AgentKit SEO

Methodology playbook

X and Twitter Profile Optimization

Jump sections

X/Twitter guide

This module helps make an X profile and posting loop easier to understand, easier to follow, and less dependent on vague growth advice.

1. Overview

Posting more is not a strategy. A strong X presence needs clear positioning, useful native posts, and a repeatable engagement loop that fits the account’s niche.

X (Twitter) is no longer a simple chronological feed. Public engineering material and open-source repository snapshots show a multi-stage recommendation pipeline (candidate sourcing, ranking, filtering, and diversity controls). This module converts that evidence into practical constraints for profile and content optimization.

The primary audience for this module is AI agents generating or reviewing X content and humans looking to maximize their reach and authority on the platform. Where behavior is officially documented, we treat it as a rule. Where behavior is inferred from field observation, we mark it as a recommendation to test.

Weak post:
"Here are 10 tools every developer needs."

Stronger post:
"I rebuilt my portfolio for search visibility. The biggest gain was not metadata.
It was giving every serious project its own crawlable URL and proof-of-work page."

2. Use this module when

  • An X profile needs clearer positioning, bio structure, or pinned-post strategy.
  • Posts need stronger hooks, better formatting, or more useful native value.
  • The account needs a repeatable engagement loop instead of random posting.
  • Advice must separate documented behavior from contested algorithm claims.

3. Fast path

  1. For profile setup, start with profile-optimization.md.
  2. For individual posts and threads, use posting-strategy.md.
  3. For growth loops, use engagement-and-growth.md.
  4. For Premium-specific choices, use premium-strategies.md.
  5. Use algorithm-phoenix.md when a task depends on ranking assumptions.

4. Module index

Agents must route to the specific file needed for the task at hand. Do not load the entire module unless requested.

  • algorithm-phoenix.md - Historical engagement-weight baseline, documented architecture components, and clearly marked inferences.
  • profile-optimization.md - Treating the profile as a searchable landing page (bio, display name, handle).
  • posting-strategy.md - Content SEO, opening-hook heuristics, link placement experiments, and native media.
  • engagement-and-growth.md - Conversation depth, replies, and community building via SimClusters.
  • premium-strategies.md - Officially documented Premium capabilities and practical tactics.
  • sources.md - Primary and secondary sources with confidence notes.

5. Rules

Rule: When optimizing an X profile or generating a tweet, the agent must consult the relevant file from the index above before generating output.

Rule: Output must prioritize algorithmic readability and high-weight engagements (replies, retweets) over generic social media fluff.

Section: algorithm phoenix

This file separates confirmed X (Twitter) ranking behavior found in the open-source codebase from inferred behavior related to Grok and the “Phoenix” updates.

1. Overview

X open-sourced major parts of its recommendation stack (twitter/the-algorithm and twitter/the-algorithm-ml), giving high-confidence visibility into the For You pipeline architecture and historical ranking signals. Later snapshots (such as xai-org/x-algorithm) provide additional architectural detail, but should not be assumed to be a full, immutable mirror of live production.

When optimizing content, prioritize documented architecture and official platform documentation first. Treat tactical creator heuristics as testable hypotheses, not guaranteed ranking laws.

2. The historical baseline (Heavy Ranker)

The values below are historical examples from the 2023 open-source release and are most useful as directional priors. They should not be treated as guaranteed live constants.

Rule: Optimize for conversation depth and dwell time, not just likes.

  • Reply Engaged by Author (~75.0): Historically one of the strongest positive signals.
  • Reply (~13.5): A strong conversational signal.
  • Good Profile Click (~12.0): Profile visit and downstream engagement.
  • Good Click / Dwell (~11.0): Conversation click with meaningful dwell.
  • Retweet (~1.0): Positive but lower than deep conversation.
  • Favorite / Like (~0.5): Positive but comparatively low signal.

Rule: Avoid negative feedback signals.

  • Report (~-369.0): Historically a severe negative outcome.
  • Negative feedback (~-74.0): “Show less often,” mute, and block events.

3. SimClusters and Two-Tower Retrieval

X has historically sourced out-of-network content using graph and embedding-based retrieval systems.

Recommendation: Stay in your lane to improve out-of-network fit. Historically, SimClusters modeled community affinity from follow and engagement graphs. More recent architecture snapshots describe Two-Tower retrieval (User Tower + Candidate Tower) for semantic matching. Strategic takeaway: strong topical consistency can improve retrieval quality and out-of-network fit.

4. The current architecture (Phoenix and Grok)

Recent public architecture snapshots describe a Phoenix pipeline with:

  • In-network and out-of-network candidate sourcing.
  • Multi-action prediction and weighted scoring.
  • Author diversity scoring to avoid feed monotony.
  • Filtering layers for policy, safety, duplication, and user preferences.

These are useful design signals, but not a full public contract for every live ranking behavior.

Not publicly documented as deterministic rules:

  • A universal penalty for external links.
  • A strict sentiment-only rule that dictates distribution.
  • Exact posting-frequency thresholds that trigger suppression.

Recommendation: Use controlled A/B tests for contested tactics (for example, inline links vs first-reply links), and prefer clear, value-dense posts over engagement bait.

Recommendation: Pace posts to reduce self-competition in the same feed session.

5. Examples

Good example:

<!-- CORRECT: optimized for replies and conversation depth -->
We migrated our entire backend from Node.js to Go last month. Memory usage dropped by 60%, but the developer learning curve was steeper than expected.

Has your team attempted a similar migration? What was the hardest part?

Bad example:

<!-- WRONG: optimized for low-value likes and engagement bait -->
10 tools every developer needs. Like and bookmark this right now!

6. Anti-Patterns

The Like-farming thread

What it looks like: “10 tools you need. Like and bookmark this!” Why it fails: Historically, likes were much lower-weight than replies and conversation depth. What to do instead: End the thread with a specific, open-ended question to invite substantive replies, then continue the discussion in follow-up responses.

Section: engagement and growth

This file defines tactical behaviors that improve conversation depth and out-of-network retrieval, while separating documented signals from field-tested heuristics.

1. Overview

Growth on X generally improves when a creator sustains deep conversation in a consistent niche and triggers semantic matching for out-of-network retrieval. Public material supports graph and embedding-based retrieval (for example SimClusters and Two-Tower style systems), while most day-to-day posting tactics remain empirical.

Agents providing strategic advice must steer users away from “broadcast” behavior and toward “conversational” behavior.

2. Rules

2.1 The “First Hour” velocity

Recommendation: Publish when the target audience is active and monitor replies immediately. Early engagement often correlates with broader distribution, but no public fixed “first-hour threshold” is documented.

2.2 Triggering the Author Reply multiplier

The Author Reply was one of the strongest historical engagement signals in the 2023 open-source baseline.

Rule: The author must reply to thoughtful comments on their own posts, especially within the first hour. Rule: End posts with specific, open-ended questions. “What is your stack?” works better than “Agree?” because it requires a substantive answer.

2.3 Niche engagement (SimClusters & Two-Tower)

Recommendation: To build initial authority, identify 10-15 high-authority accounts within the user’s target niche. Leave high-value, constructive replies on their posts immediately after they publish. This associates the user’s account with the target audience’s embedding space (whether via follow-graph SimClusters or semantic Two-Tower models).

2.4 Out-of-network retrieval

Out-of-network reach is the primary driver of follower growth. It occurs when a post is recommended to users who do not follow the author.

Rule: Maintain topical consistency to improve out-of-network matching. When content drifts too far off-niche, retrieval quality can drop because the post is less clearly aligned with the audience’s learned interests.

3. Examples

Framing for the Author Reply

Good example:

We just migrated our database to Postgres. Query times dropped 40%, but the migration scripts were a nightmare to write.

If you've done a major DB migration recently, what was the unexpected bottleneck you hit?

Why this is good: It shares a concrete insight and asks a highly specific question. When another engineer replies, the author can easily reply back, instantly triggering the multiplier.

Bad example:

Postgres is the best database. Agree?

Why it fails: It asks a closed “yes/no” question that tends to generate low-quality, one-word replies. It provides little substance for meaningful follow-up conversation.

4. Anti-Patterns

The silent broadcaster

What it looks like: Publishing a high-quality thread, logging off, and ignoring the 20 comments it receives. Why it fails: It leaves strong conversational signals unused and trains the audience that replying is pointless, which can decrease future engagement. What to do instead: Allocate 15 minutes after publishing strictly for replying to commenters.

Section: posting strategy

This file defines how an individual post should be formatted to capture search intent, increase readability, and improve accessibility.

1. Overview

Content formatting influences how quickly users understand and engage with a post. Historical ranking snapshots suggest dwell and conversation depth matter, but exact live thresholds are not guaranteed. Treat format rules below as high-confidence best practices, with explicit notes when a claim is empirical.

2. Rules

2.1 Opening-hook heuristic (“First 50”)

There is no official public rule proving a deterministic “first 50 characters” ranking factor.

Recommendation: Place the primary keyword or strongest hook at the start of the post (often within the first ~50 characters) to improve scannability and intent matching.

2.2 Alt Text and AI visibility

Visuals are critical, but AI models and accessibility tools need text context to interpret images reliably.

Rule: Add descriptive Alt Text to every image or GIF when possible. Recommendation: Write Alt Text for accessibility first, then include relevant keywords naturally when accurate to the visual.

2.3 Media enrichment and Video

Native media often improves engagement compared to text-only posts.

Recommendation: Prioritize native video and image formats that are easy to consume on mobile. Recommendation: No official source currently documents a universal reach multiplier for a specific video duration range. Test duration and format against your own audience metrics.

2.4 Thread structuring for dwell time

Threads can increase reading time and create more reply opportunities.

Recommendation: Start with 4-8 posts when a topic needs depth, then adjust based on completion and engagement data. Rule: Use line breaks every 1-2 sentences for mobile readability.

3. Examples

Optimizing Alt Text for SEO

Good example (Alt Text for a chart):

A line graph showing the 300% increase in organic traffic for a SaaS website over 6 months after implementing programmatic SEO and fixing core web vitals.

Why this is good: It accurately describes the image for accessibility while preserving clear topic keywords (“organic traffic”, “SaaS”, “programmatic SEO”).

Bad example (Alt Text for a chart):

Screenshot 2026-04-24

Why it fails: It provides zero context to the AI or search engines, resulting in lost ranking potential.

4. Anti-Patterns

Main-body external links

What it looks like: Writing a short hook and immediately pasting a YouTube or Substack link. Why it can fail: A link-first post with low native value often gets weaker engagement. A universal platform-level link penalty is not publicly documented as a hard rule. What to do instead: Write a comprehensive, zero-click post or thread that delivers the core value natively. Place the external link in the first reply.

High-frequency burst posting

What it looks like: Publishing 4 separate posts within a 10-minute window to “flood the feed.” Why it can fail: Author-diversity components are designed to avoid repetitive feed sessions, so rapid bursts from one account may compete with each other for visibility. What to do instead: Space out individual posts by several hours, or combine related thoughts into a single formatted thread.

Section: premium strategies

This file outlines the documented product capabilities of X Premium and practical tactics that can improve outcomes for subscribed users.

1. Overview

An X Premium subscription changes product capabilities and conversation visibility, but should not be treated as a guaranteed viral-growth switch.

Agents must verify if a user has X Premium before recommending strategies reliant on long-form content or reply-ranking. If the user is on the free tier, agents should route back to algorithm-phoenix.md and posting-strategy.md for baseline optimization tactics, as the strategies below will not apply.

2. Rules

2.1 What is officially documented

Rule: Base Premium advice on official feature documentation, not assumed reach percentages.

  • Premium subscribers can publish longer posts (up to 25,000 characters).
  • Premium includes reply prioritization in conversations (officially described as preference, with levels being tested).
  • Premium users can access longer video upload limits.
  • Premium users may have access to paid Boost tools, which are separate from organic ranking behavior.

2.2 Reply prioritization

Recommendation: Leverage the Premium reply boost. Replies from Premium users can receive preferential placement in conversations. The best use of this is high-value commentary on relevant posts to earn profile visits and qualified followers.

2.3 Long-form formatting

Premium users can bypass the 280-character limit, writing posts up to 25,000 characters.

Rule: Treat long-form posts as mini-blog articles. Use markdown-style formatting (bold text, bullet points) to ensure readability. Recommendation: Use a strong opening hook to earn the “Show more” click and keep scanning easy with clear sectioning.

3. Anti-Patterns

Low-effort reply farming

What it looks like: A Premium user replying “Great post!” or ”💯” on dozens of viral tweets to exploit their prioritized ranking. Why it fails: Repetitive low-effort replies can hurt credibility and typically generate weak profile conversion, even when initial visibility is higher. What to do instead: Write a 2-3 sentence reply that adds a new perspective, provides a counter-argument, or shares a relevant data point.

Section: profile optimization

This file defines how to structure an X profile so it is easy to understand by people, searchable on-platform, and useful as a conversion point for profile visitors.

1. Overview

An X profile acts as a searchable landing page. Profiles and tweets can appear in Google surfaces such as tweet carousels, and profile clarity can improve who discovers and follows an account.

The primary audience for this file is an agent tasked with rewriting a user’s X profile. The outcome is a profile that clearly communicates authority, uses relevant keywords naturally, and provides a structured funnel for profile visitors.

2. Rules

2.1 The display name and handle

Recommendation: Include the primary keyword or job title in the display name (for example using | or -) to improve immediate clarity and people-search discoverability.

Recommendation: Ensure the handle (@username) matches the user’s brand across other platforms to build cross-platform entity trust for external search engines.

2.2 The Bio formula

The 160-character bio functions like a compact profile summary for people, search surfaces, and external previews. It should be keyword-rich without reading as spam.

Recommendation: Structure the bio using this formula: [Who I help] + [How I help] + [Social Proof] + [Call To Action]

Recommendation: Place the most critical industry keywords within the first 80 characters.

2.3 The funnel (Pinned post and Highlights)

Historical ranking snapshots suggest profile-click and downstream engagement are valuable signals, so the profile should immediately offer useful next-step content.

Recommendation: Use a Pinned Post when the account has a strong evergreen asset. This acts as the top of the funnel (e.g., a lead magnet, a high-performing thread, or a newsletter link). If there is no strong evergreen asset yet, leave the pinned slot empty rather than pinning weak content. Recommendation: If the account has access to Highlights, curate testimonials, high-signal threads, and case studies to build authority for new visitors.

3. Examples

Bio formatting

Good example:

Helping SaaS founders scale organic traffic via programmatic SEO. 📈 Grew 3 startups to 1M+ monthly visits. 👇 Get my free technical SEO checklist:

Why this is good: It follows the formula precisely. Who/How: “Helping SaaS founders scale organic traffic via programmatic SEO”. Proof: “Grew 3 startups…”. CTA: “Get my free…”. It reads naturally while hitting major keywords.

Bad example:

SEO Expert | Growth Hacker | Digital Marketing | Coffee Addict | Views are my own

Why it fails: Keyword stuffing without context. It lacks a value proposition, social proof, and a CTA, providing no reason for a user to follow or understand the account’s authority.

4. Anti-Patterns

The misleading bio

What it looks like: A profile bio filled with provocative keywords outside the core professional niche, or a pattern of aggressive follow/unfollow behavior. Why it fails: Misleading positioning attracts the wrong audience and can lead to negative feedback events such as blocks, mutes, or reports. What to do instead: Ensure the bio accurately reflects the content to avoid misleading users who might otherwise mute or report the account.

What it looks like: A bio that only says “Links below 👇” pointing to a Linktree. Why it fails: It wastes the limited bio space that search surfaces and profile visitors use to understand the account. What to do instead: Use at least 120 characters to define the user’s niche and expertise before deploying the CTA for the link.

Section: sources

This file contains the citations used by the x-twitter module. Sources are grouped by confidence level so agents can distinguish hard constraints from hypotheses.

1. Overview

The rules defined in the x-twitter module are grounded in two distinct types of evidence:

  1. Primary evidence (highest confidence): Official X documentation and historically official open-source repositories.
  2. Architecture snapshots and research: Useful for understanding possible system design, but not always a live production contract.
  3. Secondary evidence (lower confidence): External analyses and agency testing that should be treated as empirical guidance.

2. Primary sources (highest confidence)

3. Architecture snapshots and research

  • xai-org/x-algorithm (GitHub) - Public architecture snapshot of a Phoenix/Grok-based recommendation pipeline. Treat as a design signal, not as an official guarantee of live production behavior.

4. Secondary sources (empirical, use with caution)

5. Evidence handling rules for agents

  • Use primary sources for hard rules (for example, Premium features, official architecture stages, documented product behavior).
  • Use architecture snapshots for directional design guidance and label them clearly when they are not official product documentation.
  • Use secondary sources for recommendations only and phrase them as hypotheses to test.
  • Do not present contested claims as deterministic facts unless an official source explicitly documents them.