# AI Document Report 2026: The median AI-made PDF takes one message and about 3 minutes | BuildPDFs > Original data from 1,348 AI-made documents: what people make, how long it takes, how many edits, which fonts and images the AI picks, and how many pages a document has. Source: https://buildpdfs.com/research/ai-document-report-2026 [Research](https://buildpdfs.com/research) / AI Document Report 2026 AI Document Report 2026 # The median AI-made PDF takes one message and about 3 minutes We studied 1,348 documents that 887 people made with BuildPDFs between April 1 and October 9, 2026\. Together they sent 3,096 requests to the AI designer. This report shows what people make, how fast they get a finished PDF, and what the AI designer picks for them. All numbers are aggregates. Published October 10, 2026 · Last updated October 10, 2026 · [Download the data (CSV)](https://buildpdfs.com/research/ai-document-report-2026.csv) 1,348 documents studied 3.2 min median time to a downloaded PDF 67% of downloaded PDFs needed no edits 31% requests not in English ## Key findings 1. The median document goes from first request to downloaded PDF in 3.2 minutes, with one message. 2. 67% of downloaded documents were downloaded after the first message, with no edits. 3. Guides, workbooks and checklists are the most common type (24%), then ebooks and books (16%). 4. 42% of documents start from the person's own material: more than 1,500 characters pasted into the first request. 5. 31% of requests are in a language other than English. Spanish is the largest (10%). 6. The AI designer generates images in 52% of documents, a median of 2 images each. 7. Montserrat, Inter and Playfair Display are the fonts the AI designer picks most. 8. The median exported PDF has 7 pages. 26% of documents aim for more than 10 pages. 9. When people do ask for changes, they most often ask to add content (28% of follow-up messages). 10. 45% of documents are downloaded on a phone or tablet. ## 1\. A finished PDF takes a median of 3.2 minutes and one message We measured the time from a person's first request to their first PDF download. For the 657 documents with a recorded download, the median was 3.2 minutes. A quarter were ready in under 1.8 minutes. Nine in ten were ready in under 17 minutes. Most people do not edit at all. 67% of downloaded documents were downloaded after one message. Across all 1,348 documents, 60% have only one request, and 7% have six or more. __Requests per document__ | Group | Share | Documents | | ------------------ | ----- | --------- | | 1 message | 60% | 803 | | 2 messages | 19% | 250 | | 3 to 5 messages | 15% | 206 | | 6 or more messages | 7% | 89 | ## 2\. Guides, ebooks and brochures are what people make most Guides, workbooks and checklists lead with 24% of documents. Ebooks and books follow with 16%, then brochures, flyers and catalogs with 13%. Worksheets, lessons, reports and research papers together are 18%. __Documents by type__ | Group | Share | Documents | | -------------------------------- | ----- | --------- | | Guide, workbook or checklist | 24% | 329 | | Ebook or book | 16% | 216 | | Brochure, flyer or catalog | 13% | 179 | | Worksheet or lesson | 10% | 133 | | Report or research paper | 9% | 115 | | Proposal, pitch or business plan | 7% | 87 | | Form, invoice or letter | 5% | 67 | | Other | 17% | 222 | Sorted by keyword rules on the document title and first request. See the methodology. ## 3\. Four in ten documents start from the person's own material 42% of first requests are longer than 1,500 characters, about 250 words. These people paste in a draft, notes, a manuscript or a table and ask for a designed PDF. 46% start with a short prompt and let the AI write the content. The median first request is 530 characters. __Length of the first request__ | Group | Share | Documents | | --------------------------------------------- | ----- | --------- | | Short prompt (under 300 characters) | 46% | 621 | | Detailed brief (300 to 1,500 characters) | 12% | 163 | | Own content pasted in (over 1,500 characters) | 42% | 564 | ## 4\. 31% of requests are not in English Almost one in three documents is requested in another language. Spanish is the largest at 10%, then Arabic or Persian, French and Portuguese. The AI designer writes the document in the language of the request, with fonts that support the script. __Language of the first request__ | Group | Share | Documents | | ----------------- | ----- | --------- | | English | 69% | 806 | | Spanish | 10% | 113 | | Arabic or Persian | 5% | 61 | | French | 5% | 56 | | Portuguese | 4% | 44 | | Other languages | 8% | 89 | Base: 1,169 documents with enough text to tell the language. ## 5\. Over half of documents get AI-generated images The AI designer called the image generator in 52% of documents (705). A document with images has a median of 2. Realistic photos are the most common style, then illustrations. __Image style, share of documents with images__ | Group | Share | Documents | | --------------- | ----- | --------- | | Realistic photo | 61% | 433 | | Illustration | 49% | 347 | | Abstract | 8% | 58 | A document can use more than one style, so shares add up to more than 100%. ## 6\. Montserrat, Inter and Playfair Display are the AI's favorite fonts Before it uses a Google Font, the AI designer looks up its weights and styles. It did this in 36% of documents. In those documents, Montserrat came up in 32%, Inter in 26% and Playfair Display in 21%. __Fonts the AI designer looked up, share of documents__ | Group | Share | Documents | | ----------------- | ----- | --------- | | Montserrat | 32% | 154 | | Inter | 26% | 128 | | Playfair Display | 21% | 102 | | Plus Jakarta Sans | 9% | 44 | | Lora | 8% | 39 | | Crimson Pro | 8% | 38 | | Libre Baskerville | 6% | 27 | | Roboto | 5% | 26 | Base: 488 documents with a font look-up. A document often uses two or more fonts. ## 7\. The median exported PDF has 7 pages For the 69 documents exported with our PDF server since it launched on October 7, 2026, the median page count was 7. The middle half had 4 to 12 pages. A larger sample tells the same story. The AI designer sets a target page count when it checks the layout. Across 193 documents, the median target was 6 pages, and 26% aimed for more than 10 pages. __Target page count__ | Group | Share | Documents | | ------------------ | ----- | --------- | | 1 to 2 pages | 19% | 37 | | 3 to 5 pages | 26% | 51 | | 6 to 10 pages | 28% | 54 | | 11 to 20 pages | 15% | 28 | | More than 20 pages | 12% | 23 | ## 8\. When people ask for changes, they ask for more content We grouped the 816 follow-up messages in English-language documents by what they ask for. Adding content or sections is the top request at 28%. Colors, fonts and style come next at 21%, then images at 20%. Layout fixes are 17%. __Follow-up messages by request type__ | Group | Share | Messages | | ----------------------------------------------- | ----- | -------- | | Add content or sections | 28% | 226 | | Colors, fonts and style | 21% | 174 | | Images and logos | 20% | 164 | | Layout, spacing and page breaks | 17% | 136 | | Remove or shorten | 13% | 106 | | Fix or rewrite text | 12% | 98 | | None of these (questions, new material, thanks) | 42% | 341 | Base: 816 follow-up messages in 323 documents. A message can match more than one type. ## 9\. 45% of PDFs are downloaded on a phone or tablet Many people make their documents on a phone. Of 657 downloaded documents, 45% were downloaded on a phone or tablet and 55% on a computer. __Device of the first download__ | Group | Share | Documents | | --------------- | ----- | --------- | | Phone or tablet | 45% | 293 | | Computer | 55% | 364 | ## 10\. Most people start from a blank prompt, not a template 79% of documents start from a plain request. 21% start from one of the [ready-made templates](https://buildpdfs.com/templates). The [Company Brochure](https://buildpdfs.com/templates/company-brochure) template is the most used, with 33% of template starts. __Starting template, share of template starts__ | Group | Share | Documents | | ------------------- | ----- | --------- | | Company Brochure | 33% | 91 | | Minimal eBook | 8% | 21 | | All other templates | 60% | 167 | ## Methodology Dataset 1,348 documents from 887 people, with 3,096 requests. A document counts when its owner sent at least one request to the AI designer. Period April 1 to October 9, 2026 (UTC). Sources The production database (documents, requests, AI tool calls) and product analytics (downloads, devices, PDF exports). Excluded Team and test accounts. Time and edits Time runs from the first request to the first PDF download. Downloads come from product analytics, so ad blockers hide some. These numbers use the 657 documents with a recorded download. Document types Keyword rules on the title and the first request, in English, Spanish, Portuguese, French, Arabic, Russian and other languages. The first rule that matches sets the type. In a hand check of 60 random documents, about 9 in 10 labels were right. Most errors were documents left in Other. Languages A rule on the script and common words of the first request. 179 requests were too short to tell and are not in the base. Follow-up requests Every request after the first, in documents whose first request is in English. Keyword groups. A message can be in more than one group. Pages Exported page counts come from the PDF server (from October 7, 2026). Target page counts come from the layout check of the AI designer (from September 27, 2026). Fonts and images From the AI designer's tool calls: font look-ups and image generator calls. Privacy We publish aggregates only. Each published group has at least 20 documents and 20 people. Smaller groups are merged into Other or left out. We publish no titles, document text, names or any data that identifies a person. A script reads request text only in memory to sort documents into groups. Limitations These are BuildPDFs users, not all AI users. Keyword rules are not perfect. The page counts cover a short period. The aggregate tables are in the [CSV download](https://buildpdfs.com/research/ai-document-report-2026.csv). Each row has the count, the number of people and the share. ## How to cite this report You can quote and chart these numbers. Please link to this page. > BuildPDFs (2026). AI Document Report 2026: The median AI-made PDF takes one message and about 3 minutes. https://buildpdfs.com/research/ai-document-report-2026 ## Make your own document in minutes Describe it or paste your content. BuildPDFs designs a print-ready PDF with fonts, layout and images. [Try the AI PDF generator](https://buildpdfs.com/ai-pdf-generator)