How-To Guide
What are the social media API rate limits?
Instagram counts posts, X counts requests per endpoint, YouTube counts quota units and LinkedIn publishes nothing. Official limits and how to stay under them.
By the Superdirector editorial team. Platform limits were read from each platform's own developer documentation on 15 November 2026, and the figures below are the platforms' published wording. Where a platform publishes nothing, this page says so.
At a glance
What are the social media API rate limits?
There is no single social media API rate limit. Each platform counts something different. Instagram counts API-published posts and meters calls against impressions, X counts requests per endpoint, TikTok caps each user token at 6 requests a minute, YouTube meters cost-weighted quota units in three daily buckets, and LinkedIn publishes no standard figure at all. Stay inside them by reading the platform's own documentation, querying the account's live quota where one exists, and releasing writes from a queue instead of publishing them as they are generated.
- Instagram allows 100 API-published posts in a 24-hour moving period, and the same page also states 50. [Instagram content publishing](https://developers.facebook.com/docs/instagram-platform/content-publishing)
- X allows 10,000 calls to POST /2/tweets per app per 24 hours and 100 per user per 15 minutes. [X rate limits](https://docs.x.com/x-api/fundamentals/rate-limits)
- YouTube gives uploads and searches their own 100-call daily buckets at one unit each, with 10,000 units for other endpoints. [YouTube quota calculator](https://developers.google.com/youtube/v3/determine_quota_cost)
No two rows count the same unit, which is why a single requests-per-minute counter is wrong on four of these five platforms. TikTok's row names its window rather than a count because its daily request number is not published. Original diagram, drawn by hand from the platform documentation quoted in this section, read 15 November 2026.
Make it work for your brand
Find it. Remake it. Post it. Superdirector turns the viral formats in your niche into your brand's carousels, X posts and video scripts, and keeps the calendar full.
The short answer: no single number
There is no single social media API rate limit. Each platform counts a different thing, and the ceilings that matter are not even the same kind of measurement. Meta meters calls against a formula, Instagram meters posts against a rolling day, X meters requests per endpoint, TikTok meters requests per user token, YouTube meters cost-weighted quota units, and LinkedIn publishes no standard figure.
A rate limit is also not one ceiling. Three sit on top of each other: the request limit on the API, the publishing cap on posts themselves, and the extra caps a scheduling vendor adds before a call reaches the platform. A guide that gives one number per platform has usually collapsed all three.
The rest of this page keeps them apart. It gives the published figures, names the counting model behind each one, and shows the two habits that keep a queue inside them: read the account's own usage numbers, and release writes on a timer.
What each platform counts
The table is the reference. Every figure is the platform's own published number, and the counting model is named because it decides what your queue has to measure.
| Platform | What it counts | Published figure | Window |
|---|---|---|---|
| Instagram publishing | API-published posts | 100 in a 24-hour moving period, and 50 in the carousel section of the same page | Rolling 24 hours |
| Instagram Platform calls | Impressions | 4,800 times the account's impressions in 24 hours | Rolling 24 hours |
| Meta Graph, app token | App calls | 200 times the number of users | Rolling one hour |
| X | Requests per endpoint | POST /2/tweets: 10,000 per app per 24 hours, 100 per user per 15 minutes | 15 minutes or 24 hours |
| TikTok | Requests per user token | 6 requests per minute; the daily post cap is not published | Per minute |
| YouTube Data API | Cost-weighted quota units | 100 videos.insert and 100 search.list calls a day at one unit each, plus 10,000 units for other endpoints | Midnight Pacific |
| API calls per app and per member | Not published | Midnight UTC |
The shape to notice is that no two rows count the same unit. A limiter with one requests-per-minute counter will be wrong on Instagram, wrong on YouTube and wrong on X, because each of them meters something else.
The selection decision is separate. A comparison of the social media APIs answers which platform to build against, and the difference between API access and MCP access answers how an assistant reaches it. This page assumes the platform is chosen and the work now is to stay inside it.
Where the published numbers disagree
Instagram is the clearest case. The content publishing guide states that accounts are limited to 100 API-published posts within a 24-hour moving period. That same page, in its carousel container section, states that accounts are limited to 50 published posts within a 24-hour period. Both sentences are live in the same document.
That disagreement cannot be settled from the documentation. It can be settled from the account. The guide provides GET /<IG_ID>/content_publishing_limit, which reports current usage against the quota that actually applies. That figure is the only account-specific answer, and it is the one to schedule against.
Meta's rate-limit reference adds two counting rules that change queue design. A request that names several IDs counts once per ID, so batching reduces round trips and not consumption. And the usage header reports percentages: call_count is the share of the allowance used, not a count of calls. A team that reads 28 as twenty-eight calls learns the difference at 100.
The rule that follows is simple to state and easy to skip: when a published figure and a live endpoint disagree, the endpoint describes your account and the figure describes a policy.
The ceilings above the API limit
A publish call can fail for three different reasons, and only one of them is a 429. The request limit refuses the call, the publishing cap refuses the post, and the vendor's own anti-spam cap refuses the account before either platform number is reached.
Wrappers publish their own caps. Blotato states that it caps Instagram at 50 posts per 24 hours where Instagram's documented limit is 100, and applies similar lower caps to LinkedIn, Facebook and Pinterest. That is a legitimate choice, and it means a wrapper cannot raise a platform ceiling. It can only add a second one.
The same vendor's trial is worth reading before you test it: the trial does not include API access, so programmatic testing starts on a paid plan. A free trial can therefore tell you nothing about throughput.
TikTok's publishing cap is the other kind of silence. The documentation gives the request limit plainly, six requests per minute per user token, and names the daily post cap only as an error, spam_risk_too_many_posts. No daily number is published, so a guide that prints one is printing an observation.
One more gate sits in front of TikTok posting. All content posted by unaudited clients will be restricted to private viewing mode. The call succeeds, the post exists, and the audience is one until the client passes audit.
Read the headers, not the blog posts
Every platform that exposes usage in a response header is telling you the true number for that account at that moment. Read it after each response and let it correct your local count.
Meta returns X-App-Usage with call_count, total_cputime and total_time as percentages. A call can be throttled when total_cputime reaches 100 even if call_count is low, which usually means the request asked for too many fields. X returns x-rate-limit-limit, x-rate-limit-remaining and x-rate-limit-reset, with windows usually of 15 minutes or 24 hours and a 429 until the window resets.
TikTok returns a 429 with rate_limit_exceeded. LinkedIn returns a 429 for two different reasons, a spent quota or platform load shedding, and the body does not say which. Log the endpoint, the UTC hour and your count on every 429, because that record is the only way to tell the two apart later.
Where no header exists, the platform's own endpoint is the substitute. Instagram's content_publishing_limit is the example. LinkedIn's portal plays the same role with one catch: it lists only endpoints your app has already called today, so finding a LinkedIn limit costs one call of that limit.
Queue the writes so a burst cannot fail
An agent that generates a day of content and publishes it immediately is the shape every per-minute limit is built to reject. Queue the writes and release them on a timer. The queue also gives you one place to hold the retry policy.
- Spend one budget per account, not per app. Meta's formulas move with the account, so an app-level counter is meaningless.
- Reconcile to the platform header after every response, and treat the local count as an estimate between calls.
- Back off with jitter, and use Retry-After when the response supplies it.
- Classify before you retry. A 429 is retryable; a permission error is not, and retrying it spends the remaining budget on the same failure.
- Round-robin across accounts so one backlog cannot starve the others.
- Read the destination before you call a post published. An accepted request is a receipt, and most platforms publish asynchronously.
Two platform details belong in the queue. TikTok's upload URL is valid for one hour after it is issued, so a slow retry can lose the upload even when the publish call succeeded. YouTube charges at least one unit for an invalid request, so a malformed retry spends quota without producing a post.
Meta's own best practice is blunt. When the limit is reached, stop making calls, because continuing raises the count and lengthens the wait. Spread queries out instead of spiking them.
The queue rules of this section as one release path, with verification filled in because it is the only step that proves publication rather than acceptance. Original diagram, drawn by hand from the rules stated in this article; no vendor interface is depicted.
What this means when you post for one brand
For one brand posting a handful of times a day to one or two accounts, the published ceilings sit far above the work. A single post costs a handful of calls at most, and the daily allowances run in the thousands or the tens of thousands. The risks that stop a small account are a token that expired, a media upload that failed, and a retry loop that turned one failure into a throttle.
Volume and unattended agents change the arithmetic. Per-minute limits fail first, because an agent emits its output in a burst. Daily caps catch whatever survives the burst. LinkedIn's silence becomes a planning risk at any volume, because its ceiling cannot be computed before you spend part of it.
This is where the tool choice becomes honest. No API, MCP server or scheduling product can raise a platform's limit. The real question is who carries the accounting and who watches the queue.
Superdirector is one option, and its two plans do different jobs. Starter at $29/month is for working from Claude or ChatGPT: live social data, ideas, scripts and downloads, with the connected AI reading public content from a link and unable to post. Growth at $199/month is the operated plan: one AI-run operated account is the declared unit, two accounts are included, and extra accounts cost $99/month each. On Growth, eligible Instagram carousels and single X text posts publish by themselves once you turn posting on for an account, after one warning that names the account, the public effect and the cost. No post asks for approval after that, and you can turn it off at any time.
If you drive posting yourself, the same limits apply and the queue rules above are the work. If you would rather hand the account to an operated service, the limit becomes a question for the vendor: which platform caps do you enforce, and where do you publish them.
A first-week checklist
Six checks cover the first week, and each one produces a record rather than a feeling.
- Write down the publishing endpoint, the quota endpoint and the error code for a spent quota on each platform you use.
- Call the quota endpoint or read the usage header once, and record the number with its timestamp.
- Put every write behind one queue with one budget per account.
- Add jittered backoff and a Retry-After path before the first burst.
- Log every 429 with the endpoint, the UTC hour and the count, then review the log at the end of the week.
- Read back the destination before marking any post published.
The checklist is short because the work is not in the numbers. These numbers change, and the habit of reading them from the account survives the change.