State of brand comments 2026: what brands hide, answer and automate

Brands hide far more than they delete, and most of what they hide is negativity, not spam links. In 30 days to 25 September 2026, Blabla's paying brands hid about 42 comments for every one they deleted. 72% of the hide volume came from one broad rule against negative or harmful comments; rules aimed at spam and scam links accounted for 1.3%. This report shows what brands hide, answer and automate, with the method behind every number.

Key findings

• Moderation is the main job. 91% of hides were done by automations, and hides outnumbered deletes by about 42 to 1.

• Negativity, not link spam. Only 0.6% of a random sample of hidden comments contained a link.

• Facebook carries the load. 73% of hides happened on Facebook; TikTok came second with 16%.

• Ad comments are mainstream. 65% of paid workspaces ran ads in the last 90 days, and more than half of the workspaces that hid comments hid some under ads.

• Automation sends fixed text; AI drafts. 97% of automated reply steps send approved text with variations, while AI drafts replies for a person to approve; about a third of comment reply drafts were approved.

• Comment-to-DM runs on topic words. Most keywords are tied to the post's subject, with typo variants. Generic words like LINK are rare, and PRICE or INFO were not used at all.

Methodology

This report is built from Blabla's own production data, queried read-only and reported in aggregate. No comment text is published and no customer is identified.

Limits. Volumes are concentrated: a handful of very large accounts produce most of the hide volume and receive most of the comments, so we use shares of workspaces to show how common a behavior is. The base is mostly French (86% of workspaces with enough text to classify) and mid-to-large accounts. These are brands that chose a moderation and automation tool, so they are not a random sample of all brands.

What brands hide

Paying brands mostly hide negative and harmful comments, not spam. One broad AI rule describing negative or harmful content (scams, hate, harassment, insults, misinformation) produced 71.9% of hide and delete volume. Keyword blocklists produced 4.2%, and rules aimed at spam, scam and bot links 1.3%.

A random sample of hidden comments shows why keyword and link filters catch so little:

• 0.6% contained a link, and 4.4% mentioned money, crypto or investment.

• 44.5% were classified as negative, 9.6% neutral and 8.9% positive; 37% were too short or ambiguous to classify.

• 17% were 15 characters or fewer, and 3% were emoji only.

• By category, 24% were complaints, 10% direct interactions, 6% feedback and 6% sensitive content.

The complaint share deserves a comment. Some complaints are abusive and hidden for that reason; others are criticism that a brand may be better off answering in public. Our guide to protecting brand reputation covers where to draw that line.

Rules have moved away from word lists. Almost every moderation rule runs on an AI instruction combined with sentiment and category filters; only a handful of rules in the whole paid base are keyword-only.

Where moderation happens: Facebook first, TikTok second

Instagram is almost as widely connected as Facebook but produces far fewer hides. On Instagram, the same brands use comments for conversation: it accounts for about a fifth of all public replies, and brands sent almost as many DMs there as public replies. TikTok's weight is often underestimated: about half of the paid workspaces that connect TikTok hid comments there. For TikTok-specific settings, see our guide to TikTok moderation.

Most brands run several networks: 35% of paid workspaces use two, 23% use three and 20% use all four. The most common setup is Facebook plus Instagram.

Ad comments are a mainstream need

65% of paid workspaces had ad posts in the last 90 days. Of the workspaces that hid comments during the 30-day period, more than half (54%) hid at least one comment under an ad, and for about a quarter of them, ad comments were the majority of what they hid (measured on the 60 most recent hidden comments of each workspace).

Ad threads reach people who do not know the brand, and many ads are dark posts that never appear on the page. Our guide to monitoring comments on Facebook ads explains how to find and review them.

What brands answer, and how they use AI

87% of the public replies brands published in 30 days were automated. The way they automate is conservative:

• Fixed text inside automations. 97% of automated reply steps send approved text, usually with a few variations; only 3% generate text with AI.

• AI as a drafting assistant. About half of paid workspaces received AI drafts of comment replies, and about a third of those drafts were approved. For DMs, 42% of paid workspaces received AI drafts, and fewer than one in five of those drafts was approved.

• Knowledge first. 83% of paid workspaces have FAQ entries (a median of five per workspace), 30% use saved replies and 18% have uploaded documents.

The pattern is clear: brands trust automation to repeat what they approved, and trust AI to propose, with a person deciding. Our guide to AI for community management describes the same split.

Comment-to-DM: topic words, not LINK

28% of paid workspaces run comment-to-DM automations. Of these automations, 84% are triggered by a keyword, 8% by an AI instruction (for example "only when the person shows purchase intent") and 8% by any comment.

• Topic words dominate. About 60% of comment-to-DM workspaces use a word tied to the video or post subject, usually with two to six typo variants each; these topic words trigger about two thirds of all comment-to-DM automations.

• Generic keywords are rare. LINK is almost never used, and PRICE and INFO are not used at all.

• One keyword per post. More than half of comment-to-DM automations are scoped to a specific post or video.

• Public reply plus DM. 71% of these automations also reply publicly ("check your DMs"). Almost six in ten of the buttons configured in these DMs are link buttons.

This matches the lead magnet pattern described in our guide to Instagram comment-to-DM automation: the keyword names what the person will receive.

Other automations

Shares of runs are calculated across the five automation types above. Many workspaces with comment auto-replies use them as the public half of a comment-to-DM flow: 77% of enabled reply automations are that "check your DMs" message.

Who the brands are

• Mostly French. 86% of the workspaces with enough text to classify operate mainly in French, and French accounts for 92% of comment volume.

• Mid-to-large accounts. 36% of paid workspaces have an account above a hundred thousand followers, and 6% have one above a million.

• Media carry the volume. Media and publishers produce 82% of comments processed and 73% of hides. Creators and experts account for 25% of hides and 42% of public replies.

• Volume is uneven. A handful of very large accounts receive most of the comments. The median workspace received about two hundred comments in 30 days, and about three in ten received more than a thousand.

What this means for your brand

1. Write moderation rules for negativity and abuse, not only for spam and links. Describe what you want hidden in plain language.

2. Include ad comments, dark posts included, in the same policy as organic posts.

3. Hide rather than delete. It is what brands at scale do, and it keeps decisions reversible.

4. Automate approved text; let AI draft. Keep a person on anything sensitive.

5. Use topic keywords for comment-to-DM, with typo variants and a public reply.

6. Do not forget TikTok, the second source of hidden comments.

Where Blabla fits

The behaviors above are what Blabla customers set up. Blabla's moderation hides harmful comments on Instagram, Facebook, TikTok and YouTube, including under ads (Pro plan), with AI rules written in plain language plus keyword lists. Blabla's automations send approved replies and comment-to-DM messages, and AI drafts replies in your tone from your FAQ for your team to approve. The 7-day free trial needs no credit card.

Conclusion

Across Blabla's paying brand workspaces, the picture is consistent: moderation is the main job, negativity outweighs spam, ads are part of the problem, automation repeats approved text and AI drafts. If your comment strategy still starts with a spam word list, it covers a small part of what brands at scale actually handle.