Key takeaways
- Every comment passes through two filters before yours: YouTube's own spam systems, then whatever moderation level your channel is set to. Most creators have never opened that setting and are running the platform default.
- The four levels are none, basic, strict and hold all. Basic holds likely spam, self-promotion and gibberish; strict widens the net; hold all makes your comment section a queue you personally publish.
- Held comments sit in Studio for up to 60 days and then go. A channel that never opens the Held tab is not running a clean comment section — it is silently deleting real viewers once a quarter.
- The most damaging spam on YouTube is not the crypto copypasta, it is an impersonator in your replies wearing your avatar and promising a prize. It costs your audience money and costs you the trust that made them comment.
- Comments are not a documented ranking input. Their real value is the only free qualitative feedback you get on packaging — the viewers telling you what they thought they were clicking.
- Since June 2026 Studio can be searched by meaning rather than keyword, which turns that feedback from an unreadable pile into a question you can ask about a specific thumbnail.
The comment section is the only part of a channel that runs while you sleep, and the only part you did not make. Everything else — the video, the title, the thumbnail, the description — is something you decided on. The comments are decided by whoever showed up, and by an automated system you have probably never configured, operating on rules you have probably never read.
That is fine while a channel is small. It stops being fine around the first video that overperforms, when the volume arrives and the professionals arrive with it: sub-for-sub bots, drop-shipping links, engagement farms, and the one that does real damage — an account wearing your name and your profile picture, replying to your most enthusiastic viewers to tell them they have won something.
This is the working version of comment moderation. What the filters do before you see anything, which of the four settings you should actually be on, how to build a blocked-word list that catches spam without eating real conversation, what to do about impersonators, and — the part most guides skip — what comments are genuinely worth, which is not what the engagement-signal folklore claims.
What happens to a comment before anybody reads it
A comment does not go from a keyboard to your video. It goes through a sequence, and knowing the order explains almost every strange thing you have seen in a comment section.
First, YouTube's own spam and abuse systems look at it. That layer is neither configurable nor disclosed in detail, and comments found to violate the spam and deceptive practices policies are removed outright rather than handed to you. Second, your channel's moderation setting decides whether what survived publishes immediately or lands in a queue. Third, your blocked-word and blocked-link settings run. Fourth, your approved-users and hidden-users lists override parts of the above. Only then does a comment appear, or not.
Two consequences follow, and they explain most of the confusion. A comment can vanish without ever touching your moderation queue, because the platform removed it before your settings applied. And a comment can be held by your channel without you having chosen to hold anything, because the default was set in an interface you have not opened since.
The four moderation levels, and which one you should be on
In Studio, under Settings then Community, YouTube offers four states for comments on your videos. The labels understate the difference in workload between them.
| Setting | What it does | Your workload | The real cost |
|---|---|---|---|
| Allow all comments | Nothing is held by you; only YouTube's own filters apply | None until something goes wrong | Scam replies live publicly until a viewer reports them or you notice |
| Basic | Holds comments that may be spam, self-promotion, gibberish or potentially inappropriate | A short queue, most of it genuinely spam | A small amount of real conversation waits on you |
| Strict | Holds a broader range of potentially inappropriate comments | A longer queue with more false positives | Regulars start asking why their comments disappear |
| Hold all comments | Nothing publishes until you approve it | Every comment, every video, forever | Conversation dies — nobody replies to a comment that is not there yet |
For almost every channel the answer is basic, plus a blocked-word list, plus a habit of opening the Held tab. Basic exists precisely for the category of thing a creator would delete anyway, and it holds it before anyone has to see it.
Strict earns its place in two situations: a channel covering a subject that reliably attracts harassment, and a channel in the middle of a specific wave of abuse. It is a setting to turn on for a period and turn off again, not a default. The false positives are not theoretical — a wider net on "potentially inappropriate" catches sarcasm, profanity used affectionately, and the vocabulary of whatever your niche happens to be about.
Hold all changes what the comment section is. Conversation arrives in bursts and is answered in bursts; a queue flattens that into a trickle where the reply lands hours after the thought. Channels in legal or medical territory, and channels with a history of coordinated abuse, have reason to accept that trade. Nobody else does.
If comments are off entirely and you did not turn them off
Check your audience setting before you check anything else. A video or channel marked as made for kids has comments disabled by YouTube, not by you, and no setting on your side turns them back on while that flag is in place. That is a consequence of the children's privacy rules rather than a moderation decision — the made for kids setting explains what else it switches off, including notifications, memberships, cards and end screens.
The Held tab and the 60-day clock
Held comments do not sit in Studio indefinitely. YouTube keeps them for up to 60 days, during which you can approve or remove them; after that they are gone and so is the choice. This is the single most consequential piece of comment mechanics that nobody mentions.
Consider a channel that turned on basic moderation two years ago and has never opened the Held tab. Some of that queue is spam, and deletion by neglect is the right outcome for spam. The rest is every comment that linked to a source the viewer was being helpful about, every comment whose phrasing tripped a word someone added to a list in 2023, and every comment the classifier found ambiguous. Those people watched their comment never appear and drew the obvious conclusion about whether this channel wanted to hear from them.
The routine that fixes it is small. Open Comments in Studio, switch to Held for review, sort by newest, and work down until the spam becomes uniform. On most channels it takes a few minutes a week. What you are looking for is not a judgement call on each item — it is the two or three real comments hiding in a column of sub-for-sub requests. While you are there, learn the filters: Studio can show only comments you have not responded to, which is the fastest way to find the questions you meant to answer inside a thousand-comment video.
Blocked words: how to build a list that does not eat your audience
Blocked words are the sharpest tool available and the easiest to misuse. Any comment containing a word on your list is held for review rather than published, and the same list applies to live chat. Comments from you, your moderators, and your approved users are exempt, which matters more than it sounds — it means a list tuned aggressively does not silence the people you most want speaking.
The mistake is treating it as a profanity filter. Profanity is what YouTube's own strict setting is for, and a list of swear words mostly holds your regulars being enthusiastic. The high-value entries are structural, not rude:
- Off-platform contact patterns. The names of the messaging apps scams migrate to, plus the phrasings that accompany them. Nearly every comment scam ends with an instruction to continue the conversation somewhere YouTube cannot see it.
- Prize and selection language. "You have been selected", "claim your prize", "you are our winner". Your genuine giveaway, if you run one, is announced by you and pinned by you, so holding these costs nothing.
- Sub-for-sub vocabulary. The exact phrases are stable across years and cost you a single line each.
- Whatever your niche's spam looks like. Trading bots for finance channels, follower services for music, betting sites for sport. Look at your own Held tab for a fortnight and the list writes itself.
- The specific harassment your channel receives, if it receives any, in the exact spelling it arrives in.
Two practical notes. Keep the block-links setting separate in your head from the word list — holding every comment containing a link also catches the viewer linking to the study you cited. And accept that spammers substitute characters, space words out and swap in lookalike glyphs: a word list is a ratchet that raises the cost of spamming you, not a wall. It works because comment spam is a volume business, and anything that makes one channel individually annoying moves the operation elsewhere.
Two other controls sit alongside it. Hidden users removes a specific channel's comments from everything you have published, without notifying them — YouTube's documentation says their previous comments should disappear from public view within about 48 hours. Moderators can review, approve and delete comments and hide users on your behalf, which is the only mechanism that scales past one person and the one most channels adopt far too late.
Why your own comments disappear on other people's videos
The same machinery explains the other half of the question, the one creators ask when they are the ones commenting. A comment you posted that nobody can see has a short list of possible causes, and they are worth ruling out in order.
- The channel holds comments. Basic, strict or hold-all moderation on their side. You will see your comment; nobody else will, until they approve it.
- You matched their blocked-word list. Nothing to do with you, everything to do with a word list you cannot see.
- You included a link or an @handle. A frequent trigger and a frequent false positive.
- The channel hid you. Silent by design and effectively permanent.
- YouTube removed it under its spam or harassment policies, in which case it never reached the creator's queue.
- It is a display bug. These are real and recur — in June 2026 a widely reported fault left many viewers seeing only a handful of comments on videos that had hundreds, with Google acknowledging the problem while a fix went out.
The useful diagnostic is to check the comment in a logged-out browser or incognito window. If it is visible to you signed in but invisible signed out, it is being held or hidden rather than deleted. If it is gone in both, it was removed.
The impersonator in your replies
Everything above is housekeeping. This is the part that does actual harm.
The scam is simple and well documented by security researchers. An account copies your channel name, your handle as closely as it can manage, and your profile picture. It then replies to comments on your videos — not to everyone, but to the people who wrote something thoughtful, because those viewers are the likeliest to believe the creator noticed them. The reply congratulates them on winning a giveaway and asks them to continue on a messaging app. On the other end, the "winner" is asked for a delivery fee, a verification payment, or a wallet address. There is no prize. There never was a giveaway.
Two things make this worse than ordinary spam. It targets precisely the viewers you cannot afford to lose, and it does not look like spam to them, because it arrives in your voice, under your picture, inside a thread they started. When it goes wrong, a good proportion of them do not conclude that a criminal impersonated a creator. They conclude that the creator's giveaway was a scam.
YouTube's impersonation policy covers this explicitly, and the coverage extends to comments rather than only to videos and channel branding. Setting up a channel using someone's name and image and then posting comments as though you were that person is a listed violation. So is copying a channel's profile picture, banner or colour scheme with the evident intent of passing as them, and so is the "Real", "Official" or "Backup" channel-name trick. Crucially, a channel that exists to impersonate can be removed in full rather than comment by comment.
Which dictates the response. Deleting the scam reply feels like the fix and is the weakest action available — the account posts another within the hour. The report with leverage is against the channel, for impersonation, with its URL and screenshots of the replies attached, because that is the one that can take down the whole operation rather than one comment. Alongside it, three defences are worth the ten minutes they take:
- A pinned comment stating your policy. One line: you never announce winners in replies, and you never ask anyone to message you on another app. Pinned, it sits at the top of the thread the scammer is working in.
- Blocked words covering the migration phrasing, as above. It will not stop a determined impersonator but it moves the easy attempts into your Held tab.
- Moderators on high-traffic videos. The window that matters is the first day or two after publication, which is exactly when you are least able to watch the comment section yourself.
The pinned comment is packaging real estate
A pinned comment is the only text you control that sits below the fold on every view, and it is competing with nothing. Most channels waste it on a link nobody clicks. It is better used for the thing the thumbnail and title could not fit: the caveat that keeps the video honest, the correction that would otherwise generate two hundred comments, or the anti-scam line above. If your packaging makes a promise that needs one sentence of qualification, that sentence belongs there rather than crammed into the image.
Why buying comments is a worse idea than it looks
Where there is spam there is a market, and part of that market sells engagement to creators rather than at them. YouTube's fake engagement policy prohibits anything that artificially inflates views, likes, comments or other metrics, whether through automated systems or through content that exists to incentivise engagement. Enforcement works like the rest of the Community Guidelines system: content removed, normally a warning for a first violation that can be cleared with the policy training and expires after 90 days, then strikes, then termination at three inside a 90-day window. The guide to Community Guidelines strikes covers that escalation in detail.
Separately from enforcement, YouTube periodically removes artificial engagement from public counts, so the metrics tend not to survive anyway. But the strongest argument against bought comments has nothing to do with policy. The comment section is a measuring instrument — the only place viewers tell you, unprompted and for free, what they expected from your packaging and whether they got it. Filling it with purchased text is not gaming a ranking signal. It is spraying paint on the one gauge you own.
What comments are actually worth
The standard claim is that comments boost the algorithm. It is repeated everywhere and it is not something YouTube has documented. What YouTube does describe, repeatedly and in public, is a recommendation system driven by personalised prediction: what a specific viewer has watched, what they clicked, how long they stayed, and — increasingly — how satisfied they say they were, measured through survey responses rather than inferred from time alone. YouTube's growth leadership has been explicit that time spent is not assumed to be valuable in itself, and that the weighting of signals is not constant between viewers or even for the same viewer over time. The rundown of what changed in the algorithm in 2026 goes through that in more depth.
Nothing in that picture puts your comment count into a ranking formula. A comment is evidence that someone engaged; it is not a lever. A video with a lively comment section is usually a video that held attention, and the attention was the thing being measured. That does not make comments unimportant — it relocates their importance. They are worth real effort for three reasons, none of which is the algorithm:
- They are packaging feedback. "I thought this was going to be about X" is the most valuable sentence a viewer can write, and it arrives free.
- They are the idea pipeline. The question asked forty times under one video is a video, and it comes pre-validated by demand.
- They are the audience relationship. Replies, hearts and pins are cheap on your side and disproportionately remembered on theirs.
Reading comments as feedback on the thumbnail
This is the use of a comment section almost nobody makes deliberately, and it is the reason to keep it clean.
Every video you publish runs an experiment whose hypothesis is the thumbnail and title: this image plus these words will make a particular person expect a particular thing. Analytics tells you whether they clicked — click-through rate — and whether they stayed, through the retention graph. Neither tells you what they expected. The comments do, in sentences, for free.
The specific patterns to look for:
- Expectation mismatch. "I clicked because I thought this was about the new update." Repeated across a video, that is a packaging defect, and it usually shows up as a decent click-through rate with a cliff in the first thirty seconds.
- The wrong subject named. When viewers consistently reference the person, object or number in the thumbnail rather than the actual argument of the video, your image is carrying a promise the video is not built to keep.
- Questions the thumbnail should have answered. Forty people asking which model, which year, which version means the thumbnail spent its text on something less useful than the answer.
- Unprompted description of the image. Rare, and always worth reading, because it tells you what the thumbnail reads as to someone who is not you and has not been staring at it for an hour.
Pair that with the numbers rather than reading it alone. A high click-through rate with poor early retention plus mismatch comments is an over-promise; a low click-through rate with strong retention plus comments praising the depth means the packaging is underselling a good video. Neither reading is available from the comments alone or from the numbers alone, which is the argument for looking at both in the same sitting.
The 2026 Studio tools, and what they do not do
Comment tooling was static for years and then moved quickly. As of June 2026, Studio's comment search understands meaning rather than matching strings: trade press covering the rollout reported it going global on desktop on 26 June 2026, with natural-language queries of up to around a hundred characters, suggested topic filters, and a function that finds comments similar to one you are looking at. PPC Land's write-up is the clearest account of what shipped.
Alongside it, YouTube's Studio creator updates list has added comment summaries that group the themes in a video's comments, reply suggestions written in the style of your channel, and bulk hearting for comments worth acknowledging. The conversational assistant inside Studio can also summarise comment themes and sentiment as part of a wider analytics chat.
The genuinely useful one is semantic search, because it converts the packaging-feedback exercise above from an afternoon of scrolling into a question. "Comments from people who expected something different" is now a query rather than a project, run per video, in the minute after you notice a click-through rate that does not match retention.
The limits are worth stating. Summaries flatten outliers, and the outlier is frequently the comment that matters — the one person who articulated precisely what was wrong with the thumbnail is, by definition, not a theme. Suggested replies in your channel's voice are a convenience that gets less valuable the more the relationship depends on it actually being you. And none of these tools moderate; they help you read a comment section, not defend one. The defence is still the moderation level, the word list, the moderators and the Held tab.
Shorts comments are a different job
Channels publishing both formats notice quickly that the two comment sections behave differently. A Short reaches a large number of people with no relationship to the channel, which produces volume, a lower proportion of substantive comments, and a spam profile weighted towards promotion rather than conversation. Tune the blocked-word list against what your Shorts attract specifically, treat emoji-only replies as a category rather than as feedback, and accept that replying to everything does not survive a Short doing a million views. The one signal Shorts comments carry well is confusion: when a large share of them ask what the video was about, the problem is the first second, not the comment section.
A moderation routine that takes ten minutes a week
Almost all of this collapses into a short list of habits. Set once:
- Set channel-level moderation to basic in Studio under Settings, then Community, unless you have a specific reason to be stricter.
- Write the first version of the blocked-word list: contact-migration phrases, prize language, sub-for-sub vocabulary, and your niche's particular spam.
- Add your most consistent regulars as approved users so your filters never touch them.
- Add at least one moderator if anyone in your orbit is willing.
- Pin an anti-scam line on any video large enough to attract impersonators.
Then, weekly:
- Open Held for review, rescue the real comments, ignore the rest. The 60-day clock means this is the one task that cannot be indefinitely postponed.
- Scan your newest video's comments for packaging feedback, using semantic search if it is available to you.
- Report any impersonator channels rather than deleting individual replies.
- Add anything new your spam is doing to the word list.
That is the entire discipline. It is unglamorous and it compounds, in the way that most of the work that separates channels does.
The part that circles back to the thumbnail
It is easy to file comment moderation under admin — something to tidy when there is time, behind the real work of making videos. That is a fair read of the spam half and the wrong read of the other half.
Your comment section is a permanent, free, unsolicited focus group on the one decision that determines whether anything else you made gets seen. Every video, a self-selected sample of people who clicked your thumbnail tell you in plain language what they thought they were clicking. No survey you could run would produce that, and no analytics screen contains it. It is sitting there under every upload, and on most channels it is buried under crypto bots and unread because nobody spent ten minutes setting the filters.
Clean it up, read it properly, and the loop closes: the thumbnail makes a promise, the comments tell you what promise landed, the next thumbnail is better aimed. What that loop needs on the production side is more than one credible option to choose from, which is the constraint Thumblore exists to remove — enough genuinely different concepts, fast enough, that the image you publish is a choice rather than whatever the deadline allowed. Before you publish it, check how it reads where it will actually be judged with the thumbnail preview tool, and if you would rather test two candidates against real viewers than argue about them, the guide to Test & compare covers the experiment properly.