How to summarize lecture slides with ChatGPT, Claude or Gemini
Slides are made to be talked over. A chatbot reading just the deck misses a lot of what the lecture actually said, and "summarize this" usually gets you a tidy list of the headings you already had.
Can ChatGPT read a PowerPoint file?
Often, yes. Lots of chatbots accept .pptx uploads, and they all take a PDF export. Whether yours does depends on the app and your plan, so check its help page if an upload gets refused.
The harder question is what it actually reads from the file, because you can't see that. When a summary leaves something out, you don't know if the chatbot skipped it or never got it. Converting the deck to text first solves that, since you can read exactly what the chatbot will read.
Why slide decks confuse chatbots
Text order
A slide isn't a page of writing. It's a bunch of boxes placed on a canvas, and something has to decide which box comes first. The converter CourseFeed uses reads boxes top to bottom, then left to right, based on where they sit on the slide. For a title and a list of bullets, that's exactly right. But on a two-column comparison slide where the right-hand box sits slightly higher than the left, the right column comes out first. Labels scattered around a diagram come out as a loose list of words with no arrows between them.
Speaker notes
The notes are often where the real explanation lives. The slide says "Resting potential"; the notes say why it matters and what the lecturer will ask about it. Some decks have no notes at all. Others still have notes from a previous year, or from whoever made the deck originally. If your lecturer posts slides as a PDF, the notes are usually gone, because PowerPoint leaves them out of PDF exports by default.
Images
Converted to text, a picture turns into a one-line placeholder with its name or alt text. If an equation was pasted in as a screenshot, or a figure was scanned from a textbook, none of it ends up in the text. Charts built in PowerPoint do better, since they come out as a small table of the chart's data.
Shorthand
Bullets like "↑ Na⁺ → depol." make sense if you were in the room. Without the lecture, the chatbot fills the gap from general knowledge and won't always tell you it's doing it. And general knowledge might not match how your course defines things.
What the converted deck looks like
Drop the .pptx into CourseFeed and each slide comes out in order. A comment marks the slide number, the slide title becomes a heading, the body text follows, tables become Markdown tables, and speaker notes go under a Notes heading. This is one slide from a test deck, cut short:
# Source: lecture-50-slides.pptx <!-- Slide number: 10 --> # Week 6 summary table 1 | Term | Definition | Example | | --- | --- | --- | | Cell signal affinity. | Energy energy osmosis. | Protein affinity function. | ... ### Notes: Speaker notes: ...
A picture shows up as a line like . That tells the chatbot there was an image on the slide, but not what was in it.
The slide numbers are what make the rest of this guide work. You can ask about "slide 14", and the chatbot can point you back to a slide you can open and check.
If you have the .pptx, convert that instead of a PDF export, so you keep the speaker notes and slide numbers. If your lecturer only posted a PDF, convert that. You'll still get the text, but the slide breaks are harder to see and the notes are probably missing.
Prompts that work better than "summarize this"
Name the file in every prompt. The # Source: heading gives you an exact name to point at, which keeps the chatbot inside your material.
Map the deck first
"Using only the Source section for Week 5 slides.pptx, list every slide number with its title and one line on what it covers. Don't add anything that isn't on the slides."
It takes a minute to check against the deck, and it tells you whether the chatbot has every slide. If the list stops at slide 30 of 45, you find out before you rely on anything else.
Ask for explanation, not compression
"Write study notes for slides 8 to 15. Explain each idea in two or three full sentences, the way a lecturer would say it out loud. Use the Notes sections where there are any. Put the slide number in brackets after each point."
A summary shorter than the slides doesn't help much when the slides were already short. Usually what you need is the missing sentences between the bullets.
Make it flag the gaps
"Where a slide is only a heading, an image placeholder or shorthand you can't interpret, say so instead of guessing."
If you add one line to your prompts, make it this one. Without it, the chatbot quietly fills holes from general knowledge. With it, you get a list of slides to look at yourself or ask about in the next lecture.
Find the structure
"Group the slides into the three to five main ideas of this lecture. For each one, say which slides cover it and how it leads into the next."
A lecture has an argument running through it that's hard to see one slide at a time. This prompt pulls it out.
Connect it to the reading
If you converted the reading along with the slides: "Which points in the slides are explained in more detail in the reading? Give the slide number and quote the start of the matching passage."
Set a length
Say how long you want it: "one page", "under 300 words", "a table with one row per slide". Left alone, a chatbot tends to give you either a few vague lines or the whole deck reworded at full length.
Check the summary against the slides
Since every point has a slide number, checking goes fast. Five minutes covers most of it:
- Pick three slide numbers from the summary at random, open the deck, and see if the point is really there.
- Look for things on the slides that didn't make it into the summary. Check whether the last slides got as much attention as the first ones, because long decks are where things get skipped.
- Look for things in the summary that aren't on the slides, like specific numbers, names, dates or examples, and ask "Which slide is this from?" If it can't tell you, it came from general knowledge. It might be correct, but it isn't your course.
- Treat anything that depended on a picture as a gap until you've looked at the slide yourself.
When to upload the deck as it is
Some lectures are mostly pictures: anatomy, art history, architecture, maps, annotated photos. A text version can't carry that, so uploading the original deck to a chatbot that reads images is the better choice. Ask it to go slide by slide and describe what each image shows.
You can also mix the two. Convert the text-heavy files, then attach the one or two diagrams that matter as images in the same chat. Combining slides, PDFs and Word docs covers that. Photos of slides have no text to convert, so those go in as images too.
After the summary
Reading a summary feels like studying, but it doesn't do much for remembering. Once you've checked the summary, use it as the starting point for questions. Flashcards and practice exams from your notes picks up from there.