What Threads actually returns when you are not logged in

A measured reference. August 2026.

Every number here came from a real request against threads.com, logged out, and is dated. Where we could not establish something, it says so. If you find a number that no longer holds, open an issue — a stale measurement presented as current is the same as a wrong one.

Summary

The second and third rows are the interesting ones. Meta caps each entry point rather than the total, so querying every public entry point and merging returns substantially more than any single one — without an account, and without touching anything that is not already public.

1. Replies really are capped at ~20

Request a post page logged out and you get the post plus roughly its first 20 replies. Past that, Threads renders one of:

Log in to see more replies.
Log in to see more from <username>

Both strings are client-side rendered. They are not in the HTML you receive, which is why a naive scraper reports success and silently hands you a truncated thread. Detecting the wall means rendering the page or inferring it from the reply count in the payload.

2. Search has five public surfaces, not one

Threads’ own website queries several different search endpoints:

Each is capped around 20. Each returns a different result set. Measured, deduplicated by post ID:

That is 2.4–2.7× more unique posts, from public endpoints the Threads website itself uses.

3. Profiles have four public tabs

Each holds roughly 5 posts logged out, and they hold different posts. Measured:

4. Meta serves different pages to different IPs

This is the part that surprised us, and it is not documented anywhere we could find.

Request the same profile two ways on the same day:

The practical consequences, measured on the Apify platform, same target, same day (2026-08-19):

45× slower and 15× more expensive, for identical output. If your Threads scraping suddenly got slow and expensive, this is very likely why.

Two things follow:

• Size is a usable signal. We treat a response under ~500 KB that yields no posts as “the target genuinely does not exist” and stop immediately, and a large response that yields no posts as a transient failure worth retrying. Getting this backwards burns money on missing targets and gives up on recoverable ones.
• Watch your render ratio. We log what fraction of fetches fell back to rendering. When that number climbs, Meta has started serving shells to the IP range you are on. It is the earliest warning you get.

5. Structure changes, signatures do not

Most Threads scrapers hard-code GraphQL doc_id values or fixed JSON paths. Both change without notice, and when they do the scraper returns zero rows rather than an error.

We identify a post by structural signature instead: a node carrying pk and user.username and at least one of caption, text_post_app_info or taken_at. A refactor that moves the data degrades the result instead of killing the run.

One thing to filter explicitly: Threads returns recommended posts in the same payload as genuine replies. Keys like relatedPosts, recommendedPosts and suggestedUsers sit alongside the real thread. Parse the payload naively and your “30 replies to my post” quietly include 10 strangers discussing something else.

6. Not every brand is on Threads

Worth checking before you blame your scraper. Of 40 well-known brand handles we tested, these had no Threads account as of August 2026:

tesla · spacex · apple · stripe · vercel · anthropicai

A missing target is not a failure, and a tool that counts it as one is inflating its own error rate while telling you nothing useful. Separate the two, and report which inputs were bad.

Method

Measurements were taken via the Actors this client wraps, running on Apify with residential proxies, reading only public pages — no login, no cookies, no CAPTCHA solving, no access-control circumvention.

Benchmark referenced above: 24 tasks (8 profiles, 10 posts, 6 keyword searches), 647 records, 0 failed tasks, 5.5 s average per task, 72 fetches with zero render fallbacks.

If you want this data without building it

The three Actors behind these measurements are public:

• Threads Scraper — posts, replies, profiles and search in one run
• Threads Search Scraper — keyword search, deduplicated across keywords
• Threads Creator Finder — creators ranked by engagement on your topic

Or read the client and do it yourself. The limits above apply either way — they are Meta’s, not ours.