5.7% of the sources ChatGPT cites are PDF documents, against 0.32% for Google AI Overviews and 0.42% for Google AI Mode. Across five AI engines, 1.5% of citations are documents rather than web pages, and 57% of the PDFs cited come from government, university or military websites.
Analysis of 155,312 cited URLs · answers from ChatGPT, Gemini, Perplexity and Google's AI features, September 2026
Key Findings
- 1.5.7% of ChatGPT citations are PDF documents, 3.5 times Perplexity's 1.6% and more than 13 times Google AI Mode's 0.42%.
- 2.Google AI Overviews cite a PDF in 0.32% of their citations, the lowest rate of five AI engines, about one citation in 300.
- 3.1.5% of citations in AI answers point to a document file rather than a web page, and nearly all of those documents are PDFs.
- 4.57% of the PDFs cited in AI answers sit on government, university or military websites, led by consumerfinance.gov, noaa.gov and usda.gov.
- 5.PDFs make up 3.5% of citations in AI answers about health, against 0.1% in AI answers about software.
Summary
A family asks ChatGPT what Medicare Part B covers. Among the sources under the answer is a PDF: the government's official Medicare handbook. Ask Google's AI Overviews the same question and the sources are web pages. The two engines treat documents very differently.
Most organizations publish their most detailed information in PDFs: annual reports, product specification sheets, price lists, research papers, policy documents. Whether AI engines cite those files decides whether that information reaches people asking AI, or stays locked in a download.
ChatGPT is the engine that reads them. 5.7% of ChatGPT's citations are PDFs, about one source in 18, against 0.32% for Google AI Overviews, about one in 300. Across all five engines, 1.5% of citations are documents, and most of the PDFs cited come from government agencies.
What we measured
On September 23, 2026 we asked 2,984 everyday questions, drawn from real search demand, to five AI engines from a US location: ChatGPT, Gemini, Perplexity, Google AI Mode and Google AI Overviews. We collected every source cited in the answers, 155,312 in all, and checked the file type of each one, overall, by engine, by topic and by the website it came from.
A citation is a source link an AI engine shows with its answer. A document is a downloadable file rather than a web page: a PDF, Word document, Excel spreadsheet, PowerPoint deck, CSV, text or RTF file. Nearly every document cited was a PDF, so the document and PDF figures are almost identical.
ChatGPT cites PDFs far more than Google
That is 3.5 times the rate of Perplexity (1.62%), ten times Gemini's (0.57%) and more than 13 times Google AI Mode's (0.42%). Google AI Overviews cite a PDF in 0.32% of their citations, the lowest rate of the five, about once in every 300 sources.
Source: Suff Digital analysis of 155,312 cited URLs · answers from ChatGPT, Gemini, Perplexity and Google's AI features, September 2026
| AI engine | Citations that are documents | Citations that are PDFs |
|---|---|---|
| ChatGPT | 5.70% | 5.69% |
| Perplexity | 1.62% | 1.62% |
| Gemini | 0.57% | 0.57% |
| Google AI Mode | 0.42% | 0.42% |
| Google AI Overviews | 0.32% | 0.32% |
| All five engines | 1.51% | 1.51% |
Google's AI features draw their sources from the pages that rank in Google Search, which are overwhelmingly web pages. ChatGPT searches more widely and goes to primary documents when they hold the answer: the official handbook, the regulation, the research report. For an organization whose best information sits in PDFs, ChatGPT is the engine most likely to find it.
One AI citation in 66 is a document
Across all five engines, 1.5% of citations point to a document file rather than a web page, about one in 66. The share changes with the topic. PDFs make up 3.5% of citations in answers about health, 3.2% about cars, 3% about money, 2.9% about weather, 2.7% about taxes and 2.6% about nutrition.
Those are topics where the authoritative source is often a formal document: a drug label, a vehicle recall notice, a tax form's instructions, a dietary guideline. Answers about software (0.1%), beauty (0.2%) and shopping (0.2%) almost never cite a PDF, because the information people need there lives on product and retailer pages.
Government agencies supply most cited PDFs
57% of the PDFs cited in AI answers come from government (.gov), university (.edu) or military (.mil) websites. The top sources are federal agencies publishing consumer guides, weather data and nutrition resources: the Consumer Financial Protection Bureau, NOAA and the US Department of Agriculture. The National Association of Insurance Commissioners and the University of Wisconsin also make the top ten.
Source: Suff Digital analysis of 155,312 cited URLs · answers from ChatGPT, Gemini, Perplexity and Google's AI features, September 2026
Public agencies publish their guidance as PDFs because the documents are official and fixed. AI engines, especially ChatGPT, treat those documents as authoritative. Businesses rarely appear on this list: their PDFs are brochures and price sheets that rarely match the question someone asks.
What this means for businesses that publish PDFs
If an important fact lives only in a PDF, publish it as a web page too. That includes pricing, specifications, service areas, research findings and policy details. Keep the PDF as a download, but give AI engines a page version as well, since Google's AI features cite web pages almost exclusively.
If you publish research or reports as PDFs, add a summary page for each one with a clear title, the key numbers stated in plain sentences and a link to the full document. ChatGPT can then cite either, and Google's AI features have a page to point to. Suff Digital is an ai seo company that turns document-heavy content into pages every AI engine can cite.
Embed this research
Paste this on your site to embed the charts. It links back to the source automatically.
<iframe id="sd-pdfs-in-ai-search" src="https://www.suffdigital.com/embed/data-studies/pdfs-in-ai-search" width="100%" height="600" style="width:100%;border:1px solid #E5E7EB;border-radius:12px" loading="lazy" title="ChatGPT Cites PDFs in 5.7% of Its Sources vs 0.32% for Google AI Overviews - Suff Digital"></iframe>
<script>window.addEventListener("message",function(e){if(e&&e.data&&e.data.sdEmbed==="pdfs-in-ai-search"&&e.data.height){var f=document.getElementById("sd-pdfs-in-ai-search");if(f){f.style.height=e.data.height+"px";}}});</script>
<p style="font:14px/1.5 system-ui,sans-serif">Source: <a href="https://www.suffdigital.com/resources/data-studies/pdfs-in-ai-search">ChatGPT Cites PDFs in 5.7% of Its Sources vs 0.32% for Google AI Overviews - Suff Digital</a></p>
Cite this study
Suff Digital. (2026). ChatGPT Cites PDFs in 5.7% of Its Sources vs 0.32% for Google AI Overviews. https://www.suffdigital.com/resources/data-studies/pdfs-in-ai-search
ChatGPT Cites PDFs in 5.7% of Its Sources vs 0.32% for Google AI Overviews - Suff Digital - https://www.suffdigital.com/resources/data-studies/pdfs-in-ai-search
Frequently asked questions
Related studies
- AI SEOAI Answers Link to a Website's Homepage Only 0.9% of the Time
- AI SEOWhat AI-Cited Web Pages Have in Common: 72% More Likely to Show a 'Last Updated' Date
- AI SEOGoogle's Gemini Cites No Source in 61.2% of Its Answers vs 2.2% for ChatGPT
- AI SEOGoogle AI Mode Puts Only 7.8% of Its Sources in the Answer Text vs 100% for ChatGPT