AI in Prepress: Where It Helps, and the New Problems It Creates

AI has arrived in prepress from two directions at once. Customers are sending files created with AI tools, and software vendors are adding AI to the tools printers use. Both callas and Enfocus have written about this in 2026, and their experience lines up with ours: AI is genuinely useful in places, but it is not magic, and it introduces new problems as well as solving old ones.

We looked at the bigger picture in AI in print: necessary evil or genuine game-changer? This post is more practical: what is actually happening in prepress right now.

Problem one: AI-generated PDFs that break preflight

In September 2026, callas documented a specific issue with PDFs created by ChatGPT. These files carry Content Credentials, the C2PA provenance data that records how a file was made. The credentials themselves are fine, but the way they are embedded includes an extra entry that tells PDF software the data lives in an external file that does not exist.

For a print workflow, that has two consequences:

• The file cannot be PDF/X compliant as delivered, because PDF/X does not allow that entry.
• Preflight checks that inspect embedded files can fail on it.

callas has addressed this in pdfToolbox 17.1, which removes the faulty entry automatically whenever a Fixup runs and keeps the credentials intact. On earlier versions there are workarounds, such as discarding embedded files or excluding them from checks. Whichever tool you use, the lesson is wider than one bug: AI tools produce PDFs in ways the print industry has not tested, and preflight needs to keep up.

Problem two: provenance data you may need to keep

Content Credentials are likely to become more common, and some brands and publishers will want them preserved. That creates a new question for prepress: when a fix strips embedded data to make a file compliant, are you removing something the customer needs? It is worth deciding now how your workflow should handle credentials, rather than discovering it on a live job. callas has noted that specifying C2PA properly for PDF is on the standards agenda, which we cover in PDF/X is moving.

Where AI genuinely helps: writing scripts and configurations

In July 2026, David van Driessche at callas tested whether AI could write pdfToolbox imposition scripts. His verdict was "yes, but only if the AI is handled correctly." His findings match what we see when using AI for workflow work:

• Public AI models have little product-specific training, so they produce plausible code that gets details wrong.
• Results improve dramatically when the AI is given the product documentation, clear rules and a small set of clean, consistent examples.
• The AI should be told to ask about assumptions such as bleed, gaps, margins and rotation, rather than guess.
• Output is not repeatable: the same prompt can give different code, and models change over time.
• Everything must be tested in the real product. AI speeds up the draft, test and correct cycle; it does not remove it.

In other words, AI is a capable assistant for someone who already understands the workflow, not a replacement for that understanding.

Where AI genuinely helps: help inside the tools

Enfocus took a different route with PitStop 25.11, adding Botus, an AI chatbot that answers questions about using PitStop. It was launched as a beta, with Enfocus inviting feedback from users, and in PitStop 26.03 it sits in a new My PitStop panel alongside documentation and resources. For operators learning the software, or trying to remember how a rarely used function works, that kind of in-context help can save real time.

What we recommend

1. Update your preflight tools. Make sure you are on versions that handle files from AI tools properly, for example pdfToolbox 17.1.
2. Test AI-created files deliberately. Ask a few customers for files they produced with AI tools and run them through your workflow before they arrive on a deadline.
3. Decide your policy on Content Credentials. Keep, strip or keep a separate original: choose deliberately.
4. Use AI to accelerate configuration, not to replace testing. Give it documentation and examples, and verify every result.
5. Keep humans on the exceptions. Automation and AI handle the routine; experienced people handle what they cannot.

If AI-created files are already causing problems in your workflow, talk to us and we will help you put the right checks in place.

Sources: Understanding and fixing an issue in AI-generated print PDFs (callas); Using AI to create pdfToolbox imposition scripts (callas); Enfocus reveals new AI chatbot feature in PitStop 25.11 (WhatTheyThink); Enfocus unveils PitStop 26.03 update (DPS Magazine).