How to build a Twitter search query that returns what you meant
Most reports that come back wrong were wrong before they ran. The rules that decide what a query actually collects, and the mistakes that cost money.

Most reports that come back wrong were wrong before they ran. Not in the analysis, not in the tool, in the query. Somebody typed what they meant in the language of a search box, X read it as something else, and the tweets were paid for either way.
A query is not a sentence. It is a set of instructions with rules about what binds to what, and the rules are not obvious.
Start from the question, not from the operators
Write the question in words first. Who talked about our campaign. What people complained about after the outage. Which accounts spread the announcement.
Then pick the smallest set of terms that answers it. A query that collects more than the question needs is not more thorough, it is more expensive and harder to read.
A subject is a hashtag, a phrase or an account
Three shapes cover almost everything. A hashtag is matched whole, so #thanku does not match #thankunext. A phrase in double quotes matches those words in that order. An account can mean two different things, and the difference catches people out daily.
#COP30
"climate summit"
@nasa
from:nasa
@nasa finds tweets that mention the account. from:nasa finds tweets written by it. If you want both, ask for both.
Brackets decide what binds to what
This is the rule that ruins more queries than any other. Operators bind to the term beside them, not to everything you wrote.
#launch OR #release lang:en
That reads as everything tagged #launch, plus everything tagged #release written in English. Half the query is not filtered at all. What you meant was this.
(#launch OR #release) lang:en
Same words, different result, and nothing on screen tells you which one you got. When a query has an OR in it, put the brackets in before you do anything else.
The period is not part of the search
In the search box of X you write since: and until:, and the dates bind like any other operator, which is to say badly. In the official API the period is two parameters that sit beside the query, and in a report it is the two dates you pick.
One detail worth knowing either way. until: stops at the start of that day, so an end date of today gives you nothing from today. Leave it out and the search runs to now.
Cut the noise with a floor, not with more words
A hashtag with forty thousand tweets does not need a cleverer query, it needs a threshold. Three operators set one.
#COP30 min_reposts:300
#COP30 min_likes:500
#COP30 min_replies:100
Pick the number after looking at the conversation. Three hundred reposts is nothing in a football match and enormous in a B2B launch.
The mistakes that cost money
A plain word next to a hashtag. iPhone OR #MyEvent collects every tweet in the world that says iPhone. Your report is then mostly noise you paid for. If the word only matters inside your conversation, quote it or make it a hashtag.
AND. It is not an operator. A space between two terms already means both, and writing AND in capitals is refused.
-RT, filter:images, min_faves:. All from the old search API, all still in guides written years ago. The modern names are -is:retweet, has:images and min_likes:.
Punctuation. Commas, semicolons, a full stop at the end, a dash copied from a document, curly quotes from a word processor. X either ignores them or refuses the query, and neither is what you wanted.
A term and its own negation. (#brand -#brand) comes from editing a query rather than writing one, and it matches nothing at all.
Test it before you spend
Run the base term first and look at the volume. Then add one operator at a time. When the result collapses to nothing you know exactly which one did it, which you will not know if you add five at once.
And read the query back in words before you run it. "Every tweet containing #x or the phrase y, in Spanish, published between these two dates." If that sentence is not what you wanted, the query is not either.
We run the query, return the full report and the dataset behind it, gathered within the Twitter compliance rules. Tell us what you need to measure and we will tell you whether your query does it.