For YouTube creators

Scam comments gone before your fans see them.

The "I'm the real you, DM me on Telegram" bots, the fake giveaways, the lookalike accounts. CommentSafe hides them within minutes of posting. Real comments stay. Anything uncertain waits in a short queue for you.

Telegram bots, fake giveaways and lookalike accounts hidden within minutes. Real comments stay.

Score a comment yourself
Connects to your channel in one clickNever posts as you$19 a month, cancel anytime
I built a backyard pizza oven for $212
@KaiBuilds · posted 41 min ago
watching
MI
mira.k2m
the brick math at 4:12 saved me, thank you
Kept
KA
Kai Builds2mlooks like the channel
I'm the real Kai Builds, I didn't upload this. DM me on Telegram @kai_official_giveaway to claim your prize
Phishing 0.97Impersonator 0.96Spam 0.81Abuse 0.02
Rejected · author banned
TO
tobi_w2m
first
Kept
hidden today 27banned 4held 2
median time to hide 1m 48s
Try it

Paste any comment. See the four scores.

Every comment gets four independent judgments: phishing, impersonation, spam, abuse. Each one is a probability. Your thresholds decide what happens, and you can change them.

Reject phishing or impersonation at 0.90 or aboveHide spam or abuse at 0.90 or above, restorable for 7 daysHold anything between 0.55 and 0.90 for youKeep everything else, without storing it
Demo scores from keyword rules in your browser. On a connected channel the model scores every comment.
How it works

Connect once, then it runs on its own.

New uploads reach us by push notification and your channel's new comments are listed every couple of minutes, so a scam is usually gone within minutes of being posted.

1

Connect your channel

Sign in with Google, then grant one YouTube permission. Your channel name, picture and subscriber count come back, and a bounded first sweep of your recent and popular videos starts.

2

Every new comment is scored

Top-level comments and replies, plus whatever YouTube has already put in your held and likely-spam queues. Each one gets four independent judgments: phishing, impersonation, spam, abuse.

3

Your thresholds decide

Code decides, not the model. Scams are rejected, spam and abuse are hidden with a week to change your mind, the uncertain middle waits in a short queue, and everything else stays up untouched.

What we hide

Four categories, two very different outcomes.

Rejecting a comment on YouTube is permanent: the API cannot put it back. So only the two categories nobody ever wants back are rejected outright. Everything else is hidden first and stays restorable.

Impersonation

Rejected
"I'm the real [your channel], I didn't upload this. Message me on Telegram @…"Example

Anyone posing as you, your team, YouTube support, or handing out prizes on your behalf. Lookalike display names count as evidence, and so does the text on its own. Banning the author is off by default.

Phishing

Rejected
"Free iPhone if you verify at bit.ly/…"Example

Fake giveaways, "verify your account" links, wallet and seed-phrase asks, refund bait, and anything that moves your viewer off-platform to hand over money or an account.

Spam

Hidden · restore within 7 days
"Check out my mixtape channel" under a cooking videoExample

Unsolicited promotion, engagement bait and template text. Hidden from the public straight away; if you never touch it, a nightly job rejects it after seven days. That delay is yours to set.

Abuse

Hidden · restore within 7 days
"kys" aimed at the creatorExample

Threats, slurs aimed at a person, harassment and hate. Hidden, not rejected, because the line between abuse and a bad day is exactly where a permanent decision should not be made automatically.

What stays up

Kept
  • Harsh criticism. "This take is brain-dead, you didn't read the paper" is rude, not abuse.
  • Your own links. "Merch restock at shop.example.com" stays up when that domain is on your known-domains list.
  • Timestamps, questions, jokes, praise, and ordinary off-topic chatter.
  • Anything the model is unsure about is held for you to look at rather than hidden. A false hold costs you a click; a false reject cannot be undone.
What we ask for

One permission, in plain words.

CommentSafe asks for youtube.force-ssl, so we can hide and restore comments on your channel; we never post as you.

What that permission is used for

youtube.force-ssl
  • Reading your uploads and the comments on them, so there is something to judge.
  • Hiding a comment, which takes it out of public view but keeps it recoverable.
  • Restoring anything you want back, from the queue or the history.
  • Rejecting a scam outright. YouTube cannot republish a rejected comment, which is why it is reserved for phishing and impersonation.

What we never do

  • Never post, reply, heart, pin or edit anything as you.
  • Never store a comment that stays up. It is judged in flight and then dropped.
  • Comment text in your queue and history is deleted after 28 days, inside the 30-day limit YouTube sets for this data.
  • Disconnect revokes our access at Google, stops the worker, and leaves you able to handle anything we hid from YouTube Studio.

The full detail, including Google's limited use requirements for YouTube API data, is in the privacy policy and the terms.

Pricing

$19 a month, per channel.

Fourteen days free, so you can see what it catches on your own comments before you pay for it. One flat price per channel, no per-comment billing, cancel anytime from the billing portal.

$19per channel, per month
  • One YouTube channel
  • New comments and replies checked within minutes
  • Review queue, 28-day history, and the daily digest email
  • Your own thresholds, categories, allowlist and known domains
14 days free, then $19 a month. A card is required to start the trial. Cancel anytime.
Questions

The things worth asking first.

Can a rejected comment come back?

No. YouTube's API cannot republish a rejected comment, and there is no way to reverse a ban. That is exactly why only high-confidence phishing and impersonation are rejected, while spam and abuse are hidden first and stay restorable for seven days.

Will it hide real fans?

It is built to fail the other way. Anything the model is not confident about is held for you instead of hidden, and the reject, hide and hold lines are yours to move in settings. We would rather miss a scam than delete a fan.

Does it ban people?

Only if you turn it on. Auto-ban is off by default, applies only to impersonation and phishing rejects at or above 0.95, and every ban confirms first, because neither this tool nor the API can undo one.

How fast is it?

New uploads arrive by push notification, and your channel's new comments are listed every two minutes, easing off when nothing is happening and snapping back the moment something lands. Most scams are gone within a few minutes of being posted.

What do you store?

Comments that stay up are never stored at all. Comments in your queue or history keep their text, author name and scores for 28 days, then the text is deleted. Daily counters for your charts are aggregates and stay.

Does it cover Shorts and replies?

Yes. Shorts are ordinary videos to the YouTube API, and replies are scored alongside top-level comments. Live chat and community posts are not covered.

What happens if I disconnect?

The worker stops, we revoke our YouTube access at Google, and your stored comment text follows the same 28-day deletion. Anything we hid stays hidden on YouTube and you can still handle it in YouTube Studio.

See what is in your comments.

Connect your channel in one click, watch the first sweep, and cancel if it is not catching anything. Fourteen days, then $19 a month.