Back in lesson 2.4 you met writing blocks, Pages, and a first look at Sites. Text you can shape and share. This lesson is about everything you look at: images, slide decks, and Sites. Three tools, one way of working. You describe what you want, ChatGPT makes a first version, and you refine it until it's right.
You don't need to be a designer. You do need taste and a clear brief, which you already practised in Part 3.
Three ways to make something you can see
Here's the map before we go in:
- Images. Pictures, illustrations, diagrams, mock-ups, product shots, social visuals. Made in chat or in the Images section in the sidebar.
- Slides. A presentation you'll show or send: a board update, a proposal, a training deck. Built in Work with the Presentations plugin, or inside PowerPoint with ChatGPT for PowerPoint.
- Sites. An interactive web page or small app: a dashboard, a tracker, a launch calendar, an internal portal. Built in Work and shared with a link.
The question to ask is: what will the reader do with it? Glance at it: an image. Sit through it: slides. Click around in it, again next week: a Site.
Images: create and edit
Ask for an image in any chat, or open Images in the sidebar and type a prompt. It can take a minute or two, and you can keep chatting while it works. Everything you make is saved in Images.
A few things worth knowing:
- Aspect ratio. Use the picker, or say it in the prompt: "16:9 for a slide", "square for LinkedIn".
- Text in images. ChatGPT can put words on an image, like a title on a poster or labels on a diagram. Check the spelling; it's good, not perfect.
- Transparent background. Ask for it when you want to drop the image onto a slide.
- Templates. Under Images, Templates gives ready-made starting points by category. ChatGPT asks a few questions, such as the style, and you can add a reference image. (Templates aren't in Work mode yet.)
- Images with thinking. On Plus, Pro, and Business, ChatGPT can think before it draws, which helps with complex scenes, diagrams, and anything with lots of text.
Editing is where it gets useful. Two ways:
- Select and describe. Open the image, use the selection tool to highlight an area ("the sky", "the logo on the mug"), and say what should change. OpenAI warns that a highlight isn't always precise and an edit can spill over, so check the edges.
- Just say it. "Same image, but warmer light and no people in the background." Fine for changes to the whole picture.
On your phone you can even sketch: draw a rough shape, add a line of instructions, and let ChatGPT turn it into a finished picture.
Getting a good image: the brief
The recipe from lesson 3.2 works for pictures too. A strong image prompt names:
- The subject. What's in it, and what's happening.
- The purpose. "Header image for a blog post about remote onboarding."
- The style. Photo, flat illustration, line drawing, 3D render, watercolour.
- The mood and colour. "Calm, lots of white space, our green as the accent."
- The format. Aspect ratio, and whether there's room for a headline.
Compare "a team meeting" with "a flat illustration of four colleagues around a laptop, friendly and relaxed, white background, one accent colour, landscape 16:9, empty space on the left for a headline." The second one gets you there in one go.
One rule that matters at work: don't make images of real people (colleagues, clients, public figures) without their clear consent, and don't put real logos on things that aren't theirs. Images made with AI can look very real; use that power in good faith.
Slides: decks you can actually present
There are two routes to a deck.
In Work, with the Presentations plugin. Give Work your source material, such as a report, a brief, or a process document, and describe the deck: who it's for, what decision you need, how many slides. Work drafts it and delivers a real PowerPoint (PPTX) file, with guidance if you need it in Google Slides. OpenAI's own examples: an executive deck that leads with the decision needed, a proposal from a brief, a training deck with a worked example and a knowledge check.
Inside PowerPoint, with ChatGPT for PowerPoint. An add-in that sits in a sidebar in PowerPoint itself. It can draft a deck from your notes or a document, add or revise slides in an existing deck, answer questions about the story, and tailor a deck for a different audience. It works in your existing template where it can, but OpenAI is honest that slides don't always match your house style, and some advanced chart and formatting features are still limited. Their own tip: duplicate the file first. Your IT team may need to install it for you.
Which route? Starting from nothing, with lots of source material: Work. Polishing or extending a deck that already has your template: the PowerPoint add-in.
Sites: from idea to working page
A Site is a small, interactive web page that ChatGPT builds and hosts. Think dashboards, project trackers, launch calendars, prototypes, internal portals, and interactive reports.
You make one in Work (on the web, or in Work or Codex in the desktop app): describe what you need, say "website" or mention @Sites, add your content and data, and refine it from the preview. Sites is in public beta on Plus, Pro, and workspace plans.
Three things to understand before you share:
- It starts private. A new Site is visible only to you and your workspace admins. You then choose: specific people or groups, named external viewers, the whole workspace, or the public internet if your admin allows it.
- Deploys go live. Each deployment URL is a production URL. Sending edits in the chat doesn't change the live Site until you publish, so save a version and review it before you deploy.
- Connected data is read-only. On Business, a Site can show data from connected apps, but every visitor sees it through their own account and permissions, and the Site can't change anything in those apps.
A worked example. Lena runs a product launch with twelve people across three teams. Every week someone asks "where are we?" In Work she attaches the launch plan and the task list, and asks: "Build a website that shows our launch timeline, who owns each milestone, and what's late, with a filter per team." The first preview is close but cluttered, so she asks for one page per team and a red label on anything overdue. She saves a version, checks it, shares it with the launch group only, and deploys. The weekly "where are we?" mail stops. When dates move, she updates the Site from the same chat and publishes again.
The core loop: describe, generate, refine
Whatever you're making, the loop is the same one you learned for text in lesson 3.3:
- Describe the outcome, the audience, and the constraints.
- Generate a first version. Don't judge it on polish; judge it on direction.
- Refine in small, specific steps. "Make the title bigger" beats "make it pop." Change one thing at a time so you can see what worked.
Steer early. If the first version is heading the wrong way, say so straight away and restate the brief, instead of polishing something you'll throw out.
Stuck after three rounds? Stop refining and step back. Ask ChatGPT what it understood from your brief, fix the misunderstanding, and start a fresh version. For images, it often helps to describe what you don't want ("no stock-photo smiles, no gradients"). For decks, give it the one-sentence message each slide must land.
Feed it context, and your brand
A design looks generic when ChatGPT has to guess. Give it something to work from:
- A reference. Upload an image or a slide you like and say what to borrow: the layout, the colours, the mood.
- Your brand guide. Put the colours, fonts, logo rules, and tone in a project (lesson 4.3), so every image, deck, and Site made there starts on-brand.
- Your template. For decks, work in your company's PowerPoint template with the add-in, or hand Work an existing deck and say "same style."
- A skill. If you make the same kind of visual every week, package the brief as a skill (lesson 4.2): "monthly KPI slide in our style."
Where it fits: Page, image, slides, Site, or Codex
- A Page or writing block when the thing is mostly words.
- An image when one picture says it.
- Slides when someone will present it or flip through it.
- A Site when people will click, filter, or come back to it.
- Codex (lesson 5.1) when it needs to become real, maintained software in your own codebase.
Staying in control
Visual work is persuasive, which is exactly why it needs checking:
- Check every number. A chart on a slide looks authoritative even when it's wrong. Compare it with the source.
- Read every word. Typos in images and slides are easy to miss and hard to un-send.
- Mind the audience. Before you share a Site, check the access level. Before you go public, ask whether it should be.
- Respect people and brands. No realistic images of real people without consent, no logos that aren't yours to use, and follow your company's rules.
- You own the result. OpenAI says output can be incomplete or incorrect. Whatever goes out carries your name.
Try it
Pick one real thing you need this week.
An image: "A flat illustration for the header of our team newsletter: [topic], calm colours, one accent in [colour], 16:9, space on the left for a title." Then make one targeted edit with the selection tool.
Or a deck: in Work, attach a report and ask for "five slides for [audience] that lead with the decision we need, in plain language." Check every number against the report before you show it to anyone.
(General exercise. Role and industry versions come once accounts are in.)