You find customers on X by reading what people wrote, not by reading who they claim to be. A bio gives you a job title. A post gives you a problem, a deadline, and sometimes a budget.
Twitter search finds customers when you stop searching profiles and start searching post text. Circleboom collects public posts matching your keywords on X through the X Enterprise API, then pulls every account behind those posts into a deduplicated list you can filter, follow, add to a list, or export.
→ Twitter search to find customers
The tool is the easy half. Writing a query that catches intent instead of chatter is the work.
How to use Twitter search to find customers on X
Here is the flow, cheapest decisions first.
Set the search up before you spend anything
- Log in to Circleboom Twitter, granting access for whichever X profile will be running these searches.
- Switch to the Advanced X Search menu. Take either the archived collection or the live one that runs forward from a start date.
- Describe the posts you want in plain language, then read the AI-suggested variations before you accept one.
Narrow the query until only intent survives
- Add exclude terms for the words that drag in noise: hiring posts, giveaways, screenshot jokes, and your own brand name.
- Set the keyword match type to exact phrase when the signal is a sentence people literally type, and to contains when you want the variants around it.
- Set the window and the collection size, because both decide what you pay before a single post arrives.
Turn the posts into people
- Click Display Profiles of this search to swap the post list for the deduplicated account list behind it.
- Read the rows before you act on them, then follow, add the account to a Twitter List, or export the set as CSV.
That order matters because everything before step 7 is still editable and everything after it is already collected.
Cutting the noise words while the query is a draft costs nothing. Cutting them after 2,000 posts have landed costs you the collection you already spent.
At a glance: connect, pick your search, draft the query, exclude the noise, set the window, collect, then pivot to profiles.
Screen recording: a live keyword collection on X running from the query box through to the accounts behind the matching posts.
https://www.youtube.com/watch?v=dZ4djvmOqTo
Why X's own search hides the posts you most want
X's default search view is ranked, not complete. X's own search result FAQs describe Top results as selected through an algorithm, chosen on the popularity of a post along with other factors.
Now hold that against the shape of a real buying-intent post.
It comes from a small account. It has no thread, no image, no hook. It usually has zero likes, because nobody applauds a person asking for a recommendation.
Popularity ranking and buying intent point in opposite directions. The better a post performs, the less likely it is that the author needs anything from you.
Switching to the Latest tab helps, since X documents that view as reverse chronological with only global visibility filters applied. But Latest hands you an infinite scroll, not a dataset. You cannot dedupe it, sort it, or hand it to a colleague.
The narrowing tactics described by how to use Twitter advanced search all still apply. What they cannot do is keep the result once you close the tab.
That is the practical reason to find customers through Twitter search inside a collection tool rather than inside the timeline. X's own Twitter advanced search fields are good at narrowing. They are not built to hand you the list afterward.
What a buying-intent post actually looks like
A buying-intent post has a recognizable grammar, and the grammar is easier to search for than the topic. Five markers show up again and again:
- First person, present or future tense: "we are moving off," "I need to replace."
- A named constraint: a renewal date, a headcount, a budget, a deadline.
- A request pointed at the author's own followers: "any recommendations," "what are you all using."
- A named alternative they are already on, or one they are leaving.
- No punchline markers: no screenshot quote, no "lol," no emoji pile.
Tense is the cheapest filter you own, and almost nobody uses it. "We switched last year" is a case study. "We are switching this quarter" is a prospect. The words in between are nearly identical, so a keyword-only query treats them as the same result.
The constraint marker is the second cheapest. People who are only venting do not name a renewal date. People who are actually shopping almost always do, because the date is the reason they are posting at all.
This is where a deliberate search and keyword strategy earns its keep. You are not searching for your product category. You are searching for the sentence shape a person uses in the week before they go looking.
The three queries worth running first
Three post shapes carry most of the usable intent in any category, and each one needs a different query.
The recommendation ask. Someone points a question at their own followers: "anyone got a recommendation for," "what are you all using for," "looking for something that does."
Search the request phrasing plus one category noun. Exclude your own brand name, and exclude the word "hiring," which otherwise supplies half the false matches. This query is small and high quality. Ten rows a week is a good result.
The exit post. Someone names the thing they are leaving: "moving off," "canceling our," "finally switching from." Search the rival's name plus the leaving verb, and set the match type to contains so you catch the tense variants. This query is the one most people run backward, searching the rival's name alone and drowning in their marketing.
The evaluation post. Someone says the decision is already running: "we are shortlisting three of these," "picking a replacement before the quarter ends."
The verb is the query here, not the category. Search the deciding verb and let the category noun sit in a contains match, so the people mid-decision separate from the vendors who write about that category all day.
Point all three at the same window, then compare the account lists they produce. Overlap between the exit post and the recommendation ask is the strongest single signal you will get, because it means the same person said both things within days.
Where the accounts come from once the posts land
Every collected post carries its author, and Circleboom separates the two into views you can move between.
The post view holds the text alongside its impression count, likes and replies, retweets and quotes, bookmarks, and the moment it went up. The Display Profiles button pivots to the account view.
The account view is deduplicated. Someone who posted four matching complaints appears once, not four times.
What each account row tells you before you act
Each row carries the follower and following totals, the ratio between them, a post count, a join date, and an active or inactive flag. That is enough to judge whether the account is still worth an approach.
The join date and the active flag do most of the work here.
An account created three weeks ago that posted one perfect complaint is more likely to be a burner than a buyer.
The actions sit on the same rows. Per account you can follow, whitelist, blacklist, or open the profile. In bulk you can follow, add to a Twitter List, a mass block list, a whitelist or a blacklist, or export the whole set as CSV.
Circleboom is an official X Enterprise Developer company. The posts and profiles come out of X's own data pipes rather than a scraper, so your account is not the thing paying for the collection.
Follow actions still sit under X's platform limits, and Circleboom paces them instead of firing them all at once.
The account list is where this stops being research. It is a working list you can search X for customers with today and hand to a colleague tomorrow.
Planning to listen on the same keywords for months rather than days changes the calculation. The case made in social media listening and customer advocacy is worth reading next to this one.
For signals that are old rather than live, the Twitter historical data route collects from a past window instead of forward from today.
Is a buying signal on X the same thing as a buyer?
No, and no search tool closes that gap for you. The product documentation does not hedge on the point: a keyword match in a post does not guarantee that the account's intent matches your goal, and profiles should be reviewed before any bulk action.
What the search gives you is a set of posts whose language matches the language of someone who is shopping. Reading intent into that language is your inference, not a product capability, and it is wrong often enough that a review pass is part of the workflow rather than an optional extra.
That inference is still worth making. Reading a public move as meaningful is the whole premise behind watching smart money in real time, and the logic here is identical.
The alternative is waiting for someone to fill in a form on your site, which is a signal that arrives much later and from a much smaller pool.
The signal is real. The certainty is not.
Treat every collected account as a candidate, not a lead. Open two or three profiles before you commit to a batch action, and you will catch:
- The parody accounts.
- The ones that went quiet the week after they posted.
- The accounts suspended or switched to private since.
The review pass is also where you decide what the account is worth to you. A person asking for a recommendation wants an answer, not a pitch, and the difference between those two replies decides whether the search was worth running at all.
Summing up
The objection worth answering is that this only works if your buyers happen to post on X. Most teams assume theirs do not.
They are usually testing that assumption with the wrong query. A search for your category name returns your competitors' marketing.
A search for the sentence a frustrated person types in 40 characters returns people. Run one of each on the same week and the difference is obvious in the first twenty rows.
The other half of the objection is volume, and it is fair. You will not find hundreds of buyers a week in most niches.
You will find a handful whose language says they are deciding something right now, which is a different and better thing than a bigger list.
→ run a Twitter search for customers
Quick questions before you run one
Do I need a paid X plan to run this kind of search?
No. Circleboom collects public posts through the X Enterprise API on its own access, so your personal X plan does not change what comes back.
What it does change is what you can do on X afterward, since replies and follows still happen from your own account under X's normal limits.
Should I set an engagement minimum when I am hunting intent?
Set it at zero, or leave it off. Engagement minimums are useful when you want the loudest posts on a topic, and buying-intent posts are almost never the loudest.
A minimum of even five likes will quietly delete most of the people you are looking for.
How current does a buying signal have to be before it is worthless?
It depends on the purchase, not on the platform. A complaint about a tool someone renews annually stays useful for months.
A post asking for a recommendation this afternoon has a window measured in days, because the person will have chosen something by the time you finish reading it.
A rough test: ask how long the decision itself takes at your price point. That is roughly how long the signal keeps its value, and it is why an archive search and a live search suit different products rather than different budgets.
