People have changed how they search. They ask ChatGPT a question and read one answer. They see a Google AI Overview and stop scrolling. Research from Frase shows that when an AI Overview appears, clicks on regular search results drop by 61%.
That sounds like bad news for anyone who writes online. The full picture says something different. Pages cited inside those AI answers gain traffic. A Seer Interactive study found that brands appearing in Google’s AI Overviews earned 120% more organic clicks per impression than brands left out.
So one question now decides who gets found. What makes an AI engine cite your page?
I went through the published research on this. Studies, not opinions. Here is what the data shows.
AI engines read the top of your page first
AI systems do not treat every part of a page the same way. Work by Dan Petrovic found that Google’s Gemini pulls a limited amount of text from each URL. Content near the top makes the cut. Content buried under a long intro often never gets read. A separate Search Engine Land analysis found that 44% of ChatGPT citations come from the first third of a page.
Cyrus Shepard reviewed 54 experiments and case studies on AI citations in his May 2026 analysis. Placing the answer near the top scored 8.8 out of 10 as a citation factor. It was one of the strongest signals he found.
The lesson is simple. Answer the question in your first two paragraphs. Save the backstory for later, or cut it.
Numbers earn citations
Researchers at Princeton, Georgia Tech, the Allen Institute for AI, and IIT Delhi ran the first peer-reviewed study on this topic. It is called the GEO study, presented at KDD 2024. They tested which changes made content more visible in AI answers.
The single most effective change was adding statistics. It improved AI visibility by 41%.
This makes sense once you think about how these systems work. An AI engine wants claims it can repeat safely. A specific number from a named study is safe. A vague opinion is not.
Named sources build trust
Shepard’s analysis also scored pages that cite their own sources. Linking to studies and naming where a claim comes from scored 8.0 as a citation factor. Pages that show their evidence get picked. Pages that assert without proof get skipped.
Self-contained passages scored 8.0 as well. That means each section of your page should make sense on its own. An AI engine may lift one paragraph, not your whole article. Write each paragraph so it can stand alone.
Popularity is not the filter
Here is a finding that should encourage small publishers. ZipTie reviewed AI citation data and found that 40.83% of AI-cited YouTube videos had fewer than 1,000 views. View count showed almost no link to citation odds.
AI engines look for reference-quality content. Clear, sourced, and specific beats viral. A small site with strong data can outrank a big site with thin posts.
For B2B, your own pages matter most
A 12-week study by XFunnel tracked which content types ChatGPT, Google AI Overviews, and Perplexity cite. For B2B questions, company product and service pages made up close to 56% of citations. Blog content earned only 3 to 6%.
This changes how B2B service pages should be written. Your services page needs direct answers, clear pricing signals, FAQs, and specifics. It is now a citation target, not just a sales page.
Being known helps too
Research by The Digital Bloom found that brand search volume predicts AI citations better than backlinks, with a 0.334 correlation. AI models favor names people already search for.
Where you show up matters as well. A Peec AI analysis of 30 million cited sources found LinkedIn among the top three most-cited domains in AI answers, behind Reddit and YouTube. Your LinkedIn presence now feeds your citation odds.
Reddit is winning right now. One category is beating it anyway.
A recent study looked at 8,566 B2B search terms across 14 domains (13 SaaS vendors plus Reddit) in four verticals. Reddit outranked every single software vendor at the same time on more than half of them. In sales tech, it was two out of three searches.
Most people read that and think Reddit ate their traffic. The real story is smaller and more useful.
One category did not lose. Business phone software. Companies like RingCentral and Nextiva kept beating Reddit on the same searches, in the same industry, under the same Google algorithm.
The difference was not budget. It was not brand size. These companies wrote real answers to real questions years before anyone asked them to. Glossaries. “How to choose” guides. Plain explanations of how the product category works, not just their product.
Their competitors wrote product pages and gated PDFs. Product pages sell. They do not answer questions. So Google, and now AI search, skip past them and go straight to a stranger on Reddit who actually explained something.
This is not a Reddit problem. It is the same problem AI search creates for everyone. Machines are not choosing content because it is polished. They are choosing content because it answers the question first, in plain language, with something specific behind it.
What I do with this in client work
Here is the checklist I apply to every article and service page I write:
- Answer the main question in the first 100 words
- Add at least three real statistics with named sources
- Write each section so it stands alone
- Use headings phrased the way people ask questions
- Cut warm-up intros
- Put the author name and credentials near the top of the page
- Keep sentences short so extraction is clean
None of this is new to me. I have written research-backed content for seven years. AI search rewards the habits careful writers already had. Sourced claims. Clear structure. Direct answers. The fast, thin content that flooded the internet loses under these rules. Researched writing wins.
The takeaway
AI search did not kill content. It raised the bar for it. If your pages answer fast, cite real data, and stand on named sources, AI engines will pick them up. If your pages open with fluff and assert without proof, they will stay invisible.
I offer content written to be AI-citable as a service. If you want your pages showing up in AI answers instead of disappearing under them, get in touch.
Sources: Princeton/Georgia Tech GEO study (KDD 2024), Cyrus Shepard/Zyppy Signal (May 2026), Seer Interactive, Search Engine Land, XFunnel, Peec AI, The Digital Bloom, ZipTie, Dan Petrovic, Frase, Foundation Inc./Ross Simmonds (Reddit B2B SaaS study, July 2026).