Van Born Labs / Guides / YouTube & Creators

Scrape YouTube Comments Without an API Key

The official YouTube Data API is the right tool when you have a quota. Here is what to do when you do not, and how to keep the work limited to public comments you actually need.

By Arkadiusz Van Born. Updated 2026-11-18 (America/Phoenix).

If you have ever tried to pull YouTube comments through the official YouTube Data API, you already know the bottleneck. You need a Google Cloud project, an API key, and a daily quota that is surprisingly easy to spend. Comment threads are not cheap in quota units, replies are a second request pattern, and a research job that looks modest — “the last 200 comments on these 40 videos, with replies” — can stall mid-afternoon.

That does not make the official API a bad product. If you are building a production integration for a product that YouTube has approved, or you need authenticated, supported access, you should use it and stay inside its rules. This guide is for a different, very common case: you want a structured export of public comments for research, sentiment, moderation review, or an LLM pipeline, and you do not have quota to spare.

What you can actually collect

Public comments on a public video are visible in the YouTube UI without signing in. A typical export that is useful later includes:

  • The comment text and a stable comment ID, plus a link back to the comment.
  • The author’s public handle and channel ID, and whether the author is the channel owner.
  • Like count, published time, and reply count.
  • Whether the comment is pinned, and whether the creator hearted it.
  • Replies, if you asked for them, with a parent ID so you can rebuild the thread.

You should not expect, and you should not try to obtain, data YouTube does not show publicly: private videos, members-only comments, email addresses, or anything behind a login. If a video has comments disabled, the honest result is an empty list, not a workaround.

Pinned and hearted flags are easy to underestimate. For community research they are often more valuable than raw volume. A hearted reply is a primary-source signal of what the creator is willing to endorse. A pinned comment is often an FAQ, a correction, or a call to action.

Van Born Labs

Try it live

Paste a YouTube URL or video ID. The demo returns a handful of comments. Three free checks a day.

3 free checks per day, per visitor

Why the official API is painful for this job

The YouTube Data API’s commentThreads.list and comments.list methods work, and the documentation is clear. The constraints show up in practice:

  • You must have an API key and a project with the API enabled.
  • Each page of results costs quota. Deep threads, many videos, and “newest first, then also fetch replies” multiply quickly.
  • Default quotas are finite. You can request more, but that is not a plan for a one-off research day.
  • Some environments (shared servers, cloud IPs) get additional friction even when you are well-behaved.

If your daily need is “ten videos, 100 top comments, no replies,” the official API is still a good default. If your need is “every comment on this week’s uploads, with replies, for a language model to summarise,” you will hit the ceiling.

When a scraper is appropriate — and when it is not

“Scrape YouTube comments without an API key” is a popular search, and it is also a phrase that can be used badly. A few boundaries worth stating in plain language:

  • Collect public comments for a purpose you can defend: research, journalism, brand monitoring, academic work, or building a dataset you have a right to use. Do not collect comments in order to spam authors, inflate engagement, or deanonymise people.
  • Respect volume. Pull the comments you need, not the entire history of a channel “because you can.”
  • Store only what you need, and do not republish someone’s comment as if it were your copy.
  • YouTube’s terms restrict automated access in ways that change, and unofficial collection always carries a risk of blocks or policy issues. Many research teams still do it on public data; you should go in with your eyes open, use moderate rates, and prefer a hosted actor over an aggressive homemade loop.

If you are unsure, use the official API, or collect less.

A practical workflow that stays small

A workflow I recommend, whether you write the requests yourself or you use a hosted actor:

  1. Write the question first. “Which objections appear under the last ten product-review videos?” is a question. “Get all comments” is not.
  2. Start with one video. Confirm that the fields you care about (pinned, hearted, replies, timestamps) are present in the export.
  3. Decide on sort and depth. “Top” comments are useful for popular sentiment. “Newest” comments are useful for a breaking video. Replies are useful when the question is conversational, and expensive when it is not. Cap replies per thread.
  4. Cap the count. 100 to 300 comments is enough for a first pass on most videos. You can always go back for more on the three videos that actually matter.
  5. Export to CSV or JSON, then read a sample by hand before you point a model at the file. Encoding, deleted authors, and empty text show up immediately in a sample of 20.
  6. Pair comments with a transcript when the question is about the video, not only the audience. A comment that says “the part at 8:12 is wrong” is more useful next to the caption at 8:12. A bulk transcript run, such as the YouTube Transcript Scraper, is the companion job, not a substitute.

Doing it without managing quota

If you do not want to stand up your own collector, I maintain a YouTube Comments Scraper that returns the public fields listed above, including replies, pinned and hearted flags, from a video URL, a Short, a channel, or a search term. It does not use a YouTube Data API key, so it does not spend your Cloud quota.

The Apify Store listing is the right entry point for a one-off or a scheduled run: you paste the inputs, export JSON, CSV or Excel, and you are charged per comment ($1.00 per 1,000 comments at the time of writing), not per failed video. That last point matters. A private or deleted video should not cost you a unit.

If you are wiring this into an application rather than a console, the same kind of endpoint can sit behind a RapidAPI listing (YouTube Comments API) once that listing is live; until then, the Apify API is the supported programmatic path. Either way, you bring your own account. There is no shared key.

A minimal Apify-style input for a single video looks like this in spirit: a video URL, a maximum comment count (start with 100), sort by top or newest, and includeReplies only if you need them. Run it once, open the dataset, and only then scale to a channel or a list of URLs.

What the export is good for

Used carefully, a comments export is genuinely useful:

  • Sentiment and objections. What do people argue with? Which claims in the video get challenged?
  • FAQ mining. Pinned comments and repeated questions are a ready-made FAQ for a follow-up video or a docs page.
  • Creator-endorsed replies. Hearted comments are a short list of what the channel is willing to amplify.
  • Moderation review. If you run a channel, an export with timestamps is easier to scan than the native UI when you are looking for a pattern, not a single insult.
  • LLM input. A clean JSON thread, plus a transcript, is a better prompt than a screenshot.

Used badly, the same export becomes a spam list. If your next step is “message every author,” stop.

A note on quality and failure modes

Unofficial collection is not magic. You will occasionally see empty results, truncated threads, or a video that returns no comments because the page refused the request. That is a reason to retry later, to reduce concurrency, or to fall back to the official API for that video — not a reason to rotate through aggressive workarounds.

You will also see comments that were edited, authors that no longer exist, and reply trees that were partly deleted. Keep the comment ID. It is the only stable handle you have when you re-run the job a week later and want a diff.

Finally, language and sarcasm still need a human. A sentiment model that reads “great, another change” as praise is not a research method. Sample the output.

If you want a starting point, use one video you know well, cap the count, and read the file. The official API remains the right tool when you have quota and a supported integration. When you do not, a bounded export of public comments is enough for most research questions — and it is usually enough at 200 comments, not 20,000.

Independent note: the comments scraper and transcript scraper linked above are software I make at Van Born Labs. They are not affiliated with YouTube or Google. Public data only, and no YouTube API key on those actors.

Questions

Can I scrape YouTube comments without an API key?

You can export public comments without a YouTube Data API key by using a hosted actor or a careful collector. The official API remains the right tool when you have quota and a supported integration.

What fields should a comments export include?

Text, comment id, author handle and channel id, likes, published time, reply count, pinned and hearted flags, and replies with a parent id if you asked for them.

Is it okay to collect comments in bulk?

Only for a purpose you can defend, at a moderate volume, and only for public comments. Do not use an export as a spam list, and do not try to collect private or members-only data.

Tools mentioned

Related guides

Canonical URL: https://arkadiuszvanborn.com/labs/guides/scrape-youtube-comments-without-an-api-key. Support: arkadiuszvanborn@gmail.com.