Keep the YouTube transcript when you may need to verify wording, revisit context, or recover a timestamp. Keep an AI summary when you need a shorter navigation or review aid. For study, research, or consequential claims, keep both: treat the transcript and video as the source record, and treat the summary as a derived index. A summary is easier to scan, but it cannot preserve every spoken qualification or visual detail.
Publication and evidence note: Heni Hazbay, creator of Summarise Visually, has a direct interest in the app. This guide was drafted with AI assistance and editorially checked against the cited YouTube Help sources and current project evidence. The retention matrix, timestamp protocol, claim ledger, and fictional lecture example are original editorial assets, not independent results. No current-device compatibility or summary-accuracy benchmark across video types was run; product claims stay within the verified boundaries stated below.
Understand what each record can do
A YouTube transcript and an AI summary solve different problems. The transcript follows the video’s spoken content in sequence. The summary selects and compresses information for a purpose, such as finding the main points or preparing review questions.
YouTube Help’s transcript instructions explain that transcripts are available for videos with captions and that selecting a transcript line can move playback to the corresponding part of the video. That connection between text and playback makes a transcript useful for verification. It does not make the transcript a substitute for everything visible or audible in the video.
An AI summary changes the source. It may regroup ideas, omit repetition, or express a point in new wording. Those changes can reduce the amount of text you scan, but they also create a need to check the output against the transcript and the video.
Use the keep, archive, or delete decision matrix
Decide what to retain according to your task and the cost of an error.
| Task | Keep transcript? | Keep AI summary? | Why |
|---|---|---|---|
| Find one statement again | Yes | Optional | Transcript lines and playback provide a source route |
| Review the main ideas | Yes, at least until checking ends | Yes | Summary supports scanning; transcript supports correction |
| Create study questions | Yes | Yes | Questions may omit context or imply an answer the source did not give |
| Cite or quote a speaker | Yes | Optional | Verify wording, speaker context, and required citation details in the video |
| Record a procedure | Yes | Yes | Sequence, warnings, and on-screen steps need separate checks |
| Make a consequential decision | Yes | Yes, only as an aid | Return to the original video and other authoritative evidence |
| Keep casual personal notes | Your choice | Your choice | Retention depends on whether you expect to revisit the source |
Delete the transcript only when you have no foreseeable need to verify the summary and no retention requirement applies. If you created the transcript outside YouTube, also consider permission, privacy, storage, and deletion obligations before keeping a copy.
Prefer the transcript when source fidelity matters
Keep a transcript when exact language, sequence, or nearby context matters. Examples include a definition, a speaker’s qualification, a named source, a number, or a step in a process.
The transcript gives you a location, not automatic truth. A speaker can be mistaken. YouTube Help’s automatic-caption guidance warns that automatic captions may misrepresent spoken content because of factors such as mispronunciations, accents, dialects, or background noise. A textual record may also omit speaker changes or vocal cues. Replay the relevant moment and inspect the screen before quoting or relying on a claim.
Keep enough source metadata to reopen the right version:
- Video title and channel
- Video URL
- Publication or update information visible on YouTube
- Date accessed
- Timestamp for every consequential note
- Whether you used the YouTube transcript or another permitted record
- Any visible caption or transcript limitation you observed
This record prevents a detached sentence from becoming an unattributed fact in later notes.
Prefer a summary for navigation and review
A summary helps when the transcript is too long to scan during review. It can group related ideas, separate main points from examples, and turn a sequence into a compact outline.
Define the summary’s purpose before generating it. “Give me the main argument and its limits” requires different selection from “list the procedure in order” or “create questions for retrieval practice.” One generic summary should not be expected to serve every task.
Preserve a source pointer beside each important note. A summary without timestamps can save reading time once, then cost time when you need to find the evidence again.
Follow the timestamp verification protocol
Use this protocol for every name, number, definition, recommendation, causal claim, or safety-related step.
1. Create a claim ledger
Copy each consequential summary statement into its own row:
| Summary statement | Transcript timestamp | Playback check | Visual check | Status |
|---|---|---|---|---|
| Unchecked |
Do not group several claims under one timestamp. A sentence that contains a conclusion, number, and exception may need three source locations.
2. Find the matching transcript line
Search the transcript for a distinctive term. Select the line to return to the corresponding part of the video where YouTube supports that interaction. Read the lines before and after it so that you do not remove a condition or mistake an example for the speaker’s position.
3. Replay the relevant segment
Listen for wording, emphasis, speaker identity, and corrections. Extend the playback window when the sentence depends on an earlier definition or later qualification.
Do not treat a transcript mismatch as permission to choose whichever version suits the summary. Record what the video appears to say and note uncertainty when the audio is unclear.
4. Inspect visual evidence
Check slides, demonstrations, charts, labels, equations, and on-screen text. A transcript can represent speech while omitting the material that makes the speech intelligible.
If the speaker says “this increased,” the referenced chart may define what “this” means and over which period. Your summary should not resolve that reference without inspecting the frame.
5. Mark the result
Use one of four statuses:
- Verified: wording, context, speaker, and relevant visual agree with the note
- Revised: you corrected the note to match the source
- Qualified: the note now states uncertainty or a narrower scope
- Unresolved: the source remains unclear or incomplete
Keep unresolved items out of a confident final summary. A visible question is more reliable than an invented resolution.
Worked retention example
Imagine a fictional 18-minute public lecture with captions. At 04:10, the speaker introduces a three-part framework. At 09:30, a slide adds an exception that the spoken explanation mentions only as “the second condition.” At 15:20, the speaker narrows the recommendation to first-year projects.
A generated summary says: “The speaker recommends the three-part framework for all projects.” The transcript can help you locate the framework and the late recommendation, but the slide supplies the missing condition.
The claim ledger would show:
| Summary statement | Transcript timestamp | Playback check | Visual check | Status |
|---|---|---|---|---|
| Framework has three parts | 04:10 | Speaker lists three parts | Slide matches | Verified |
| Framework applies to all projects | 15:20 | Speaker says first-year projects | Earlier slide adds a condition | Revised |
The corrected note might read: “The speaker presents a three-part framework and recommends it for first-year projects under the condition shown on the slide at 09:30.” This example is fictional and does not represent an app output or product test.
Use Summarise Visually within the verified YouTube boundary
Current project review recognizes YouTube URLs and requests English timed text when it is available. The verified route does not record audio, transcribe speech, guarantee that a transcript exists, or guarantee complete source coverage.
Use a bounded workflow:
- Open the video and confirm that YouTube exposes a transcript you can inspect.
- Copy the eligible video URL into Summarise Visually.
- Generate a summary in a mode that matches your purpose.
- Keep the YouTube page open as the source record.
- Add timestamps to consequential summary statements.
- Replay each source moment and inspect related visuals.
- Save the checked summary with the video URL and access date.
If transcript text is unavailable to the route, do not describe a generated answer as a summary of source text the app did not receive. Use manual notes or another permitted, verifiable workflow.
For setup details, follow the iPhone YouTube summarizing workflow. Use the video notes template when you need a structured timestamp record.
Source and evidence note
The source-interface and caption-accuracy statements in this guide come from YouTube Help’s “View video transcripts” and “Use automatic captioning” pages, accessed on July 20, 2026. The retention matrix, verification protocol, claim ledger, statuses, and fictional lecture example were created for this guide.
Product statements reflect current project review. No current-device test across captioned, captionless, restricted, live, long, or multilingual videos was completed for this page.
Limits and verification
The verified route does not record audio, transcribe speech, guarantee a transcript, or guarantee complete source text. It requests English timed text when available. Link recognition alone does not establish that a particular video can be processed.
Neither a transcript nor an AI summary independently verifies the speaker’s claims. Transcripts can omit visual context, and summaries can omit spoken qualifications. Replay the video, inspect the screen, and consult authoritative underlying sources before relying on consequential information.
Respect access, privacy, copyright, educational, and workplace rules when storing or processing transcripts. Keep source locations and label unresolved wording. For lecture-specific retention, compare a lecture transcript with lecture notes or follow the workflow for turning a lecture transcript into study notes.