A buying-intent post on X (Twitter) rarely announces itself. It is short, usually unpopular, and frequently does not contain your category's name — which is why keyword lists built from your own marketing vocabulary catch so little. The reliable signals are linguistic patterns about the author's situation, not terms from your product page.
Five stages, in order of value
Intent is not binary. Sorting matches into stages tells you which deserve an immediate reply and which deserve a note for later.
| Stage | What it sounds like | Action |
|---|---|---|
| Switching | "cancelling X", "migrating off X", "anyone moved from X to" | Reply within the hour |
| Evaluating | "X vs Y", "alternatives to X", "is X worth it" | Reply same day |
| Seeking | "anyone know a tool for", "recommendations for" | Reply fast — competitors will |
| Struggling | "doing this in a spreadsheet", "wrote a script", "this takes hours" | Reply with help, no pitch |
| Complaining | "why is X so slow", "X broke again" | Note it; reply only if you can help |
Counter-intuitively, the struggling stage is often the most valuable over time even though it converts slowest. Those people have a live, expensive problem and no vendor is competing for their attention, because they have not used a category term that anyone is tracking.
The linguistic markers
First person, present or recent past. "I am", "we just", "spent the morning". Third-person and future-tense posts are commentary.
A stated constraint. Team size, budget, a deadline, an integration requirement. Constraints indicate an actual situation rather than an abstract opinion.
A named incumbent. Mentioning a specific tool they use or used means they are already a buyer of something in your category.
A request marker. A question mark, "anyone", "recommendations", "how do you". These convert the post from a statement into an opening.
Effort language. "manually", "by hand", "takes forever", "every morning". This is the strongest early signal there is, and it almost never contains a category term.
Two or more markers in the same post is a reliable qualification threshold. One on its own is usually not enough.
The false positives that cost the most time
Marketers describing the problem to sell something else. Your pain phrases are also their hooks. Check the bio before the post.
Competitors doing the same thing you are. They appear in every keyword you track. Maintain an exclusion list.
Engagement bait. Accounts posting "what tool do you use for X" as a growth tactic. These are recognisable by cadence — the same account asks a similar question weekly.
Job posts. Hiring language overlaps heavily with problem language. Negate "hiring", "role", "we are looking for a".
Retrospective posts. "We used to do this manually before we found X" reads as intent to a keyword matcher and is in fact a closed deal for a competitor.
The last one is worth dwelling on: the difference between "we do this manually" and "we used to do this manually" is one word, and it is the entire difference between a lead and nothing. This is the category of error that pure keyword matching cannot avoid and semantic scoring can.
The profile check
On X, unlike Reddit, the author's profile is usually informative, and thirty seconds there prevents most wasted replies.
Does the bio indicate a role and company that plausibly buys your product. Does the account have a history, or was it created last month. Are the recent posts consistent with the one that matched, or is this a person who complains about everything for engagement. Is there a link to a real company site.
This step cannot be automated well and does not need to be. It is the cheapest filter available and the one that most improves your reply-to-response rate.
Turning this into a score
If you are doing this manually, a crude three-factor score works: stage (switching and evaluating high, complaining low), marker count (two or more markers), and fit (profile suggests a plausible buyer). Anything scoring high on all three goes to the top of the queue.
MentionSpot automates the first two by passing each match through an LLM that reads the post and returns a relevance score plus an intent stage, which is what makes the difference between a feed and a queue. The fit judgement stays with you, because it depends on knowing who your customers actually are — and so does every reply.
Calibrating against reality
Whatever scoring you use, calibrate it monthly against outcomes. Take the posts you replied to, mark which produced a response and which produced a conversation, and check whether your high-scoring items actually outperformed your low-scoring ones.
Most teams find one of two things. Either their intent definition is too narrow, and the struggling-stage posts they deprioritised converted better than the evaluating-stage ones where they were one of six vendors replying. Or their fit filter is too loose, and they are having pleasant conversations with people who will never buy. Both are fixable; neither is visible without the monthly check.
FREQUENTLY ASKED QUESTIONS
Questions about buyer intent on X
What is the strongest single intent signal on X?
Explicit switching language — cancelling, migrating off, or leaving a named competitor. The author has already decided to change vendors, so the only open question is what they move to.
Do high-engagement posts indicate higher intent?
Generally the reverse. Mundane complaints from real buyers attract little engagement, while posts that go viral are usually commentary. Applying engagement thresholds to intent queries filters out your best prospects.
How do I avoid replying to competitors and marketers?
Check the bio before replying, and maintain an exclusion list of competitor accounts and employees. These two habits remove most false positives at almost no cost.
Can intent be detected without reading each post?
Keyword matching gets you candidates, not intent — it cannot distinguish 'we do this manually' from 'we used to do this manually'. Distinguishing those requires reading, either by a person or by a model.
SCORED, NOT JUST MATCHED
Keyword matches are candidates. Intent is a judgement.
MentionSpot reads every X and Reddit match with an LLM, returns a relevance score and an intent stage, and ranks the queue so the switchers surface first.
Get the 7-day pass