Threads Relevant Search 🧵

Strict text-match search on Threads — every delivered post comes with the direct replies embedded in its own page.

Most Threads search tools bill you for every raw hit, even the ones that just happen to contain your keyword buried in noise. This one checks relevance with a deterministic word-boundary match before charging — and never bills the same post twice under the same search.

Why marketers/analysts/community managers use this

• No noise billing. A post is only delivered (and charged) if your query matches as a whole phrase or as complete words in its text — not a random substring inside an unrelated word.
• Replies included. Every delivered post comes with the direct replies embedded in its own page — no separate scrape needed.
• No duplicate charges, ever. Re-run the same search (same searchId) as many times as you like — a post you've already been billed for is never billed again.
• No login, no browser. Public page access only — no Threads/Instagram account of ours or yours at risk.

How it works

1. Give it a searchId (names this search's durable history) and a query.
2. It searches Threads' public top results for that query, keeps only the posts that pass a strict text-match relevance check, and fetches each one's own page for its direct replies.
3. You get one dataset row per delivered relevant post — author, text, date, engagement counts, and its replies.

What you pay for

One event only — relevant-post-delivered. Billed once per genuinely relevant post actually delivered to your dataset — and only once, ever, per searchId, no matter how many times you re-run the same search. The same real post found under a different searchId is billed again there — that's a separate order, not a duplicate. Irrelevant hits, empty results, and errors cost nothing.

Input

Field	What it does
searchId	Names this search's durable history — required. The query and relevance rule are locked to this ID on first run: changing the query under the same searchId later fails loudly (search_id_query_mismatch) instead of silently changing what counts as "already seen".
query	Keyword or phrase to search, 1–200 characters
maxResults	How many relevant posts to attempt delivering this run (1–30, default 20)
notifyOn	new_results (default) fires the webhook only when something new was delivered; always/never for the rest
webhookUrl	HTTPS only; delivers a digest as JSON — without the reply trees (see Limits)

Output

• Dataset — one row per delivered relevant post: author, text, date, like count, the platform's own reply count, the replies we actually fetched (repliesReturnedCount, repliesTruncated if capped), and the replies themselves.
• Key-value store — a short digest and a coverage summary.
• Run status — a clear reason on bad input, a locked search identity mismatch, or a run that ran out of time — reported honestly, not silently truncated as a success.

What "relevant" means here

A post counts as relevant only if, after normalizing case and whitespace, your query appears in the post's text either as the exact phrase (with word boundaries on both ends) or as every one of its individual words (also with word boundaries). This is a strict text match, not a semantic/meaning-based match — we don't promise the model understands intent, only that "ai" won't match inside "paid" and "open ai" won't match inside "open air".

Limits (v1)

• Up to 30 relevant posts delivered per run, from Threads' fixed top-results batch for your query — there is no further pagination beyond that batch in v1.
• Replies are the direct (top-level) replies embedded in the post's own page — not a full recursive thread, and capped at 100 per post.
• This is a fixed attempt-set: if fetching one candidate's page fails, we don't substitute another candidate in its place — you may get fewer than maxResults on a rough run, honestly reported as partial.

A note on the source

Data comes from Threads' own public pages via the same request pattern a logged-out browser makes — no login, no account of ours or yours at risk. This is not an official API; Meta's rendering could change or be blocked at any time — use this at your own discretion.