How to ask AI about a PDF, photo or screenshot on your phone

Typing out a contract clause or describing an error message on a phone is slow, and details get lost on the way. It is usually faster, and more accurate, to give the AI the file itself: the PDF of the report, a photo of the letter, a screenshot of the error. Models that accept images and documents can read them directly and answer questions about them.

This guide covers what works well, how to prepare a file so the model reads it correctly, the exact steps in Cortex AI and the checks that keep you from trusting a misread number.

What AI is good at with files

• Summarising long documents. Reports, terms of service, meeting notes or research papers, reduced to the points you care about.
• Pulling out specific facts. Dates, amounts, deadlines, names and obligations, collected into a list or a table.
• Explaining visuals. Charts, diagrams, slides and dashboards described in plain language.
• Reading screenshots. Error messages, app settings or a confusing form, with a suggestion of what to try next.
• Translating photos of text. Menus, signs, labels and letters in another language.
• Answering questions about a page. "What does clause 7 mean for me as a tenant?" works better than retyping the clause.

Prepare the file before you upload

• Check the format and size. In a Cortex AI chat you can attach images and PDFs of up to 10 MB each. Split or compress larger documents, or upload only the pages you need.
• Prefer the original over a photo of it. A PDF exported from the source keeps its text clean. A photo of a printed page works too, but small print, glare and skewed angles cause misreadings.
• Crop to what matters. For a screenshot or photo, cut away everything unrelated so the model focuses on the part you are asking about.
• Leave out what the task does not need. Files are sent to the AI provider that answers. Crop or cover account numbers, ID numbers and other people's personal details before uploading.

Step by step in Cortex AI

1. Pick a model that can read files. Open the model picker and tap the Vision filter to list the models that accept images and documents. With other models, the attachment options do not appear.
2. Attach the file. Tap the + button next to the message field and choose Camera to photograph a page, Photo Library for an existing image or screenshot, or Document for a PDF.
3. Ask a specific question. Instead of "what is this?", say what you need: "Summarise this lease in five bullet points and list every date and amount I have to pay." Point to a section or page when you can.
4. Ask for the evidence. Add "quote the sentence you used" or "tell me which page" so you can check the answer against the file in seconds.
5. Re-attach the file for new questions. The model reads a file together with the message you attach it to. Follow-up messages carry the conversation so far, including the earlier answer, but not the file itself. When you ask about a different section, or switch to another model for a second reading, attach the file again with that question.

Prompts that work well with files

• Summary: "Summarise this PDF for someone with five minutes. Start with the three decisions it asks me to make."
• Extraction: "List every deadline in this document as a table with the date, what is due and the page number."
• Explanation: "Explain this chart in plain language. What changed, and by how much?"
• Troubleshooting: "This is a screenshot of an error on my phone. What does it mean, and what should I try first?"
• Translation: "Translate the text in this photo into English and keep the original line breaks."

Limits worth knowing

Models can misread digits in low-quality images, mix up rows in dense tables, skip handwritten notes and lose track of details in very long documents. Always check figures, dates and names against the file before you rely on them, and ask about one section at a time when a document is long. An AI explanation of a contract, a medical result or a financial statement is a starting point, not professional advice.

File attachments depend on your plan, and the app shows which features yours includes. When you no longer need a conversation, you can delete it in the app.