If you are a marketing director or media planner deciding where next year’s content budget goes, you have probably seen a headline this year telling you YouTube is now the most-cited source in AI answers. You may also have seen one saying it is a rounding error. Both were published in 2026. Both are about YouTube. They disagree by more than tenfold, and neither one published an uncertainty range. Here is what our own data says, and why the honest answer changes your media plan more than either headline does.

TL;DR

• The published YouTube numbers disagree by about 13x. One 2026 study puts YouTube at roughly 1.8% of all AI citations. Another puts it at 23.3% of Google AI Overview citations. Neither reports a confidence interval.

• In our own August run, YouTube was the single most-cited domain. 11.5% of all citations (95% CI 9.5–13.5) across 1,960 citations in one category.

• That blended number is misleading, and so is everyone else’s. Split by engine, YouTube took 23.1% of citations on Google’s AI answers (19.2–27.2) and was effectively absent from ChatGPT’s.

• G2 is the exact mirror image. 22.7% on ChatGPT (15.8–29.7), 0.1% on Google. Two domains, both averaging around 11%, neither behaving like an 11% domain anywhere.

• You cannot buy one video strategy for “AI search.” There is no such channel. There are engines, and they source differently enough that a blended plan will underfund one and overfund the other.

The two numbers that cannot both be right

Start with what is already published, because it is a good illustration of the problem before we get to our own data.

Otterly AI ran a large study, published March 2, 2026, covering more than 100 million citation instances over 30 days across ChatGPT, AI Overviews, AI Mode, Perplexity, Copilot, and Gemini. Social platforms accounted for 5.54% of all citations. YouTube took 31.8% of that social slice. Do the arithmetic and YouTube lands at roughly 1.8% of all citations. Reddit beat it inside the social category at 46.4%.

5W Public Relations published a citation share report the same year putting YouTube at 23.3% of Google AI Overview citations, drawn from 46 million citations synthesized from six other vendors’ datasets.

Roughly 1.8% against 23.3%. Some of that gap is denominator: all citations across six engines is not the same universe as AI Overview citations only. That is precisely the point. Neither number is published with the denominator, the engine split, and the uncertainty range together, which is what you would need to plan against either one. Neither reports a confidence interval at all. And 5W, who published the larger number, sells citation share audits, which is worth knowing when you read it.

We are in the same position. We sell in this category too. So here is our data with the parts that make it checkable.

What we measured

At the end of August we ran our own category as a measurement program. 123 distinct buyer queries about AI visibility tools, run out to 250 responses, producing 1,960 citations across 525 distinct pages and 248 domains. The run converged on both of our stability conditions: the rank ordering of cited domains had settled into a plateau, and the established domains were far enough apart to be told from each other. It cleared the second condition by a wide margin, not by a whisker, which is why we are comfortable publishing the splits below.

One category, one window, one set of buying queries. That is the honest scope, and we will come back to what it does and does not license you to conclude.

Here is the leaderboard as most dashboards would show it to you.

YouTube first, ahead of every trade publication, every review aggregator, every vendor blog. Roughly five times Reddit’s share in this category. If you stopped reading here you would go build a video program, and you would be about half right.

The blended number is the problem

Now split the same data by engine.

Read those rows twice. YouTube is not an 11% domain. It is a 23% domain on one engine and close to nothing on the other. G2 is not an 11% domain either. It is a 23% domain on the opposite engine and a rounding error on Google. Both averaged to about the same blended figure, and the blended figure describes neither one accurately in either direction.

This is the failure mode we keep running into across client categories, and it is not a YouTube story. It is an arithmetic story. When a dashboard pools across engines without weighting, the number you read is an average of two behaviors that never occur together, and it is wrong for every actual buyer session. Nobody queries “AI search.” They query Google, or ChatGPT, or Perplexity, and they get sourcing patterns that barely overlap.

Our own July analysis across a much larger query pool, 147,934 queries dated July 31, pointed the same direction on video routing. We are treating that figure as internal until we publish the method, and you should discount it accordingly until we do.

Marketer-Side Impact

Take a $600,000 annual content budget with a video line. Planned off the blended 11.5%, YouTube reads as one important source among several, and video gets maybe 15% of the budget, spread evenly. Planned off the engine split, video is the single largest lever on Google’s AI answers in this category and close to irrelevant on ChatGPT’s. If your buyers skew to one engine, the blended plan misallocates in the range of $90,000 to $150,000 a year, in both directions at once: underfunding video for the Google-heavy audience, overfunding it for the ChatGPT-heavy one. At roughly $750 per briefed content piece, that is 120 to 200 pieces pointed by an average that describes nobody. And the failure is invisible for two quarters, because the blended number keeps reading as stable while the underlying engines diverge.

What this actually means for the media plan

Three things follow, and only three.

One. Stop planning against blended AI visibility numbers. Ask your vendor to split every figure by engine before you allocate. If the product cannot do it, you are planning against an average of behaviors that never co-occur.

Two. Match the channel to the engine your buyers use. In this category, Google’s AI answers routed heavily through video and ChatGPT routed heavily through review aggregators. Those are different production budgets, different teams, and different lead times. A video program and a review-generation program are not substitutes.

Three. Do not port our numbers into your category. This is one category, measured in one window. YouTube’s 23.1% on Google is a finding about AI visibility buying queries in August 2026. Your category will have its own structure, and the only way to know it is to measure it. More prompts alone will not fix a blended read, and neither will broader engine coverage, faster refresh cycles, or overlaying agentic workflows on top of an average.

I will admit the part I got wrong. For most of last year I treated the engine split as a reporting nicety, something to add once the core numbers were solid. It is not a nicety. In this run it was the whole finding, and the all-engine leaderboard would have sent a competent planner in a defensible, expensive, wrong direction.

Three questions for your next media planning meeting

1. “What does this number look like split by engine?” If the answer is a single blended figure, you do not yet have a media planning input.

2. “Which engine do our actual buyers use, and how do we know?” Analytics referral data, sales call notes, and win-loss interviews all carry signal here. The answer changes the channel mix more than any content tactic will.

3. “If we re-ran this measurement next month, how much would the engine split move?” Sourcing patterns shift when engines change their retrieval behavior. A split measured once is a snapshot, not a plan.

Todd Paris is a co-founder of IQRush. Chinmay Dhamapurkar works on IQRush’s measurement pipeline.

One other note, a falsifiable version: we will re-run this exact category in January 2027 and publish the engine split again. If YouTube’s share on Google’s AI answers has not stayed above 15%, the pattern was a moment rather than a structure, and we will say so on this site.

Book a walkthrough →

Published October 13, 2026. Measurement window: August 25, 2026.

Sources: Otterly AI YouTube citation study · 5W citation share report