--- name: happycouple-creator-partnership-agent description: Build and score relationship creator/specialist partnership target lists for HappyCouple organic validation, with strict schema, outreach angles, source-attribution checks, and Obsidian logging. tags: - happycouple - partnerships - creators - influencer-research - organic-distribution - consumer-ai --- # HappyCouple Creator Partnership Agent ## When to use Use this skill when Antoine asks to: - Find relationship creators, coaches, therapists, specialists, newsletters, podcasts, or communities for HappyCouple partnerships. - Build an influencer/creator target list for HappyCouple. - Score partnership candidates by fit, trust, distribution, AI/product-building gap, and likelihood. - Draft outreach messages for HappyCouple creator/specialist tests. - Turn HappyCouple organic distribution work into reusable Obsidian artifacts. If the request is creator/influencer research for a non-HappyCouple Consumer AI Factory venture, load `consumer-ai-creator-research` instead or alongside this skill. Use this HappyCouple skill only for the relationship-domain specifics: Emma/HappyCouple positioning, outreach constraints, attribution/logging paths, and relationship creator archetypes. ## Core stance HappyCouple should not pitch creators as “AI couples therapy.” Better framing: ```text A lightweight tool for helping your audience handle a concrete relationship moment and learn what to say next. ``` Do not start with the biggest category authorities. Use them as benchmarks, not first outreach targets. Early outreach should prioritize mid-sized specialists/creators who have: - trusted relationship/dating/marriage/attachment/intimacy audiences - repeated audience questions around acute moments - practical communication/script content - likely low AI/software-building fluency - plausible willingness to test a lightweight audience-value tool ## Known artifacts Creator target note: ```text /Users/antoinelevy/Documents/Obsidian Vault/Consumer AI Company Factory/05 Ventures/Venture 001 - HappyCouple/03 GTM/Partnerships/HappyCouple - Creator and Specialist Partnership Targets.md ``` Creator target CSV: ```text /Users/antoinelevy/Documents/Obsidian Vault/Consumer AI Company Factory/05 Ventures/Venture 001 - HappyCouple/03 GTM/Partnerships/happycouple-creator-targets-v0.1.csv ``` Outreach drafts: ```text /Users/antoinelevy/Documents/Obsidian Vault/Consumer AI Company Factory/05 Ventures/Venture 001 - HappyCouple/03 GTM/Email/HappyCouple - Partnership Outreach Messages.md ``` Source attribution check: ```text /Users/antoinelevy/Documents/Obsidian Vault/Consumer AI Company Factory/05 Ventures/Venture 001 - HappyCouple/04 Analytics/HappyCouple - Source Attribution Readiness.md ``` Process log: ```text /Users/antoinelevy/Documents/Obsidian Vault/Consumer AI Company Factory/05 Ventures/Venture 001 - HappyCouple/05 Build Logs/HappyCouple - Organic Distribution Process Log.md ``` ## Important lesson from first run Do not delegate “find 30 creators” as one broad open-ended web-research blob. It can run long, trigger tool/model failures, or return no usable final output. Better approach: 1. Split research by niche/platform or by 3–5 named targets. 2. Prefer official-site extraction first: homepage, about, blog/podcast, contact/media pages, official social links. 3. Verify identity before drafting; if a saved URL resolves to the wrong person/brand, mark the target blocked rather than improvising. 4. Use strict schemas. 5. Cap each worker’s output. 6. Merge/dedupe centrally. 7. Mark each candidate as “outreach now,” “later,” “blocked/unverified,” or “benchmark only.” 8. Save artifacts to Obsidian immediately. ## Recommended subagent split If using delegate_task, run 3–5 narrower workers: 1. Attachment / boundaries / anxious-avoidant creators 2. Marriage / chores / mental-load / repair educators 3. Dating / commitment / texting / breakup creators 4. Sex / intimacy educators 5. Podcasts / newsletters / communities Each worker must return exactly this schema: ```text name brand niche platforms primary_url secondary_urls rough_audience_size_if_available recent_relevant_content_examples why_fit cautions contact_path outreach_angle benchmark_or_outreach_now ``` Cap each worker at 10–15 candidates. ## Scoring rubric Score each 1–5: - Audience fit: does their audience have HappyCouple problems? - Trust: do followers likely believe their advice? - Distribution: can they move users? - Partnership likelihood: would they plausibly reply/collab early? - Brand safety: controversy, clinical boundary issues, or mismatch risk. - AI/product-building gap: do they likely have distribution/domain expertise but limited product-building ability? Suggested total max = 30 if including AI/product-building gap, or 25 if using the original five-factor table. ## Candidate classifications Use these labels: ```text outreach_now later_after_more_traction benchmark_only avoid ``` Benchmark-only examples from the first run: - Esther Perel - The Gottman Institute - Matthew Hussey They are useful for category language and credibility benchmarks, not first cold outreach. First-wave candidate archetypes: - mid-sized attachment educators - relationship communication creators - marriage/mental-load specialists - intimacy educators with practical audience questions - dating podcasts/newsletters willing to test tools ## Partnership hypotheses ### 0. Product-accurate creator moment lab Do not pitch creators a generic `bad text → better text` before/after unless the product surface is explicitly a text rewriter. Antoine flagged this as fuzzy: if HappyCouple is broader than rewriting, a creator post with only a rewritten message does not clearly explain the product or why the audience should care. Better format: ```text messy relationship situation → hidden pattern → conversation goal → safer next words ``` What to send creators is a 3-card mini pack: 1. The messy moment from their audience. 2. What HappyCouple surfaces underneath it: pattern, risk, likely goal, conversation type. 3. What to say next, framed as the output of the HappyCouple flow. Language: ```text HappyCouple is for the moment when someone has a messy relationship situation and does not know what to say next. It does more than rewrite a text: it helps identify whether the conversation is about repair, clarity, reassurance, boundary, apology, or space, then suggests safer words. ``` Product implication: a lightweight `Rewrite this before I send it` entry point may still be valuable as top-of-funnel because it is easy for creators to demo, but it should feed into the broader moment-understanding product. ### 1. Creator lead magnet ```text Send your audience here when they ask: “what should I say?” ``` Creator gets a co-branded landing page if the test progresses. Audience-language insights can be a secondary upside, but do not lead with “anonymized data/report” as the benefit. ### 2. Relationship moment challenge Creator posts 1–3 common moments. Audience tries HappyCouple with one situation. Measure: ```text landing → first message → report viewed → save/signup intent ``` ### 3. Content research swap HappyCouple mines audience/Reddit moment patterns and gives creators content themes/scripts. HappyCouple gets distribution and qualitative feedback. ### 5. Comment-to-tool bridge Antoine sharpened the strategy: do not only ask “which creators have the right audience?” Ask which creators are already receiving acute audience comments like “what do I say?”, “how do I respond?”, “am I overreacting?”, “he asked for space”, or “I need this script.” Pitch Emma/HappyCouple as the tool creators can point to when their audience is stuck at that exact moment. This is stronger than generic sponsorship, affiliate, or broad “AI relationship tool” positioning. Creator pitch frame: ```text Your audience already asks you what to say in messy relationship moments. Emma gives them a place to bring the exact situation and get what may be underneath + safer next words. ``` Operationally, creator discovery should include comment-section evidence where possible, not only follower count, niche, and contact info. A creator with visible `what do I say?` comments may outrank a larger generic relationship account. ## Outreach rules - Outreach should come from Antoine personally, not from a generic HappyCouple brand identity. - Every outreach message must pass the 3-part clarity test, but the final message should usually be natural prose, not visible callouts. Internally confirm the message communicates: 1. What it does: HappyCouple helps someone describe a concrete relationship moment, understand what is really happening underneath it, and figure out what to say next. 2. What I want: usually a 15-minute feedback call or tiny private/co-created test around one scenario from their audience — not sponsorship, not a big launch. 3. Why it is relevant to them: connect HappyCouple to a problem their work/audience already cares about (mental load, attachment, dating anxiety, intimacy conversations, boundaries, repair). Avoid abstract founder-y lines like “chance to be involved early in an AI project in your field” in final copy unless it is rewritten very naturally. - Do not literally include labels like “What it does:”, “What I want:”, “Why you might care:”, or “Why I thought of you:” unless Antoine explicitly requests a structured callout version. State the substance after those labels in normal language. - Include the live product link in creator outreach by default: https://www.happycouple.ai. The link can replace heavier explanatory/value-prop paragraphs. - Remove filler disclaimers like “No sponsorship ask. No big launch. Just validating whether this creates real value” from final copy unless needed for a specific recipient. If included, compress it into the ask sentence. - Personalization is subordinate to clarity. A recent-content hook should support those three ideas, not replace them. - Test channel fit as part of the machine: email, LinkedIn, and possibly Instagram DM depending on creator type. - Do not ask for a big launch. - Ask for a tiny conversation or tiny audience test. - Lead with the concrete user/audience moment, not broad company vision. - For Emma-era outreach, prefer the brand frame `Emma by HappyCouple` when using a venture-owned account or social page. Emma is the consumer persona/product; HappyCouple is the parent trust/domain wrapper. Avoid making creator-facing social/outreach feel like a generic relationship-advice brand. - Do not pitch therapy replacement. - Personalize with a recent piece of content and one concrete HappyCouple moment. - Hermes drafts; Antoine sense-checks/approves before anything is sent. Do not autonomously send creator outreach. If Antoine explicitly approves outbound in the current conversation (for example, “you should for sure” / “send them”), treat that as approval for a small guarded batch, then still run duplicate/contact checks before dispatch. - For HappyCouple creator/influencer outbound, use `happycouple@agentmail.to` as the venture-owned AgentMail sender when the user asks for agent-mail outreach or when operating a venture-owned email loop. Useful ask: ```text Would you be open to a 15-minute private review of one prototype scenario? I’m not asking for a post or sponsorship — I’d value your read on whether this could help people with the relationship moments your work already focuses on. ``` If a co-branded test comes later, frame it around helping their audience with one concrete moment. Audience-language insights can be secondary, not the headline benefit. ## Example natural outreach shape ```text Hi [Name] — Antoine here. I’m building HappyCouple. HappyCouple helps someone take one messy relationship situation, understand what is happening underneath, and figure out what to say next. For example: “[specific audience moment].” You can see it here: https://www.happycouple.ai I thought of you because [specific connection to their work/audience]. Would you be open to a 15-minute feedback call or private review of one prototype scenario around [their domain]? Worth a quick chat? Antoine ``` Keep it plain. If a sentence sounds like a founder positioning deck, cut it. ## Approved batch execution lessons When Antoine explicitly approves a creator batch for sending: 0. Before sending anything, read the current Interaction Ledger and outreach queue for the target names/emails. If a target was already contacted in the last 30 days, classify the action as a follow-up/re-touch and only send if Antoine explicitly asked for that; otherwise skip to fresh targets. Do this duplicate check before dispatching through AgentMail/Gmail/forms, not after. 1. Re-verify each contact path immediately before sending. Do not use privacy/deletion emails, customer-support-only emails, shop/order emails, or press/literary-agent addresses for creator/product outreach unless they are the only official route and the fit is clearly defensible. 2. Prefer official direct email when visible on the creator/site contact page; use one recipient per target unless there is a clearly designated team inbox. 3. For forms, submit only if there is no CAPTCHA/login/paywall and the form terms do not reject pitches/partnerships. If reCAPTCHA/Turnstile/arithmetic CAPTCHA/spam protection appears, mark blocked rather than bypassing. 4. Split execution into small chunks. A 10+ email batch can time out at the tool layer even after messages have been sent; verify through Gmail search for the subject/thread IDs before retrying to avoid duplicates. 5. Log every sent/submitted/blocked target in both the batch artifact and the Interaction Ledger with channel, destination, Gmail thread/message ID or form confirmation, UTM content, response state, and follow-up date. 6. Schedule a 48-72h reply check that drafts follow-ups only; do not send follow-ups without Antoine approval. ## 48h no-response follow-up and expansion loop Creator outreach should not be a one-shot batch. If no meaningful response arrives within 48 hours: 1. Search only HappyCouple-scoped Gmail/outreach context for replies; do not perform cross-mailbox searches unless Antoine explicitly asks. 2. Classify each sent target: no response, auto-response only, positive, neutral, negative, bounced, unclear. 3. For no-response targets, draft one concise follow-up email. Do not send without Antoine approval. 4. For auto-response/bounce/no-response targets, find one alternate path where possible: LinkedIn, Instagram, official site form, podcast form, or newsletter reply. Verify identity by matching official site/social/brand signals; mark blocked/unverified rather than guessing. 5. If there are no meaningful positive responses, source 5-10 new outreach_now creators/specialists from adjacent niches. 6. Update the Creator Outreach Queue, target note/CSV, and the loop note with outreach learning and product/GTM learning. Loop artifact: ```text /Users/antoinelevy/Documents/Obsidian Vault/Consumer AI Company Factory/05 Ventures/Venture 001 - HappyCouple/03 GTM/Partnerships/HappyCouple - Creator Follow-up and Expansion Loop.md ``` Follow-up drafts should be short, useful, and non-needy. Do not send more than one follow-up per email thread unless there is engagement, and do not submit LinkedIn/IG/forms autonomously. ## Review and tracking workflow Before each outreach draft, prepare: - creator name - profile/contact channel - recent content hook - why their audience fits - proposed tiny experiment - outreach draft - risk/caution - CTA Use the review queue: ```text /Users/antoinelevy/Documents/Obsidian Vault/Consumer AI Company Factory/05 Ventures/Venture 001 - HappyCouple/03 GTM/Email/HappyCouple - Creator Outreach Queue.md ``` Track results by channel and message variant: - sent date - channel: email | LinkedIn | Instagram | X | other - message variant - response - positive / neutral / negative - booked call - asked for demo - willing to share - objection - next step If Gmail labels are available, use company-specific labels such as: ```text HappyCouple HappyCouple/Outreach - To Send HappyCouple/Outreach - Sent HappyCouple/Outreach - Replied HappyCouple/Outreach - Positive HappyCouple/Outreach - Follow Up HappyCouple/Outreach - Not Fit HappyCouple/Partnerships - Active ``` Email labels are for inbox hygiene; the Obsidian queue/sheet is the learning system. ## Attribution gate Do not overbuild attribution before anyone is interested. Correct sequence: 1. Build target list. 2. Draft outreach. 3. Antoine sends outreach personally. 4. Have conversations. 5. If a creator agrees to share/test, define the campaign. 6. Only then create a tracked link/co-branded page and verify attribution. 7. Then let them send traffic. Before meaningful creator traffic, confirm source attribution works. Minimum requirement: - UTM/referrer/partner_slug captured on landing - attribution persisted through localStorage/session/person properties - attribution attached to custom activation events: - `first_user_message_sent` - `first_coach_response_received` - `report_viewed` - `save_report_clicked` - `signup_completed` If attribution is not ready: - outreach and conversations can continue. - do not give a creator a traffic link yet. - patch attribution only once there is a plausible traffic commitment. ## Obsidian behavior Save or update: ```text HappyCouple/Partnerships/HappyCouple - Creator and Specialist Partnership Targets.md HappyCouple/Partnerships/happycouple-creator-targets-v*.csv HappyCouple/Partnerships/HappyCouple - Partnership Outreach Messages.md HappyCouple/HappyCouple - Organic Distribution Process Log.md Consumer AI Company Factory - Build Log.md ``` Log what went poorly. Antoine explicitly wants process improvement notes so agents can be optimized over time. ## Verification Before final response: 1. Confirm files exist by reading first lines. 2. Report exact paths. 3. State whether list is outreach-ready or needs contact/current-audience verification. 4. State attribution readiness. 5. Add process lessons to the Organic Distribution Process Log.