The 2026 AI Citation Report: Who AI Search Engines Actually Cite

What we found

Everyone wants to know who AI search engines actually cite. We combined the major 2026 studies with our own tracking data to answer it, and the picture is consistent and a little uncomfortable for most brands. AI engines overwhelmingly cite communities and user-generated content, not company websites. They barely agree with each other about what to cite. And the old assumption that ranking first on Google gets you into the AI answer is no longer true.

The four findings that matter most:

• Community sources dominate. Reddit, YouTube, and forums get cited far more than brand-owned pages. In our own Google AI Overviews data, Reddit alone took 47.8 percent of all citations.
• The engines disagree with each other. Independent studies find only 11 to 14 percent of cited sources overlap between platforms. ChatGPT leans on Wikipedia, while Google and Perplexity lean on Reddit.
• AI citations no longer follow Google rankings. The share of AI Overview citations coming from Google's top 10 fell from 76 percent to 38 percent in under a year.
• Getting cited is not the same as getting traffic. When an AI summary appears, only about 1 percent of searches end in a click on a link inside it.

How we put this together (and what to trust)

This report has two data sources, and we want to be clear about both before you read another number.

The first is the published research: large studies from Ahrefs, Profound, Semrush, SE Ranking, Pew Research Center, and the Columbia Journalism Review's Tow Center. Every figure we quote from them is linked at the end, with the sample size and date, so you can check it yourself.

The second is our own data, from the AI search tracking inside SEOitis. We query real AI engines on our customers' behalf and record which sources they cite. The numbers we share here are aggregated and anonymized: a source only appears once it has been cited across multiple customer sites, and we never expose any individual customer, prompt, or URL. Three honest caveats about our data:

• It is early. This snapshot covers roughly two weeks (June 8 to June 22, 2026). We will update it as the dataset grows.
• It is a niche sample. Our customers skew toward software, SaaS, and indie tech, so our prompts ask the kinds of questions that audience asks. That shapes what shows up. It is not a representative sample of the whole web.
• Only two engines have enough volume yet. Google AI Overviews and ChatGPT cleared our reporting threshold. Claude, Perplexity, Gemini, and Grok did not have enough cross-customer data to report responsibly, so we are not going to invent numbers for them.

So treat our data as a sharp, current signal from one corner of the web, and the published studies as the broad, web-wide picture. Where the two agree, you can be confident. Where our niche data diverges, we will say so and explain why.

Who gets cited most

Start with the web-wide picture. Ahrefs analyzed more than 3 million US queries and found Google AI Overviews citations heavily concentrated in a handful of giant platforms.

Profound, looking at 680 million citations, found the same names but with a per-engine twist worth remembering: ChatGPT leans hardest on Wikipedia (which makes up almost half of its top-10 sources), while Perplexity leans hardest on Reddit (almost half of its top-10). The platforms are the same. The mix is different on every engine.

Now our data. Over the two-week window, Google AI Overviews cited 50 distinct sources across 2,385 citations in our sample. Here is the top of that list.

Those top seven sources account for 88 percent of every AI Overview citation in our data, and all seven are community, video, code, or social platforms. Not one is a traditional brand marketing site. This is the single clearest pattern in AI search: the engines trust places where real people talk, build, and review far more than they trust company copy.

One number jumps out. In our sample Reddit took 47.8 percent of AI Overview citations, more than double the roughly 20 percent Ahrefs found web-wide. That gap is not a contradiction, it is the niche showing through. Our prompts ask software and SaaS questions ("best tool for X", "is Y any good"), and those are exactly the queries where Google leans hardest on Reddit threads full of real user opinions. If your audience asks product-comparison questions, Reddit is not one lever among many. It is the lever.

The engines do not agree with each other

The most expensive mistake in AI search is treating "AI" as one channel. It is not. Profound compared 100,000 prompts and found ChatGPT and Perplexity share only 11 percent of their cited domains. Ahrefs compared Google's own two surfaces, AI Overviews and AI Mode, across 540,000 query pairs and found just 13.7 percent of cited URLs overlap, even though the two reach 86 percent semantically similar conclusions. They arrive at the same answers from almost entirely different sources.

Our data shows the same split, even at small scale. Here are the top sources for the two engines we can report, side by side.

AI Overviews is a Reddit-first engine. ChatGPT, in our data, is a reference-and-directory engine, led by Wikipedia, the software directory SaaSHub, and the tech-media site TechRadar. Only YouTube and dev.to appear on both lists. (Our ChatGPT sample is small, 127 citations across 8 sources, so treat it as directional. The direction matches every large study: different engine, different sources.)

The practical takeaway: a strategy that wins you Reddit threads might do nothing for your ChatGPT presence, and vice versa. You have to track and work each engine on its own terms.

Ranking on Google no longer means getting cited

For years the assumption was simple: rank in the top 10 and you will show up everywhere that matters. AI search broke that assumption, and the data is stark. Ahrefs tracked the share of AI Overview citations that come from Google's organic top 10 and watched it fall from 76.1 percent in July 2025 to 37.9 percent by March 2026. In other words, roughly five of every six AI Overview citations now come from pages that are not on Google's first page for that query.

A separate Ahrefs study of 15,000 prompts across ChatGPT, Gemini, and Copilot found that only about 12 percent of AI-cited URLs rank in Google's top 10 at all, and around 80 percent rank nowhere on the first page. Perplexity was the one outlier that still tracks Google rankings closely, at roughly 29 percent. For every other engine, AI visibility and Google ranking are now largely separate games.

What the cited sources have in common

Pull the patterns together and a clear profile of a "citable" source emerges.

They are third-party, not self-published

Across studies and in our own data, the large majority of AI citations go to sources other than the brand being discussed. People are not cited because they described themselves well. They are cited because other people, on Reddit, YouTube, forums, and review sites, talked about them. This is the hardest pattern for marketers to accept and the most important one to act on.

Brand mentions beat backlinks

Ahrefs studied 75,000 brands and measured what correlates with AI visibility. Branded web mentions correlated far more strongly (around 0.66) than backlinks (around 0.22). Domain Rating, the classic SEO authority metric, was a weak predictor (around 0.27 to 0.33). The signal AI engines respond to is being talked about, not just being linked to.

The indie launch-platform pattern (from our niche)

Here is something the big, web-wide studies miss because they average across everything. In our software-and-SaaS sample, a specific cluster of sources punches well above its weight: IndieHackers, DevHunt, Uneed, SaaSHub, and AppSumo all appear repeatedly across both AI Overviews and ChatGPT. These are launch directories and founder communities. If you are building a product for a technical or founder audience, being present and well-reviewed on these platforms is a concrete, under-discussed way to show up in AI answers. It will not appear in a generic citation report, but it is real in the niche.

Freshness and structure help

Two smaller but well-supported factors: AI-cited pages tend to be newer than typically-ranked pages (Ahrefs found AI-cited content about 26 percent fresher across 17 million citations), and content that is easy to extract (clear answers near the top, clean tables, plain HTML) gets pulled more readily than long narrative. For the full playbook on this, see our guide on getting cited in Google AI Overviews.

The catch nobody puts in the headline

Being cited by AI is worth chasing, but go in clear-eyed about three things the data shows.

Citations do not equal clicks. Pew Research Center studied 900 US adults across 68,879 searches and found that when an AI summary appears, users click a result link only 8 percent of the time (versus 15 percent without one), and just 1 percent of searches end in a click on a link inside the summary. AI search is increasingly an answer-in-place experience. Being the cited source is often about influence and brand presence, not a flood of referral traffic.

AI engines are frequently wrong about their sources. The Columbia Journalism Review's Tow Center tested 1,600 queries across eight engines and found incorrect source attributions more than 60 percent of the time. Perplexity was the most accurate (37 percent wrong) and Grok the least (94 percent wrong). ChatGPT misattributed most of the articles it was given and almost never declined to answer. Citation does not guarantee accuracy, so monitoring how you are described matters as much as whether you appear.

It moves constantly. Semrush documented ChatGPT's Reddit citations crashing from around 60 percent to around 10 percent within weeks in late 2025 as OpenAI adjusted its systems. Any "most-cited source" number is a snapshot, including ours. The only durable response is continuous tracking, not a one-time audit.

How many sources does an AI answer cite?

It varies enormously by engine, which matters when you think about your odds of being one of the cited sources. Qwairy analyzed 118,101 answers and found the average number of citations per answer ranged from very generous to very stingy.

Perplexity citing almost 22 sources per answer means more room to get in. Copilot citing fewer than 3 means the bar is brutal. The same content can have very different odds depending on where the question is asked.

What to actually do with this

Five takeaways that follow directly from the data:

• Get into the community sources. Reddit, YouTube, and relevant forums are where AI looks first. Genuine, helpful presence there (not spam) is the highest-leverage citation work you can do.
• Earn mentions, not just links. Because brand mentions out-predict backlinks, PR, reviews, comparisons, and being included in third-party roundups matter more than ever.
• Work each engine separately. They cite different sources and drift over time. What wins Google AI Overviews will not automatically win ChatGPT.
• Mind your niche's specific platforms. For software and SaaS, that means launch directories and founder communities. For your field it may be different. Find where AI pulls from in your category.
• Track it continuously. The data moves week to week. You cannot manage what you do not measure. This is exactly what our AI visibility tools and our ongoing tracking are built for, and the live source data behind this report lives on our AI citation data pages.

One thing this report does not recommend, because the evidence does not support it: do not expect a magic file to fix your AI visibility. We dug into that claim separately in our piece on whether llms.txt actually gets you cited. The honest answer is that citations are earned in the sources above, not declared in a text file.

We will keep updating this

Our dataset is young and growing. As more engines and more sources clear our anonymization threshold, we will expand this report to cover Claude, Perplexity, Gemini, and Grok with the same honesty. When we have those numbers, we will publish them. Until then, we would rather show you two engines we can stand behind than seven we cannot.

Sources

• Ahrefs: most-cited websites in AI Overviews, 3M+ US queries, 2026.
• Profound: AI platform citation patterns, 680M citations, 2025.
• Profound: cross-engine citation overlap, 100K prompts, 2025.
• Ahrefs: AI Overviews vs AI Mode overlap, 540K query pairs, December 2025.
• Ahrefs: AI Overview citations from the top 10, 863K SERPs, March 2026.
• Ahrefs: AI-cited URLs vs Google rankings, 15K prompts, August 2025.
• Ahrefs: AI brand visibility correlations, 75K brands, December 2025.
• Qwairy: citations per answer by engine, 118K answers, Q3 2025.
• Pew Research Center: clicks and AI summaries, 68,879 searches, July 2025.
• Columbia Journalism Review (Tow Center): AI search citation accuracy, 1,600 queries, March 2025.
• SEOitis first-party AI search tracking, June 8 to June 22, 2026 (anonymized, aggregated; Google AI Overviews and ChatGPT).