Search YouTube Comments by Meaning—but Only Across the Last 90 Days

To find YouTube comments by topic or intent, open YouTube Studio on a computer, select Community and choose the Search filter. Describe the concept in plain language instead of guessing every synonym: YouTube’s desktop guidance says Search interprets meaning and intent, accepts queries up to 100 characters and covers comments from only the last 90 days.
The feature is useful when viewers express the same question, complaint or request in different words. It creates a set of candidates to inspect; it does not judge the comments or take moderation action, so every reply, removal, report or hide decision remains yours.
Choose Search or Keywords before writing a query

Use Search when the concept is clear but the wording is unpredictable. A request for another episode might appear as “make part two,” “continue this series” or “please cover the next step.” A semantic query can retrieve comments related by intent without requiring a separate search for every phrase.
Use Keywords when the literal text matters, such as a product name, coupon code, disclosed sponsor, recurring phrase or exact error message. PPC Land’s account of the update distinguishes the new meaning-based Search filter from the older exact-match field, which remains available under the Keywords label.
- Choose Search: questions about setup, complaints about pacing, requests for future topics or reactions to a product.
- Choose Keywords: “Model X200,” “SAVE20,” “error 403,” a person’s name or another phrase that must appear verbatim.
- Use them in sequence: search by meaning to discover recurring language, then use Keywords for a precise pass once the relevant terms are known.
Write one focused category per query

A useful semantic query identifies both the subject and the kind of response you want. “Audio” is vague; “complaints about background noise” defines the topic and the viewer’s intent. Combining unrelated goals produces a mixed set that is harder to assess consistently.
These are natural-language starting points, not special commands:
- Questions: “questions about how I installed the software”; “viewers asking which camera I used”; “questions about shipping outside the US.”
- Complaints: “complaints about audio quality”; “viewers saying the tutorial moves too fast”; “problems following the final step.”
- Product feedback: “requests for missing product features”; “positive feedback about battery life”; “comparisons with competing products.”
- Content planning: “requests for a beginner version”; “people asking for part two”; “topics viewers want explained next.”
- Moderation review: “personal attacks about someone in the video”; “comments pressuring viewers to make contact elsewhere”; “hostile remarks about personal appearance.”
Run separate searches when the categories call for different responses. A product complaint may deserve a reply, while a personal attack may require a moderation decision. Keeping those queues separate makes the reason for each match easier to evaluate.
Seed Find Similar Comments with a clear example
Use Find Similar Comments when one published comment already captures the pattern. Open that comment’s three-dot menu and select the option; YouTube then looks for other comments with a similar topic or meaning. This avoids reverse-engineering all the vocabulary that might express the same idea.
Choose a seed that communicates one main issue. A comment combining praise, a technical question and an insult contains several competing signals and may produce an unfocused set. If the results drift, start again with a cleaner example or replace the seed with a narrowly written query.
This method is particularly useful for an emerging complaint whose wording was not anticipated. Once repeated terms become apparent, Keywords can provide a narrower follow-up search for a particular model, phrase or error code.
Review context before taking any action

Semantic similarity is not a policy decision. Two comments can concern the same subject while differing in tone, target and context: criticism of an idea is not automatically an attack on a person, while an offensive phrase may have been quoted rather than endorsed.
Open each comment, identify the video and inspect the surrounding reply thread where relevant. Separate comments that warrant an answer from those that may require removal, reporting or hiding a user. An irrelevant match can simply be left alone; appearing in the results is not evidence of a violation.
The TeamYouTube announcement confirms that the optional tools do not automatically remove anything and that creators decide which moderation action to take. The same announcement identifies the feature as available globally to creators on desktop.
Work within the 90-day boundary
The Search filter cannot provide a complete historical audit because it finds comments from only the last 90 days. It also excludes your own comments on your channel and cannot search for comments from a specified user. Treat the feature as a view of recent audience activity, not a channel-wide archive.
- Define one review goal, such as unanswered setup questions or complaints about a new product.
- Run one focused semantic query and inspect whether the results share the intended meaning.
- Use Find Similar Comments when a representative published comment appears.
- Run Keywords for exact names, codes or phrases discovered during the semantic pass.
- Review every candidate in context and use an existing engagement or moderation action only when warranted.
- Repeat time-sensitive reviews periodically when the category must remain visible beyond the 90-day window.
The division of labor is straightforward: Search retrieves varied wording connected by meaning, Keywords finds text you can specify exactly, and Find Similar Comments expands from a concrete example. All three help locate comments; none replaces the creator’s judgment.
Also read:
Subscribe to our newsletter
Get the latest Web3, AI, and crypto news delivered straight to your inbox.