A
AIOS Wiki
read-only · public mirror
Open AIOS
Wikisaysonotesmarketing-researchsayso/notes/marketing-research/sayso-x-marketing-playbook.md

SaySo / PM Apps — X Marketing Playbook

Hand-authored·9 min read·16 sections·Last edited May 12 by initial import·View history
TL;DR

**Last updated**: 2026-05-07 **Audience**: developer-founder, time-constrained, learning marketing from zero, focusing on X (Twitter) as primary channel **Context**: companion to SAYSO-ORACLE-MARKETING-STRATEGY.md (Apr 30); this is the day-to-day operator's manual


The audience-first take

Developer-marketers tend to outperform marketing-marketers in this audience. PM/crypto/AI-quant Twitter is intensely anti-cringe and pro-craft — show your work, post your data, hold opinions backed by evidence, and don't try to sound like a marketer.

X is the right primary channel for this audience. Confirmed by every prior dossier: "Twitter is the front page of crypto. If it doesn't work on Twitter, it doesn't work." Add Discord (builder communities) and dev.to/arXiv (credibility) later; ignore TikTok / IG / LinkedIn for this ICP.


The mental model shift (first principles)

Marketing on X for technical products is not "write content, get followers, sell." It's earning the right to be in the conversation, then showing real work. Three primitives:

  • Replies > posts. Your replies in threads from accounts with audiences put you in front of those audiences for free. Your own posts (without an audience yet) reach ~50 people.
  • Show, don't describe. A screenshot of an agent's live PnL > a paragraph about what you're building. A working leaderboard > a roadmap.
  • One narrative thread, not five. "AI agents trade on Polymarket; I publish their PnL daily" is one. "MCAP infra + multi-tenant + sports mini-apps + sweeps regs + agent reputation" is five = zero traction. Pick one for 90 days.

The daily ritual (manual baseline — 30 min/day)

Treat it like a dev loop. Same cadence every day for 30 days minimum.

  1. 3 substantive replies to threads from accounts with engaged audiences. Substantive = adds data, contrarian take backed by evidence, or a shipped artifact. Never "this!" or "great thread."
  2. 1 short post of your own. Either: a data point only you have, a working artifact, or one strong opinion with evidence. ≤200 chars usually wins.
  3. 1 DM to someone whose work you've engaged with for ≥7 days. Soft, no ask: "Been following your work on X. Saw you mention Y. Curious what you'd think of Z." Relationships before you need them.

Don't post 5×/day. Don't write threads yet. Don't try to "go viral." Cadence + reply discipline beats output volume 10× for this audience.


The daily ritual (AI-assisted compression — 12 min/day)

Critical principle: AI for acceleration of your voice, not substitution. Crypto/AI Twitter has the highest GPT-slop detection rate on the internet; your audience sees through it instantly. Use AI to research, draft, and analyze; never to autopost as yourself. Rewrite every draft in your own words.

  1. Briefing (3 min) — Saved Claude/Perplexity prompt: "What happened in prediction markets, agent-trading, and AI-on-chain in the last 24h? List 5 X threads/posts with engagement. Cite sources." Skim. Pick 2–3 to reply to.
  2. Reply drafting (5 min) — Paste each thread into Claude: "Draft 3 substantive reply options that add data, a contrarian take, or a shipped artifact. ≤200 chars. Technical, opinionated, anti-cringe, no hype words." Pick the best, rewrite it in your own words, post.
  3. Substantive post (3 min) — One post. Data point from your product, screenshot of working thing, or one opinion + evidence. Claude prompt: "Agent #4 closed +12% today on contrarian Lakers market. Make this a 200-char tweet, dev voice, no emojis." Edit, post.
  4. DM (1 min) — One DM to someone you've engaged with for 7+ days. Don't AI-draft this one. The "I'm a real human paying attention" signal is the entire value.

The assets to build (in order)

For a technical product, marketing is building the artifact that markets itself, then pointing at it.

  1. A pinned tweet: 2 sentences + screenshot of the working thing. Update weekly. First version: "Building Oracle Sports — agent-vs-agent prediction game on Polymarket-style markets, free, in Telegram. Live leaderboard: [link]."
  2. A live leaderboard with named agents and PnL. Tweet a daily/weekly card from @SaySoArena (auto-generated). This becomes content forever — per the marketing strategy doc, this is your "leaderboard-as-content" flywheel.
  3. One unique data point nobody else has. E.g., "edge decay 12.3s → 2.7s on Polymarket — 30-day arb-window study" (the dossier already cites this). Become the citation.
  4. One co-authored research drop with someone whose audience you want — David Minarsch (Olas), ChainStory, a Polymarket affiliate with badge. Their audience becomes your audience overnight.

The build-your-own automation (the developer-leverage move)

While other founders manually post, you ship code that markets while you sleep. One day of work with Claude Code, then runs forever.

Spec for a marketing daemon:

  • Cron at 09:00 PT pulls top 10 agents from your DB
  • Generates a card image (Satori + Tailwind → PNG)
  • Posts to X via API as @SaySoArena: "Top 10 Agents — [date]. #1: [name] +X% week. Live: [link]"
  • Auto-tweets tier promotions in real time
  • Auto-tweets agent payouts >$100

Cost: ~$0 ongoing (X API free tier + Cloudflare Worker). Generates 7–10 high-signal tweets/week of pure data content, zero marginal effort. The single highest-ROI marketing artifact a technical founder can ship. Already specified in the marketing strategy doc as the "leaderboard-as-content" flywheel — go build it.

For the curious: there's an X MCP server (read/write tweets via Claude). Useful for one-off "draft and queue" workflows from your terminal, but the daemon above is more reliable than chat-driven posting.


Tool stack (May 2026)

Drafting + research (the core):

  • Claude Pro ($20/mo) — reply/post copilot. Save 3–5 Projects with system prompts.
  • Perplexity Pro ($20/mo) — daily briefing, fact-checking, finding quotable data. Faster than Claude for "what happened today."

Posting + scheduling (pick one):

  • Typefully ($15–29/mo) — best X-specific UI, integrated AI, scheduling, analytics. Recommended default.
  • Hypefury ($19–49/mo) — better for automation (auto-retweet evergreen, auto-DM new followers). Use if you want more set-and-forget.

Analytics + intel (free Chrome extensions):

  • Twemex — sidebar with analytics on any tweet/profile. Identifies real reach vs follower-count vanity.
  • Black Magic — engagement quality metrics. Shows which of your tweets actually performed.

Other:

  • X Premium ($8/mo) — long replies + analytics + blue check (matters for engagement on this audience).
  • Claude Code / Cursor — already in stack; the daemon above is 4–8 hours with it.
  • Twitter lists (free): segment follows into 3–5 lists — PM operators, agent builders, crypto VCs, sports-betting Twitter, regulators/lawyers. Read lists, not the algo feed. This is the #1 lever.
  • socialdata.tools (free tier) — find who follows the accounts you target.

Total monthly: ~$63. Below noise.


Claude Projects to save (5-min setup, big payoff)

  1. "PM Twitter reply drafter""Draft substantive Crypto Twitter replies for PM/AI-agent topics. Voice: technical, opinionated, evidence-backed, anti-cringe. No hype words ('revolutionary', 'disruption', 'game-changing'). ≤200 chars. Always include data point, contrarian take, or shipped artifact reference. No emojis except 📊 (data) and 🧵 (threads)."
  2. "Data-point post drafter""Convert product/market data into one tweet. ≤200 chars. Lead with the number. Imply the implication; don't moralize."
  3. "DM ice-breaker drafter""3-sentence DM to a named PM/agent builder. Reference their specific recent work. Soft, no ask. Close with a curious question, not a CTA."
  4. "Daily PM briefing""What happened in prediction markets, agent-trading, on-chain AI, Polymarket/Kalshi/Limitless ecosystem in last 24h? Top 5 conversations with most engagement. Cite source URLs."

Who to study (read 30 min/day for 2 weeks; reverse-engineer)

  • @sterlingcrispin — quant + opinionated + ships in public. Coined the "73.4% No" data point.
  • @aulekator, @manja316, @warproxxx, @suislanchez — small-account PM builders; see how they post P&L screenshots.
  • @dminarsch (Olas) — PM agent ecosystem; co-authoring with him is in the dossier as the unlock.
  • Polymarket affiliates with badges (search "Polymarket affiliate" on X) — see how they package picks. ~30% direct / 10% indirect rev share is a real lever.
  • @PolyMarketIntel, @Polymarket_News — neutral data accounts; templates for leaderboard-as-content.
  • @dwr.eth, @VitalikButerin — different audiences but watch how technical founders write replies.

For higher-tier playbooks: Andrew Chen on cold-start (andrewchen.com), Brian Chesky's Founder Mode essay (2024), Patrick Collison on writing in public. Skip generic marketing books — they're for B2B SaaS, wrong audience.


Anti-patterns to avoid (developer-specific)

  • Tutorials — near-zero engagement on Crypto Twitter; the audience already knows.
  • Long stack threads ("here's how I built X") — no audience yet to consume them.
  • "Building in public" without specifics"shipped something today" is noise; "agent #4 closed +12% today, $850 payout queued" is signal.
  • Generic crypto-bro voice — instantly detected. Be a developer, not a "founder."
  • Roadmaps before shipped artifacts"we will" always loses to "here is."
  • Feature posts without context — same feature framed as a data point gets 100× the engagement.
  • Engagement farming ("RT if you agree") — fastest way to get muted by the tier-1 accounts you actually want.

What AI doesn't solve

  • Voice authenticity — AI patterns are detectable. Rewrite every draft. Treat AI output as a first-draft, never the post itself.
  • Relationship depth — DMs that work are written by humans. AI can suggest the opening; you write the rest.
  • Showing up daily — compounding only happens with 30 consecutive days. AI doesn't solve discipline; it makes the per-day cost low enough that discipline becomes feasible.
  • Audience knowledge — AI doesn't know @sterlingcrispin posted about edge decay 3 weeks ago. You learn that by reading lists 10 min/day. The reading IS the moat.

The 30-day starter sprint

If you do nothing else:

  • Week 1: Audit follows. Build 3 lists. Pin a tweet with a working artifact. Run the daily ritual.
  • Week 2: Ship one data-point post that gets ≥10 quote-tweets from named accounts.
  • Week 3: Send 5 DMs to people you've engaged with for 14+ days. Aim for one to amplify your work.
  • Week 4: Co-author one piece with someone in their audience. Measure follower delta.

By Day 30 you should have 50–100 PM-relevant followers and one warm relationship with a known builder. If not, the issue isn't marketing technique — it's product/positioning, and the answer is in the dossier.


Minimum viable starter sprint (one Saturday, ~6 hours)

  1. Set up Typefully + X Premium + Claude Projects (1 hr)
  2. Build the leaderboard daemon with Claude Code (4 hrs)
  3. Audit follows + build 3 X lists — PM operators, agent builders, crypto VCs (1 hr)

Then run the 12-min daily ritual for 30 days. The daemon tweets 7×/week regardless of whether you show up. By Day 30: an auto-content engine, a daily reply habit, 3 quality lists you actually read, and a measurable follower delta. That's the foundation.


What NOT to do as your first move

Don't write a "launch manifesto" yet. Don't build a landing page funnel. Don't run paid X ads (CAC math is ruinous; per dossier $300–800 FTD CAC for sports books). Don't run influencer giveaways (Drake/Stake collapsed for a reason). All of that is M2+, after you have a shipped artifact + named-account engagement.


The 80/20

If you only do one thing: pin a tweet with a working leaderboard screenshot today, then reply 3× and DM 1× per day for 30 days to the named accounts in the dossier. That alone outperforms 90% of "marketing strategies."

The compounding insight

The AI-assisted approach doesn't make you 2× faster — it makes the daily ritual low-friction enough that you actually do it. The real lever is consistency × 30 days; AI only matters because it lowers the cost-per-day to a level where you'll keep showing up.