thinner retrieval layer. Perplexity leans on real-time retrieval that favors Reddit and Quora. Same brand, same query, a different answer and a different fix per engine.
• Your buyers’ real queries. Not “best CRM” in the abstract, but the comparison, problem, and feature phrasings your actual buyers use, at the point where they are ready to choose.

No study can publish that, because it is the cross-product of your name, your engines, and your buyers. It is specific to you by construction. The upside is that this is exactly the kind of thing you can measure yourself, and the manual version costs nothing.

How to get your own number by hand

You do not need a tool to start. You need an afternoon and a willingness to read the answer instead of the links under it. Here is the manual pass I run before I trust any paid number.

• Write 5 to 10 real buying queries. Use the phrasings a ready buyer would type, not your brand name. Mix category queries like “best outstation cab service in India” with comparison ones like “X vs Y for small teams” and a few problem queries. If you have Google Search Console, pull the low-CTR long-tail queries you already get impressions for. Those are the closest free thing to a real prompt set, because they are actual demand, not invented phrasings.
• Run each across the engines your buyers use. ChatGPT, Perplexity, Gemini, Copilot. Run every query two or three times, because the answers vary and one run is a coin flip.
• Read the answer text, not just the citations underneath. This is the step almost everyone skips. A page can sit in the source block while your brand never appears in the sentence the user actually reads. Sourced is not the same as presented. Note whether your brand is named in the answer itself.
• Tally four outcomes per query. Named and linked, named but not linked, linked but not named, or absent. Those four states are the difference between a content-structure fix and an off-site placement fix, and a single blended score hides which one you need.

Do this once and you will know more about your real brand AI visibility than any “Reddit is 40%” headline could tell you. For the structured version of the same pass, including how to organize the results, see our guide on tracking brand mentions in AI search. And if your name never shows up at all, the diagnosis usually starts with why your website is invisible to AI.

Be honest about what this check cannot do

A manual pass is the right first move, but it has two honest limits, and pretending otherwise is how you lose people’s trust.

First, it is point-in-time. You are reading the weather on the day you check, not a continuous feed. Run the same queries next month and some answers will have moved. That is fine, as long as you treat each pass as a snapshot and re-run on a cadence you pick, rather than assuming today’s result is permanent.

Second, you cannot see the clicks AI sends you, because mostly there are none to see. Bain & Company found that about 60% of searches now end without the user going to another site, and that around 80% of consumers lean on these zero-click results at least 40% of the time. Ahrefs measured a 34.5% drop in clicks to the top organic result when an AI Overview is present, across 300,000 keywords. If the buyer reads your brand in the answer and never clicks, there is no analytics trace of it. This is a brand-recall game, and I cover the traffic side of it in the zero-click crisis in AI search. Anyone selling you AI-referral attribution is selling a number that mostly does not exist.

Want the four-state version of your own number without running all 400 checks by hand? Run a free audit at app.bluejar.ai and see, prompt by prompt, where your brand is named, cited, or invisible across ChatGPT, Perplexity, Gemini, and Copilot.

What to actually do with the category stat

None of this makes the leaderboards worthless. The category stat is real climate data, and you can use it the way you use a forecast: to decide what to pack, not whether you personally got wet.

If Tinuiti shows Reddit driving 24% of Perplexity citations and your buyers live on Perplexity, that tells you Reddit threads are a place worth being present and credible. It does not tell you that you currently are. The aggregate sets your priors about where to look. Your own check confirms whether the work is done. Use the stat to pick the battlefield, then measure your own brand to see if you are winning it.

The honest summary is short. Category citation statistics answer “where does AI pull from, on average.” Your own check answers “does AI name me, here, for the queries my buyers ask.” Only the second one changes what you do on Monday.