More founders now hand their X account to an AI agent, and more agents offer to post, reply and engage for them. X allows some of this and suspends accounts for the rest. This guide reads X's Automation rules (updated April 2026) and the February 2026 change to replies through the X API for a founder deciding what an agent should be allowed to do: a short answer, a table covering each action, what the reply limit means in practice, why browser agents are a special case, a checklist for each approval and a policy you can copy. It is useful without a Soloop account and is not legal advice.
The short answer
An AI agent may draft anything for you. Through the official X API and with your consent, it may publish ordinary posts that are not spam, duplicates or posts about trending topics. Replies are narrower: X's rules allow automated replies only to people who invited them, such as someone who replied to your post or sent you a message, and since February 2026 the API on self-serve plans rejects a reply unless the original author @mentions you or quotes your post.
Automated replies to posts found by keyword search, unsolicited direct messages, automated likes, bulk following and scripting the x.com website are not allowed. An AI bot that generates replies needs X's prior written approval. You remain responsible for anything an app does with your account.
What counts as automation on X
X defines automation as accounts or apps that take repeated actions without a person actively performing them. Its rules page is written mainly for developers, but it says users are ultimately responsible for actions taken by applications they authorize, and that X may filter posts from search or suspend the account if that activity breaks the rules.
Connecting an app is not enough consent on its own. Before an app takes automated actions through your account, it must describe those actions, receive your express consent and stop when you opt out. If it substantially changes what it does, it must ask again.
The rules do not say whether a post an agent drafted and you approved one at a time counts as automation. Our reading: a post you read and choose to publish is closer to you posting with a writing tool than to a bot, which is why an approval step for each post is the safer setting for a founder's own account. X's rules still apply to the content either way.
Automation is repeated actions without a person actively performing them; users are responsible for apps they authorize; violations can lead to filtered posts or suspension.
X Automation rules (updated April 2026)OAuth authorization alone is not consent for automated actions: describe them, get express consent and honor opt-out immediately.
X Automation rules, section II.AWhat an agent may do on your X account
Each row is an action an AI agent might take. The middle column summarizes X's rules and the API change; the right column is how we would set up an agent for a founder's account.
| Dimension | What X's rules say | Founder setting |
|---|---|---|
| Draft posts and replies for you to publish | Nothing is automated until something is published. | Allow. This is the lowest-risk use. |
| Publish an original post through the API | Allowed if it is not spam, duplicative or substantially similar to other posts, and the app has your consent. | Allow after you approve each post. |
| Post automatically about trending topics | Not allowed, nor any attempt to influence trends with automation. | Block. |
| The same post on several accounts | Duplicative or substantially similar posts across accounts are not allowed. | Block. Write each account's post separately. |
| Reply to someone who replied to your post or messaged you | Rules allow one automated reply per interaction to users who opted in, with an easy opt-out. Through the API on self-serve plans, the reply must also meet the February 2026 summon rule below. | Allow after you approve each reply; expect some to be rejected by the API. |
| Reply to posts found by keyword search | Not allowed: automated replies based on keyword searches alone. Following you is not opt-in either. | Block. Use search for research and reply yourself. |
| Run an AI bot that writes replies | Requires X's prior written and explicit approval. | Do not run one on your founder account. |
| Send direct messages | Automated DMs only to people who asked to hear from you, with an opt-out; being able to message someone is not consent. | Draft only; send by hand. |
| Like posts or hide replies | Not allowed in an automated manner. | Block. |
| Repost or quote | Allowed for entertainment, informational or novelty purposes; bulk or spammy reposting breaks the rules. | Approve each one. |
| Follow, unfollow or add to lists | No bulk, aggressive or indiscriminate following or list adding; apps that promise followers are prohibited. | Block. |
| Operate x.com in a browser | Non-API automation, such as scripting the X website, may result in permanent suspension. | Draft only. See the browser section below. |
The February 2026 API reply limit
On February 23, 2026, X restricted replies made through the API to address what it called LLM-generated spam. A reply is accepted only if the original post's author has summoned the replying account: by @mentioning it in the post or by quoting one of its posts. Otherwise the request is rejected. Programmatic @mentions and quotes are limited too: in a conversation you are part of, an app may only mention or quote people already in the thread.
The limit applies to the self-serve plans (legacy Free, Basic and Pro, and pay-per-use). Enterprise access is not affected, and ordinary posts that are not replies remain supported. X's announcement does not say whether a comment under your own post counts as summoning you, so test it on your own account before you rely on an agent to answer commenters.
This is stricter than the rules page. The rules allow an automated reply to someone who replied to your post; the API on a self-serve plan may still reject it. If a tool promises to reply to anyone, ask which access level it uses and whether a person approves each reply.
Replies via the API only when the original author @mentions the replying account or quotes its post; others are blocked or rejected.
X Developers, Feb 23, 2026Self-serve tiers: legacy Free, Basic, Pro and pay-per-use. Enterprise unaffected. Non-reply posts supported with the new @mention and quote limits.
X Developers, Feb 23, 2026Enforcement against AI reply agents
X enforces these rules. On May 28, 2026, PiunikaWeb reported that Nikita Bier, then X's head of product, suspended accounts that used AI agents to reply automatically, and said that using AI for proofreading or grammar was acceptable while autonomous agents posing as human engagement were not. People who summon a bot such as Grok in a thread were not affected. This is one outlet's report, and the original posts were not linked.
Accounts that are bots by design can identify themselves with X's automated account label. A founder's personal account is not that kind of account, so keep a person deciding what it says.
Accounts automating replies with AI agents suspended; AI proofreading acceptable; summoning bots unaffected.
PiunikaWeb, May 28, 2026Automated accounts act through the X API and can label themselves so people know they are bots.
X Help CenterBrowser agents are a special case
Several 2026 agents work by operating a browser or a cloud computer: they open sites, click and type like a person. X's rules name non-API automation, such as scripting the X website, as something that may lead to permanent suspension. X does not say whether an agent operating a browser you supervise counts, so we treat it as a risk rather than a settled answer.
Some vendors draw the line themselves. xAI's guide to Grok Bot for founder-led sales sets X and LinkedIn to read and draft only, with nothing posted without approval, and when you tag @bot on X, @bot posts only a confirmation. A practical setting: let a browser agent read and draft, then publish through an official integration or by hand.
Do not use non-API automation such as scripting the X website; it may result in permanent suspension.
X Automation rules, Ground RulesFounder Mode: LinkedIn and X read and draft only; nothing sent, posted or published without approval.
xAI: Grok Bot for Founder-Led SalesTagging @bot hands the post to your Bot as a task; @bot posts only a confirmation.
xAI docs: Tag @bot on XCheck each post or reply before you approve it
Use this list on every approval card, whichever agent drafted it.
- It is a reply only if the person replied to you, messaged you, @mentioned you or quoted you. Do not approve replies to strangers found by search.
- It is not a copy or near copy of another post on this or another account you run.
- It does not jump on a trending topic or hashtag just because it is trending.
- It mentions or quotes only people already in the conversation.
- Facts, numbers and product claims are ones you can support today.
- Links go straight to the page they describe, with no misleading redirects.
- It asks for no personal information in public; anything private moves to a channel the person chose.
- You would be comfortable if the person knew an agent drafted it and you approved it.
Copy the agent policy
Paste this into your agent's instructions or your own notes, and replace the bracketed fields. It turns the table above into rules an agent and a person can both check.
Where Soloop fits
Soloop is an AI cofounder that runs individual go-to-market tasks for a founder. According to the Soloop manual, after you connect your X account Soloop can write posts and replies, publish them through the available account controls and track responses. It connects through X's sign-in for apps, and actions wait for your approval by default.
Soloop has no relationship with X, is not approved by X as a reply bot and cannot guarantee that X accepts every reply; it does not promise impressions, replies or followers. A bounded first request: My product is [URL] for [role]. Find recent public X posts where [role] asks about [problem] and summarize them with links. Draft two original posts and replies to the people who replied to my last three posts. Do not publish anything until I approve it.
Sources, limitations and refresh
The rules come from X's Automation rules page (updated April 2026) and the X Developers announcement of February 23, 2026, both read on October 27, 2026. The May 2026 enforcement is reported by one outlet. The founder settings, checklist, policy and our reading of approved drafts are editorial guidance, not legal advice or a statement from X.
X changes its rules and API access often. This guide will be reviewed by January 27, 2027, or sooner if X updates the Automation rules or its API plans.
The Soloop manual describes connecting an X account so Soloop can write posts and replies, publish them through the available account controls and track responses.
Soloop product manual