AI for Everyday Work: Writing, Research, Meetings & DataYour AI workspace: projects, voice, files and agents · Lesson 5 of 17
Voice, photos and files: AI beyond typing
Video lecture
Voice, photos and files: AI beyond typing
The narrated lecture is in production
Every chapter is scripted and ready. Browse the chapters and read the full transcript now — the video will appear here when it’s published.
Chapters
Transcript of the narration, chapter by chapter.
0:00 AI beyond typing
Quick question. When was the last time you typed out something that was already in front of you? A whiteboard after a workshop. An error message. Notes from a site visit. Most of us still do this, and it's a waste of time, because today's assistants can listen, talk, see images and read files. In this lecture you'll learn when to talk to your assistant, when to show it something, and when to hand it a file. You'll see a simple example and a real operations workflow, you'll watch me run a file question the careful way, and you'll leave with three tasks you can try in the next ten minutes.
0:48 Why multimodal matters
Why does this matter? Because a surprising amount of work is just moving information from one form to another. Speech to notes. Photos to a list. A PDF to three answers your manager needs. When you do that by hand, it's slow, and it's where errors creep in, like the wrong number copied from a receipt. When the assistant does the moving and you do the checking, you get faster and, often, more accurate. And these features aren't exotic anymore. ChatGPT, Gemini, Claude and Copilot all accept images and files, and all of them offer some form of voice on supported devices.
1:32 The sharp intern analogy
Think of it like working with a sharp intern. Sometimes you talk things through while walking to a meeting. That's voice. Sometimes you just point at the whiteboard and say, write this up. That's images. And sometimes you hand over a folder and say, find me the answer, and show me the page. That's files. Here's the key idea for each. With voice, set a frame and end in writing. With images, show rather than describe, but crop out anything private. With files, always ask for evidence: the quote and the page number. Those three rules cover most of what can go wrong.
2:17 Voice done right
Let's start with voice. Voice conversations drift, so open with a frame. Who the AI should be, what you're trying to do, and when to stop. For example: you're a skeptical finance director. I'm asking for budget for a new scheduling tool. Push back on cost and risk. Keep replies short. After eight minutes, stop and give me feedback. Make it realistically hard, because a friendly role play builds false confidence. And here's the golden rule. Always finish by asking for a written summary: decisions, actions, and anything you said you'd follow up on. Paste it into your notes. Voice is where you think. Writing is what you keep.
3:04 Example 1: the whiteboard photo
Here's a simple example. Your team finishes a brainstorm and the whiteboard is covered in sticky notes. Instead of typing them up, you take a photo and ask: transcribe this into a clean list of ideas grouped by theme, and mark anything you can't read as unclear. In a few seconds you have a tidy list in four themes, with two items flagged. You fix those two from memory and share it before everyone's even left the room. One tip: check the photo first. If a colleague's laptop screen with their email is in the background, crop it out.
3:47 Example 2: the warehouse walk
Now a realistic business workflow. Khalid is an operations supervisor in Riyadh. Every Thursday he walks the warehouse. He used to scribble notes, then spend an hour typing an inspection report. Now he dictates as he walks. Loading bay three, pallet wrap torn. Two fire extinguishers blocked. And he snaps six photos. At his desk, in his company's approved assistant, he uploads the photos and the transcript and asks for an inspection table: issue, location, photo number, severity, suggested fix and owner, with unassigned where the owner isn't clear. And a key rule: only include issues I mentioned or that are clearly visible, and mark anything uncertain. Two minutes later he has the table. He changes one severity, assigns owners, and sends it before lunch.
4:41 Watch me do it
Let me show you the careful way to question a file. I've uploaded our company's travel policy, about forty pages. A lazy question would be, what does this say about travel? Instead I type: answer these three questions from the attached document only. For each answer, quote the supporting sentence and give the section or page. If the document doesn't answer a question, say not in the document. My questions: what's the hotel limit, do I need approval for international travel, and can I claim airport lounge access. Here's the answer. Hotel limit, with the quote from section three. Approval, yes, quote from section five. And lounge access: not in the document. That's the valuable part. A lazy question would have given me a confident guess.
5:36 Bonus: dictation for drafts
Here's a bonus technique that people love once they try it. Dictate your first draft. Many of us think faster than we type, especially when we're walking between meetings or when English isn't our first language. So speak your rough thoughts into your phone, ums and all. Then paste the transcript and say: turn this into a clear email to my manager, keep my meaning exactly, under one hundred and fifty words, and list anything that sounds unclear. You get a tidy draft with your ideas, not the AI's. And if you're writing in a second language, add: then list the main changes so I can learn. It's one of the fastest ways to go from a busy head to a sent message.
6:29 Common mistakes
Common mistakes. Treating a voice chat as the record. Ten minutes later you won't remember what was agreed, so end in writing. Trusting numbers read from photos. The assistant might misread a seven as a one on a crumpled receipt, so check extracted numbers against the image, every time. Uploading a whole folder when one page answers the question. More input means more places for confusion and more data exposed. And recording people without telling them. In many places that needs consent, and it's always a matter of trust. Tell people, follow your policy, and when in doubt, take your own notes instead.
7:14 Recap and try this now
Let's recap. Voice: set the role, the goal and a stopping point, and always end with a written summary. Images and screenshots: show instead of describing, and crop out anything private. Files: ask for quotes and page numbers, and welcome the answer not in the document. Here's your try this now. Do three quick tasks today. A ten minute voice rehearsal of a conversation you're dreading. A screenshot of something confusing, with explain this in plain English. And one real document question using the evidence prompt from the lesson text. Notice which one saves you the most time. That's the workflow to keep.
AI beyond typing
Modern assistants don't just read what you type. They can listen, talk, look at images and read files. For everyday work, these "multimodal" features often save more time than clever prompts:
| Input | What you can do | Where it lives (2026) |
|---|---|---|
| Voice conversation | Think out loud, rehearse a hard conversation, practice a language, capture ideas hands-free | ChatGPT voice, Gemini Live, Claude voice mode, Copilot voice on supported apps |
| Dictation / recording | Speak a rough draft; record a voice memo or meeting and get a summary | Built-in dictation in most apps; ChatGPT record mode (macOS desktop app on eligible plans); Notion AI Meeting Notes; meeting tools with AI notes |
| Photos and screenshots | Read a whiteboard, explain an error message, extract a table from a photo, compare two designs | Upload or paste an image in any major assistant |
| Files | Summarize PDFs, compare documents, analyze spreadsheets, extract data from invoices | Upload in ChatGPT, Claude, Gemini, Copilot; connect Drive/OneDrive/SharePoint through connectors |
Features and limits vary by plan, device and region, so check what your account supports.
Voice: set a frame, then end in writing
Voice conversations drift without structure. Open with the role, the goal and a stopping point:
You're a skeptical finance director at a logistics company. I'm asking for budget
for a new scheduling tool. Push back on cost and risk. Keep replies under 30 seconds.
After 8 minutes, stop and give me feedback: my 3 strongest points, my 3 weakest
answers, and a better way to phrase the weakest one.Always finish by asking for a written summary of decisions, action items and follow-ups, and paste it into your notes. Voice is great for thinking; writing is what you keep.
Photos and screenshots: point, don't describe
Instead of typing out a table or describing an error, show it:
- Whiteboard photo after a workshop: "Transcribe this into a clean list of ideas grouped by theme. Mark anything you can't read as [unclear]."
- Screenshot of an error: "What does this error mean and what are the two most likely fixes? I'm not technical."
- Receipt or invoice photos: "Extract date, supplier, amount and tax into a table." Then check every number against the image.
- Two versions of a flyer: "Which is easier to read at a glance, and why? Give three specific improvements."
Images can contain more than you intend: faces, names on screens, addresses, open email tabs. Crop before you upload.
Files: ask for evidence, not just answers
When you upload a document, ask the assistant to point to where it found each answer: "Quote the sentence and give the page or section number." This makes checking fast and exposes guesses. For spreadsheets, ask the assistant to show the steps or formulas it used (you'll go deeper in the spreadsheets lesson).
Worked example: a site visit in Riyadh
Khalid, an operations supervisor, walks a warehouse with his phone. He dictates observations as he goes ("loading bay 3, pallet wrap torn, two fire extinguishers blocked") and takes six photos. Back at his desk, in his company's approved assistant, he uploads the photos and the transcript:
Here are my spoken notes and 6 photos from today's warehouse walk.
Create an inspection summary: issue, location, photo number, severity (high/medium/low),
suggested fix, owner (write "Unassigned" if unclear).
Only include issues I mentioned or that are clearly visible; mark anything uncertain as [check].He gets a clean table in two minutes, corrects one severity rating, assigns owners, and sends it before lunch. Previously this took an hour of typing.
Hands-on: three multimodal tasks to try today
1. Voice rehearsal (10 minutes). Use the finance-director prompt above (or adapt it to a conversation you actually need to have). At the end, ask for written feedback.
2. Screenshot explainer (2 minutes). Screenshot something confusing, such as a settings page, a chart or an error. Ask: "Explain this in plain English and tell me what, if anything, I should do."
3. File with evidence (5 minutes). Upload a policy or long PDF you actually need to use and ask:
Answer these three questions from the attached document only.
For each answer, quote the supporting sentence and give the section or page.
If the document doesn't answer a question, say "Not in the document".
1. [question]
2. [question]
3. [question]Before: "What does this policy say about refunds?" returns a confident paragraph mixing the document with general knowledge.
After: three short answers, each with a quote and page number, and "Not in the document" for the one the policy doesn't cover, which you would otherwise have assumed.
Consent and privacy
- Recording people requires telling them, and in many places their consent; follow local law and your organization's policy.
- Photos of people, screens and documents can capture personal data. Crop, blur or skip.
- Files follow the same data classes as text: confidential files only in work-approved tools; restricted data never.
Pitfalls
- Treating a voice chat as the record. Always end in writing.
- Trusting numbers extracted from photos without checking the image.
- Uploading a whole folder when one page answers the question.
How to measure success
Pick one weekly task you currently type from scratch (notes, inspections, expense entries). If voice, photos or files cut it by half with the same accuracy, keep the workflow.
Key takeaways
- Voice, photos, screenshots and files often save more time than clever prompts.
- Frame voice sessions with a role, goal and stopping point, and always end with a written summary.
- Show, don't describe: upload the screenshot or photo, and crop out anything private first.
- With files, ask the AI to quote and locate its evidence, and to say when the document doesn't answer.
Check your understanding
Quick questions to lock in the lesson. They don’t count towards your certificate.
Put it into practice
Do three tasks today: a 10-minute voice rehearsal ending in a written summary, a screenshot explainer, and one document question with quotes and page numbers.
Enrol for free to save your progress
Reading is always free. Enrol to keep your place, take the final assessment and earn a verifiable certificate.