How we checked (2026-11-03): Whisper Web limits from pricing and llms.txt. ChatGPT video wording from OpenAI Help, ChatGPT Image Inputs FAQ (“Can I upload videos to ChatGPT?”). Audio upload plan gates from Uploading files and audio to ChatGPT. OpenAI pages change; re-open them before you rely on a limit. We did not re-test every ChatGPT client for this draft.

What people mean by “Can ChatGPT watch videos?”

The phrase usually mixes three jobs:

1. See frames — describe what appears on screen.
2. Hear speech — turn talk into a usable transcript.
3. Open a public link — paste a YouTube or TikTok URL and expect the model to fetch the stream.

Those jobs are not the same product. ChatGPT is strong at chat follow-ups once it has text. It is weak as a dedicated, offline, export-friendly transcription pipeline. Third-party “ChatGPT watches YouTube via MCP / plugin” pitches are outside the workflow we recommend here. We also do not teach pasting watch-page URLs into downloaders to grab other people’s videos.

What OpenAI actually says about video uploads

In the Image Inputs FAQ (reviewed 2026-11-03), OpenAI states that ChatGPT can accept video files as attachments on supported methods, including some Free-plan paths, with availability that varies by platform and account. The same answer adds the important caveat: analysis can be incomplete or inaccurate, and ChatGPT may not analyze the entire video or accurately interpret its audio.

That is enough to reject the marketing claim “ChatGPT watches videos like a human editor.” For a lecture, interview, or meeting export where the value is what was said, incomplete audio understanding is the wrong risk profile.

A separate help page covers audio uploads. As of our check, audio file uploads are described for paid ChatGPT plans and workspaces, not the Free plan. Video containers identified as video are not the same as the audio-upload path. Treat video upload and audio upload as different gates, and confirm on OpenAI’s current help pages for your account.

Live camera / screen share in Voice mode is another product surface. It helps with “what am I looking at right now?” It does not replace a time-coded transcript of a file you already recorded.

Why “paste the YouTube link” fails as a transcription plan

Pasting a public watch URL into chat usually sends ChatGPT a page to talk about, not a media decoder for the full stream. Even when browsing tools run, you often get search results or a “couldn’t retrieve the transcript” style answer—not a clean, editable transcript with timestamps.

More important ethically: building a habit of “any URL → download → rehost” is how people accidentally process media they do not have rights to use. Stick to files you own, public-domain / Creative Commons material, or content you have permission to keep offline. This article only covers that authorized path.

A better split of labor: transcript first, ChatGPT second

Use a speech model for speech. Use a chat model for chat.

Practical steps for speech-heavy video

1. Save authorized media with your own tools onto your device (MP4, WebM, MKV, or exported audio). Do not treat a random watch-page paste as a download license.
2. Transcribe in the browser so free local mode can keep processing on-device.
3. Spot-check names, numbers, and quotes against the original audio.
4. Paste the cleaned transcript (or a section) into ChatGPT and ask for summaries, action items, or study notes.
5. Export TXT / SRT / VTT from the transcription tool if you need captions or a durable archive.

Browser Whisper path on whisperweb.dev (facts dated 2026-11-03)

The official product site is whisperweb.dev — not lookalike domains such as .app, .tech, or .net.

For authorized YouTube-style files already saved on disk, open YouTube to Text. That page states it does not fetch a watch-page URL directly; you upload the saved file. For a broader saved-video workflow, use Video to Text.

Product interface note: first screen of https://whisperweb.dev/youtube-to-text as captured 2026-11-03.

From pricing / llms.txt on 2026-11-03:

• Free local mode: processes in the browser; not uploaded in that mode; no sign-up; each file up to 20 minutes and 200 MB; no daily file-count limit; 100+ languages.
• Whisper Web Unlimited (paid): optional cloud upload; up to 10 hours or 5 GB per file; batch and sync features for larger jobs.
• Exports commonly include TXT, JSON, SRT, and VTT depending on the tool page.
• Accuracy still depends on clarity, accents, noise, and overlapping speech—verify before you cite anyone.

Main call to action for this article: start on https://whisperweb.dev/youtube-to-text with a file you are allowed to process. Mid-article alternative for general video files: https://whisperweb.dev/video-to-text.

First local run may download the speech model; keep the tab open. Device memory still matters for long or high-bitrate files—split long recordings if the browser struggles.

Honest comparison: ChatGPT vs browser transcription

What not to do

• Do not treat ChatGPT video upload as a guaranteed transcription service.
• Do not chase MCP or unofficial “watch YouTube inside ChatGPT” plugins for this job.
• Do not paste YouTube / TikTok watch links into shady downloaders to process other people’s streams.
• Do not invent accuracy percentages or claim HIPAA / clinical compliance for a general Whisper tool.
• Do not confuse free local (on-device) with Unlimited cloud (upload).

FAQ

Can ChatGPT watch a YouTube video if I paste the link?

Expect chat about the page or search context, not a full media decode. For speech, use a transcript (official captions you are allowed to copy, or a file-based transcription of media you may keep).

Can ChatGPT transcribe an uploaded MP4?

OpenAI allows some video attachments, but help text warns analysis may miss parts of the video and may misread audio. For a durable transcript, run a dedicated speech workflow first.

Is Whisper Web free?

In free local mode on whisperweb.dev: yes for the limits above (20 min / 200 MB per file, no account, no daily file-count cap, browser processing). Longer files need Unlimited cloud or splitting. Checked 2026-11-03 on pricing and llms.txt.

Which page should I open?

Authorized saved YouTube-style files → youtube-to-text. Other saved video containers → video-to-text. Primary CTA for this piece is youtube-to-text.

Is whisperweb.app the same product?

No. Use whisperweb.dev only.

Bottom line

ChatGPT remains useful after you have trustworthy text. It is not a dependable “watch video → perfect transcript” appliance in 2026, and OpenAI’s own caveats say so. For clips you are allowed to keep, transcribe first in the browser on whisperweb.dev/youtube-to-text, fix the draft, then paste into ChatGPT for the thinking layer.

Sources

• OpenAI Help — ChatGPT Image Inputs FAQ (video upload caveat; reviewed 2026-11-03)
• OpenAI Help — Uploading files and audio to ChatGPT (audio plan gates; reviewed 2026-11-03)
• Whisper Web — pricing, llms.txt, youtube-to-text, video-to-text (product limits 2026-11-03)

How we checked: Product limits for Whisper Web come from https://whisperweb.dev/pricing and https://whisperweb.dev/llms.txt on 2026-11-03. ChatGPT behavior summarized below follows OpenAI’s public help/docs as reviewed in our 2026-10-14 article pass (audio uploads on paid plans, size limits, Record mode on Mac, API 25 MB). OpenAI pages change; re-open their help center before you rely on a limit. We did not re-upload test files to ChatGPT for this refresh.

ChatGPT is a chat product that can accept audio

OpenAI’s help materials (as of our 2026-10-14 review) described roughly three audio paths:

1. Upload an audio file in chat (paid plans; Free plan called out as unable to upload audio at that time). Supported types included common audio formats; very large or video-container files could fail. Files go to OpenAI’s service.
2. Record mode in the macOS app on eligible paid plans, with a long per-session cap in the help text then.
3. Speech-to-text API for developers, with a separate 25 MB request limit in the docs then.

Useful follow-ups (“summarize,” “list actions”) are the chat strength. Guaranteed speaker labels, offline processing, and “no upload” are not.

Where ChatGPT stops for transcription work

• Plan gates. Free users may not be able to upload audio at all — confirm on OpenAI’s current help page.
• Uploads. The recording leaves your device. Retention depends on plan and settings (including Library vs chat deletion).
• Not a file workflow. Batch folders, SRT/VTT export, and local-only modes are not the core ChatGPT promise.
• Errors. OpenAI’s own pages warn transcripts can be wrong.

For private interviews or regulated audio, “paste into ChatGPT” may be the wrong default.

A free Whisper path for files you already have

OpenAI also released the Whisper speech model as open source. Tools can run it in new places — including inside a browser tab.

On Whisper Web’s M4A to text page (official site whisperweb.dev), free local mode works like this (pricing / llms.txt, 2026-11-03):

• Audio is processed in the browser and is not uploaded in free local mode.
• No account.
• Each file up to 20 minutes and 200 MB.
• No daily file-count limit.
• Optional Unlimited cloud: up to 10 hours / 5 GB per file (paid).

That path is for your recordings (phone voice memos, exported meeting audio, lecture files you are allowed to keep). It is not a “paste any YouTube link” downloader, and we do not recommend grabbing other people’s media.

Simple file workflow

1. Export or save the recording (M4A, MP3, MP4, etc.).
2. Open https://whisperweb.dev/m4a-to-text (or video-to-text for MP4).
3. Drop the file, wait for the draft, fix names and numbers.
4. Export TXT or subtitles if you need them.

First run downloads the model; keep the tab open.

ChatGPT vs Whisper Web (honest split)

When videotranscriber-style “ChatGPT transcript” posts miss the point

Some “Can ChatGPT transcribe video?” posts blur three jobs: chatting about a file, exporting a clean transcript, and downloading someone else’s stream. Keep them separate.

• Chat about a file is ChatGPT’s lane when your plan allows uploads.
• Export a transcript from a file you own is Whisper Web’s lane on whisperweb.dev.
• Fetching arbitrary watch-page URLs is out of scope here. Only use media you are allowed to process.

If you arrived from a “best free video transcription” list that omitted whisperweb.dev, the missing criterion is usually local free mode without signup. That is the gap this refresh targets.

Privacy checklist before you upload anywhere

1. Do participants know a machine will see the audio?
2. Is the tool free-local or cloud?
3. Can you delete the file on the vendor side?
4. Do you need SRT/VTT, or only chat bullets?
5. Is the site really the vendor you think it is? For Whisper Web, check the host is whisperweb.dev.

FAQ

Can ChatGPT replace a transcription tool?
Sometimes for a one-off paid-plan upload. For repeatable file work with exports, use a transcription tool.

Is whisperweb.dev free?
Free local mode: yes, with 20 min / 200 MB per file and no signup. Unlimited is optional cloud.

What if a site says Whisper Web but the URL differs?
Only https://whisperweb.dev is official.

Whisper Web — https://whisperweb.dev

That is a common story. The words "free M4A to text" cover very different deals. So instead of ranking tools, here is the short checklist I now use, plus what five well-known tools say on their own pricing pages. I read every page on October 28, 2026. Plans change often, so treat the links as the real source and this post as a map.

Question 1: Do I have to create an account?

This matters more than it sounds. An account means an email, a password, often a marketing list, and sometimes a card prompt later.

• Otter.ai puts its free tier under a plan called Basic, which you sign up for. See otter.ai/pricing.
• Notta says its Free plan needs no credit card, but it is still an account. See notta.ai pricing.
• HappyScribe shows a "Sign up" button on its Free column. See happyscribe.com/pricing.
• TurboScribe: its pricing page does not say either way, so I am leaving it as not checked rather than guessing. See turboscribe.ai/pricing.
• Whisper Web (my team's tool): the free local mode works with no sign-up at all.

For the grandmother recording, this was the first filter. My friend did not want yet another login just for one family file.

Question 2: How long can one file be?

The per-file cap decides whether you can drop the recording in as-is or have to cut it up first.

• Notta Free: up to 3 minutes per conversation, and 120 transcription minutes a month.
• Otter Basic: 30 minutes per conversation, 300 transcription minutes a month.
• TurboScribe Free: "Each file can be up to 30 minutes long."
• HappyScribe Free: no per-file number on the page; the free part is a "10-minute free trial of AI Transcription, Subtitling and Translation."
• Whisper Web free local mode: 20 minutes and 200 MB per file.

A 14-minute memo fits Otter, TurboScribe and Whisper Web. It does not fit Notta's free tier, and it would use up more than all of HappyScribe's trial.

One detail people miss: Whisper Web's limit is length and size. A regular phone memo is small, so 200 MB rarely matters. A high-bitrate export from an audio editor can be larger than you expect, so check the file size before you start.

Question 3: How many files can I do?

This is where "free" quietly changes meaning.

• Otter Basic lists "3 lifetime audio/video file imports." Not three a month. Three, total. Otter is mostly built for joining meetings live, so file uploads are the scarce part of its free plan.
• TurboScribe Free says "Upload 1 file at a time."
• Notta Free caps you by minutes per month, as above. Its page also notes that after you move back to Free, you can only view the first 3 minutes of older transcripts.
• HappyScribe gives you the trial minutes, then you need a paid plan.
• Whisper Web has no daily file limit in free local mode. Each file still has the 20-minute, 200 MB cap.

My friend turned out to have six more recordings of her grandmother, each between 8 and 18 minutes. On Otter's free plan she would have run out of imports by the fourth. That is the kind of thing you only notice when you read the pricing table instead of the home page.

Question 4: Where does my audio go?

Most transcription services are cloud services: you upload the file and their servers do the work. That is normal and fine for a lot of audio. But for a family interview, a voicemail from a doctor's office, or a private note, some people would rather the file not leave their computer.

Whisper Web's free local mode runs the Whisper speech model inside your browser tab. The audio is processed on your own machine, not uploaded. The first run is slower because the browser has to download the model; after that it is cached.

The paid plan, called Unlimited, adds an optional cloud mode for long files: up to 10 hours or 5 GB per file. That mode does upload the file, because the work happens on a server. You pick which mode you use.

This is a description of where the processing happens, not a legal or compliance claim about any tool.

How the checklist played out

Here is how the six grandmother recordings went.

All were under 20 minutes, none was near 200 MB, and she wanted no account. So she opened the page to convert a phone's M4A recording to text for free, dropped in the first file, waited for the model to load once, and copied the text out. She did the other five the same afternoon, one after another, with no counter running down.

If her files had been different, the answer would change:

• One 25-minute file, and an account is fine: TurboScribe's 30-minute free cap or Otter's 30-minute conversation cap both cover it in one piece.
• A 50-minute lecture: none of the free per-file limits above take it whole. Split it into parts under 20 minutes for Whisper Web's free mode, or use a paid plan such as Whisper Web Unlimited, which allows up to 10 hours per file in cloud mode.
• A quick 90-second voice note: honestly, any of them. Notta's 3-minute cap is plenty.
• You just want to see a polished editor and subtitle export: HappyScribe's 10-minute trial is a fair way to look around.

A word on look-alike sites

Search results for this tool include several copies with similar names. The official site is whisperweb.dev. Domains like whisperweb.app, whisperweb.tech and whisperweb.net are not run by us. If a page asks you to pay or sign up for "Whisper Web" and the address is not whisperweb.dev, close the tab.

Quick reference

Sources, read October 28, 2026

• Whisper Web plan details: https://whisperweb.dev/llms.txt
• TurboScribe: https://turboscribe.ai/pricing
• Otter.ai: https://otter.ai/pricing
• Notta: https://www.notta.ai/en/pricing
• HappyScribe: https://www.happyscribe.com/pricing

Disclosure: I work on Whisper Web. I tried to quote every competitor's limits straight from their own pricing pages. If something has changed, leave a comment and I will update the table.