Almost every X (Twitter) monitoring setup dies the same way: it is configured in ten minutes with three obvious keywords, produces an unreadable firehose, gets muted within a week, and is quietly forgotten. The fix is not a better tool. It is spending an hour on the query list before you turn anything on.
Three buckets, three different jobs
Separate your queries by what you want them to do, and review each bucket on a different cadence. Mixing them into one feed is the root cause of most monitoring fatigue.
Intent queries are meant to produce a small number of replyable posts. They are narrow, they contain first-person language, and you check them several times a day. Low volume is the goal, not a problem.
Brand and competitor queries are meant to produce awareness. You check them once a day. Volume is whatever it is, and most entries need no action.
Discovery queries are broad category terms meant to show you how your market talks. You check them weekly, skim them, and mine them for new phrases to promote into the intent bucket. This bucket is where your query list improves over time, and it is the one people skip.
Constructing an intent query
An intent query has three parts: a problem or product anchor, a first-person or request marker, and a set of exclusions.
Take a monitoring product as the example. The anchor is "monitor twitter" or "track mentions". The marker is language like "anyone", "how do I", "looking for", "recommend", "tired of". The exclusions strip out the marketing chatter — your own brand, "webinar", "blog post", and -filter:nativeretweets.
Combined, that becomes something like: ("track mentions" OR "monitor twitter") (anyone OR "looking for" OR recommend) lang:en -filter:nativeretweets -webinar. It will return a handful of results a week. That is correct. An intent query returning fifty results a day is a discovery query wearing a disguise.
Negative keywords do most of the work
The quickest improvement available to any monitoring setup is a good exclusion list, and it is almost always built reactively: every time a useless result appears, ask what word would have excluded it, and add that word.
Four categories cover most of it. Marketing vocabulary — webinar, ebook, giveaway, "sign up", "join us". News and commentary markers — "breaking", "report says", "according to". Job and hiring language if your keyword overlaps with a role title. Your own accounts and employees, so you are not alerted to yourselves.
Expect the exclusion list to be longer than the keyword list within a month. That is a sign the setup is maturing, not a sign it was badly built.
The operators that matter for monitoring
A small subset of X's search syntax does nearly all the useful work in a monitoring context. The full reference, including the ones that behave unreliably, is in our operator guide.
| Operator | Use in monitoring |
|---|---|
| "exact phrase" | The backbone of intent queries. Unquoted multi-word terms match far too loosely. |
| OR | Group synonyms for one concept. Must be uppercase. |
| -term | Your exclusion list. The highest-leverage operator available. |
| -filter:nativeretweets | Removes retweets. Apply to essentially every monitoring query. |
| lang:en | Restricts to a language. Cuts noise dramatically for English-language products. |
| min_faves: | Useful only in discovery queries. Never apply it to intent queries. |
| -filter:replies | Original posts only. Useful for brand tracking, harmful for intent tracking. |
| filter:links | Mostly used as a negative, to strip out automated and promotional posts. |
One warning worth repeating: min_faves feels like a quality filter and is in fact an intent filter working in reverse. High-intent complaints are unpopular by nature.
Polling windows and why latency matters
Monitoring on X is not a push system. Something has to ask the platform for new posts on an interval, and the length of that interval determines your worst-case latency. A tool polling every fifteen minutes means the oldest post you see is fifteen minutes old; a tool polling hourly means an hour.
Given that the useful reply window is a few hours, anything under roughly fifteen minutes is functionally real-time and anything over an hour starts costing you conversations. The reason cheap and free tools tend to sit on the wrong side of that line is structural — API access is metered, and frequent polling across many queries consumes the allowance fast. We go into what the API realistically permits in X API limits for monitoring.
The review cadence that keeps it alive
Daily, five minutes. Work the intent bucket. Reply or dismiss. Never let it accumulate — a backlog of forty posts is a backlog of forty dead opportunities, and seeing it is demoralising enough that people stop opening the tool.
Weekly, twenty minutes. Skim the discovery bucket. Pull out two phrases you had not seen before and test them as intent queries. Retire any intent query that produced nothing qualified for three consecutive weeks.
Monthly, an hour. Review the exclusion list, add competitors that have appeared in your market, and check whether your conversion from surfaced to replied has moved. If it has fallen, the queries have drifted broader than you think.
Where scoring changes the equation
Everything above is achievable with plain keyword alerting and discipline. What plain alerting cannot do is read a post and judge whether the author is describing your problem and how close they are to acting — and that judgement is where the time goes.
MentionSpot applies an LLM pass to every match, returning a relevance score and a buyer-intent stage, so the daily five minutes is spent on the top of a ranked queue rather than on the whole feed. It finds and ranks; it does not write replies or post anything. The reply remains a human job, and it should.
FREQUENTLY ASKED QUESTIONS
Questions about X keyword monitoring
How many keywords should a monitoring setup have?
Typically fifteen to twenty-five queries spread across intent, brand and competitor, and discovery buckets. The count matters less than the separation — one undifferentiated feed is the most common cause of abandonment.
Why do my alerts return so much irrelevant content?
Almost always because the queries lack exclusions and quoted phrases. Adding a negative keyword every time a junk result appears is the single fastest improvement available.
Can I monitor X keywords for free?
Free options exist but are limited by API access: they typically poll infrequently, cover few queries, or cover only a subset of posts. For intent work where latency decides whether a reply is seen, those limits are the binding constraint.
Should I monitor my own brand name?
Yes, but in a separate bucket and with different expectations. Brand mentions are a support and reputation signal; they are rarely new leads, since the person already knows you exist.
MONITORING THAT SURVIVES MONTH TWO
A ranked queue beats an unread feed.
MentionSpot polls X and Reddit for your queries, scores each match for relevance and intent stage, and shows you the ones that deserve a reply.
Get the 7-day pass