X (Twitter) advanced search: every operator that works, with examples

X’s search box is far more precise than it looks. Typed operators let you pull one account’s posts from one week, only the posts that took off, or every reply a brand got, with nothing else around them. The catch is that most operator lists online were written years ago, and some of what they list no longer works. So we tested them.

Every operator below was run on 31 October 2026. Where an example result is mentioned, it’s from that run.

The operators that work

Words and phrases

People and conversations

Dates

Dates need another word or operator next to them. At the edges, a post from late on the last day can fall either side, so add a day of margin when it matters.

Engagement

Media and links

What didn’t work

• Location search. coffee near:London within:10km returned nothing. Very few posts carry a location now, so geo operators are effectively dead.
• Verified filters. filter:blue_verified returned plenty of accounts we couldn’t confirm as verified, so we wouldn’t rely on it.
• Hashtags in other languages. #marketing min_faves:50 also matched posts with the Swedish and Dutch word “markering”. Add lang:en (or your language) to any hashtag search.

Ten searches worth saving

These are the ones social media managers and creators reach for most. Swap in your own brand, competitor or topic.

1. Everything people said to a brand this week
   to:Canva -from:Canva since:2026-10-24

2. A competitor's own posts, no replies
   from:hubspot -filter:replies since:2026-10-01

3. A competitor's videos only
   from:hubspot filter:videos

4. What took off in your niche
   "content calendar" min_faves:500 lang:en

5. Complaints about a category
   ("social media scheduler" OR "scheduling tool") (hate OR annoying OR "too expensive") lang:en

6. Who is sharing a site's links
   url:substack.com min_retweets:200 since:2026-10-01

7. Mentions with screenshots (often bugs or wins)
   @Canva -from:Canva filter:images

8. Someone's replies (where they say what they really think)
   from:levelsio filter:replies since:2026-10-25

9. Early posts before they get big
   "ugc creator" -min_faves:10 lang:en since:2026-10-29

10. A topic in one language, minus the job ads
   "social media manager" lang:en -hiring -job -filter:links

Watch out for brand names that are ordinary words. We searched (hootsuite OR buffer) "too expensive" and got posts about financial buffers, energy reserves and robot training, because “buffer” is an everyday word. For brands like that, search their handle (@buffer, to:buffer) or pair the name with words only customers would use.

A real run: what people told Canva after it launched Brush

On 29 October 2026 Canva announced Brush, its first AI model for editing designs. We searched every reply to Canva from the following days, to:Canva -from:Canva since:2026-10-25, and sorted what came back by hand. In the first 20 replies:

• Support problems hid among the launch replies. One user wrote twice that the app closed every time they tried to download a video: “Hello the app is down I currently cannot download any videos can you please assist?”
• Some people pushed back on more AI. “Ew. The last thing we want from you is more AI.”
• Others got specific about what they wanted. “Brush has to read those layer names, please stop calling everything ‘Rectangle 1’”, and two questions: “Is it available for education?” and “Wen API?”

For a social team, that’s three jobs in one search: a support ticket to route, a sentiment line for the launch report, and two feature questions worth answering in public. Doing it by scrolling the announcement’s replies would have taken far longer, and replies to other Canva posts wouldn’t have shown up at all.

Let your AI run the searches

The operators are the same in Claude or ChatGPT with Scrollhound connected, but the AI also does the reading: it groups the results, counts them and quotes the posts with their links.

Use Scrollhound to search X for to:<brand> -from:<brand> since:<date a week ago>.
Read up to 100 posts and sort them into support problems, complaints, praise and questions.
Quote two posts for each group with their links, and tell me how many posts you read.

Or describe what you want and let it pick the operators: “find this week’s most liked posts about content calendars, in English, no replies”. For a weekly version, schedule the prompt; see the weekly niche brief recipe. To size up a competitor across platforms, the competitor analysis template and the LinkedIn Ad Library guide go further.

Set it up

Add Scrollhound to Claude or ChatGPT as a connector, sign in with your email, and paste any search from this page. New accounts get 100 free reads, no card needed.