Everyone still plans content around Google. The keyword brief, the content calendar, the "will this rank" gut check — all of it assumes that if a page shows up on page one, it's also doing the job for AI search. Ahrefs just tested that assumption at scale. The firm's new research series, "1 Billion Data Points, 14 Studies," led by Tim Soulo with contributions from Ryan Law, found that a large share of what ChatGPT cites never shows up in Google's results at all, that most AI-cited pages are short listicles rather than deep guides, and that video presence predicts AI visibility better than backlinks do. If you're still building your AI search content strategy on top of a rank tracker alone, you're missing more of the picture than you think.
The visibility overlap is smaller than you think
Start with the headline number. Ahrefs analyzed a massive sample of ChatGPT's most-cited pages as part of its "1 Billion Data Points, 14 Studies" research series, led by Tim Soulo with contributions from Ryan Law, and found that 28.3% of those pages have zero Google organic visibility. Not "ranks on page three." Not "gets a trickle of impressions." Zero. These are pages that would not turn up if you typed the same query into Google, yet ChatGPT pulls them into its answers on a regular basis. That's more than one in four of the pages doing the heaviest lifting in AI-generated responses. If your content strategy treats organic rank as a reliable proxy for AI citation, more than a quarter of the citation opportunity is invisible to you by definition, because you're only measuring the metric that predicts three-quarters of the picture.
This breaks the working model most content teams still operate under, which goes roughly: rank well in Google, get cited by AI models as a downstream byproduct. The data says the byproduct theory doesn't hold for a meaningful share of citations. Some of what gets pulled into ChatGPT's answers is coming from pages that never had to fight through a competitive results page, never had to clear Google's ranking factors, and never accumulated the backlink profile or domain history that traditional SEO optimizes for. They got cited some other way — through tight topical match, through a structure a model can lift cleanly, through being crawlable and indexed even without earning search visibility. That's a different game than the one most editorial calendars are built to win, and it means citation tracking has to run as its own discipline, not as a Google Search Console side effect.
Most reporting stacks aren't built to catch this. If your dashboards pull from Search Console and a rank tracker, a page with zero Google visibility simply doesn't exist in your view of performance, even while it's actively shaping what an AI model tells a prospective buyer about your category. That's the practical risk in the 28.3% figure: it's not just a statistic about Ahrefs' dataset, it's a description of a blind spot most teams are currently running with. Closing it means adding a citation-tracking layer that doesn't depend on organic rank as its trigger — a separate watch list of pages worth defending because they show up in AI answers, whether or not Google ever sends them a click.
What actually gets cited: format and length
Format matters more than most editorial calendars admit. Ahrefs found that 43.8% of AI-cited pages are in a "best X" listicle format — best tools, best services, best options within a category — rather than comprehensive guides or long-form explainers. Pair that with the length finding: 53% of AI-cited pages run under 1,000 words. Put those two numbers together and the picture is clear. The single most common shape of a page that gets cited by ChatGPT is a short, ranked or grouped list, not a deep, exhaustive resource. That's uncomfortable for teams that have spent the last few years building longer and longer content on the theory that depth signals authority. Depth still has a role. It's just not the role most AI answers are asking a page to play.
| FINDING | SHARE OF AI-CITED PAGES | WHAT IT TELLS YOU |
|---|---|---|
| "Best X" listicle format | 43.8% | Structure beats depth for extraction |
| Under 1,000 words | 53% | Length isn't the differentiator you'd assume |
| Zero Google organic visibility | 28.3% | Citation and ranking are separate outcomes |
None of this means long-form content is dead — it means format needs to match the job a page is doing. A model answering "best password managers for small teams" is going to reach for a tight, clearly labeled list before it reaches for an eight-section guide with the actual list buried in section four. We've written before about which content formats AI engines actually cite most, and this new dataset backs the same conclusion from a different angle: structure is doing more of the extraction work than word count is. For the length question specifically, because it behaves differently engine by engine, see our related breakdown of why short content wins in ChatGPT and long wins elsewhere — a related but distinct finding from this one.
The practical implication is that a lot of existing pillar content is structured for the wrong reader. Humans skim; models extract. A guide written as a narrative, with the useful comparison or ranking woven into paragraph four, forces a model to do interpretive work it would rather skip in favor of a competitor's page that hands the list over cleanly at the top. That's not a reason to gut every long guide on your site. It's a reason to check whether your highest-value pages present their core answer in a form a model can lift in one pass, before you assume word count is the problem.
Run this test on your own site before you decide it's a coincidence. Take the pages you'd bet are getting cited by ChatGPT — pricing comparisons, tool roundups, "how to choose" pages — and check two things: current word count and whether the core answer sits in the first couple hundred words. If a page qualifies as a deep guide by any editorial standard but buries its list six sections in, it's fighting the shape of the data Ahrefs just published. That's a fixable problem in an afternoon, not a rewrite of the whole site.
Why YouTube outperforms backlinks for AI visibility
Here's the finding that should worry link-building purists: pages and brands with YouTube mentions or presence correlate more strongly with AI brand visibility than backlink count or Domain Rating do, according to Ahrefs' analysis. In classic SEO, backlinks and Domain Rating are still two of the biggest levers for organic ranking. In this AI citation dataset, a YouTube footprint is a stronger predictor of showing up inside an AI-generated answer. That's a meaningful reordering of priorities for any team that has spent years treating link acquisition as the primary trust signal worth chasing.
That doesn't mean links stop mattering. They still shape what a crawler trusts and what a model has been trained to weight during retrieval. But it does mean video presence is functioning as its own trust and relevance signal, one that link-building alone doesn't replicate. A brand that shows up in comparison videos, tutorial content, or review roundups on YouTube is feeding a different layer of the retrieval stack than a brand that only shows up in text-based backlink profiles. For B2B and SaaS teams especially, most of whom have under-invested in video relative to written content, that's a gap with room to close quickly, and one worth weighing against a straight link-building budget.
This lines up with a broader pattern in how AI models source their answers: they pull from wherever a topic is discussed with the most corroborating detail, not just wherever the highest-authority domain happens to rank. A written page, a YouTube video, and a handful of forum threads covering the same claim reinforce each other in a way a single link-earning campaign doesn't. Treat video as a second, independent evidence source for the claims your written content is already making, rather than a separate content line with its own disconnected calendar.
Practically, this doesn't require standing up a video department. It requires picking the handful of comparison and "how to choose" topics where your written content already exists, and making sure the same claims show up in a video a model can associate with your brand — a demo, a walkthrough, a founder explainer, a customer testimonial. The goal isn't view count. It's giving retrieval systems a second format carrying the same facts your page already makes, so a model has more than one place to confirm what you're claiming before it decides to cite you.
“"Best X" listicles under 1,000 words, with no guaranteed Google visibility, are pulling real weight in AI answers. Most editorial calendars still aren't built to produce that shape of page on purpose.”
What to change in your content plan this quarter
None of this is an argument to abandon Google-first content planning. It's an argument to stop treating it as the whole plan. Pair the moves below with your existing approach to content chunking for AI search so new and reformatted pages are structured for retrieval from the start, not patched after the fact. Here's where to spend the next few weeks.
Google rank still matters. It's still the largest single traffic source for most of the sites we work with, and ignoring it in favor of a citation-only strategy would be its own mistake. But if roughly a third of AI citations are coming from pages with no Google visibility, and more than half of what gets cited runs under 1,000 words in listicle form, your content marketing plan needs a second track — one built for extraction and citation, not just rank. Start with the audit this week. Run the format test on one guide within two weeks. Treat a YouTube footprint as infrastructure for the quarter, not an afterthought you'll get to eventually. The teams that build both tracks now will be the ones showing up in both places next year.
See where you are cited today
A free snapshot audit of your rankings and AI citations before we ever talk.
Josh leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.