marketing-ideas

Marketing ideas and experiments

Generate and prioritize context-specific marketing hypotheses instead of a list of supposedly generally proven tactics. The default is a selection list with a test sketch in the chat. Idea generation allows analysis but not execution.

Safety and evidence rules

Limited sources. Use only supplied information and approved project files. Do not use CRM, analytics, email, support, contact lists, browser profiles, communities or access credentials without specific approval.
Ideas are hypotheses. “Proven,” “quick win,” “inexpensive,” “best” or “high intent” only with comparable evidence. Separate observation, internal statement, third-party estimate, hypothesis and unknowns.
No execution by implication. No research, contact collection, message, publication, account action, ad, spend, installation, product/website change, discount campaign, partnership or contract without explicit approval.
No fabricated evidence. Do not invent demand, conversion, return, rankings, customer stories, recommendations, urgency or competitor weaknesses.
Privacy and consent. Do not recommend hidden tracking, fingerprinting, data broker enrichment, contact scraping, unauthorized profiling, private community analysis or audience/pixel sharing.
Platform integrity. No fake reviews, review gating, engagement pods, purchased followers, cloaking, identity confusion, rule circumvention or covert automation.
Search integrity. No doorway pages, parasitic reputation use, link manipulation, copied comparisons or mass thin content.
Fair comparisons. Current sources, neutral trade-offs and brand/legal review; no defamation, imitation or confidential competitor data.
Communication. Email, DM, press, podcast, influencer or partner outreach requires legitimate target selection, identity, consent/legal review, suppression lists, easy objection and separate send approval. No mass unsolicited sending.
Partnerships and benefits. Affiliate, referral, sponsorship, influencer, certification and integration programs require terms, disclosure, trademark rights, data roles, fraud protection, tax/accounting review and consent limits.
Actions and scarcity. Sweepstakes, discounts, trial phases, early access and giveaways require genuine terms, eligibility, regions, tax, platform, privacy, consumer and budget review. No fabricated scarcity.
Rights. Check sources and usage rights for quotes, images, logos, music, data, templates, open source, customer stories and research.
Product ideas are product projects. Tools, extensions, APIs, OAuth, imports, viral loops and open-source components require specification, security/privacy, accessibility, abuse protection, maintenance, support, testing and rollback.
Sensitive areas. Health, finance, law, work, housing, education, insurance, minors, politics and vulnerable groups require expert review; no sensitive characteristics for discriminatory selection or pricing.
Physical and reputational risks. Events, stunts, out-of-home advertising and controversial campaigns require rights, permits, safety, accessibility, consent of uninvolved persons, insurance and crisis plan.

Context brief

Ask only for shortlist-relevant information:

Product, country, language, market and business model;
Target organizations, buying group, job, trigger and exclusions;
Phase, goal, time horizon and reliable baseline;
approved statements and evidence;
existing channels, own reach and distribution advantages;
previous tests, methodology, results and learnings;
team, capabilities, budget ceiling, review and maintenance capacity;
sales cycle, value, margin and implementation/support load;
data protection, security, legal, brand and accessibility limits;
risk appetite and expressly excluded actions.

An approved .agents/product-marketing.md can serve on a project-local basis. Do not automatically read private systems.

Idea generation process

1. Formulate the problem as a decision

"Grow," "generate leads" or "build authority" translate into target audience and behavior, funnel/customer phase, baseline/source, outcome/time period, practical limit, early/late metric, guardrail metrics and follow-up decision. Fast acquisition is not automatically paid or outbound; speed depends on access, offer, sales cycle, proof, creative material, measurement, budget and capacity.

2. Build an evidence register

ID	Signal/statement	Class	Source/date	Relevance	Confidence	Unknowns
E-01		Observation / Internal / Estimate / Hypothesis / Unknown			High/Medium/Low

A competitor tactic proves activity, not performance, profitability, incrementality, legal compliance or fit.

3. Develop across mechanisms

Generate candidates from product benefit/customer success, education/decision support, search with standalone value, expert/founder distribution, permitted customer referrals, partnerships, respectful community participation, events/demos, controlled referral programs, paid media, genuine PR, consent-based communication with existing customers, and localization only when operational readiness exists.

4. Screen out before evaluation

Decline or escalate in cases of unclear data use, manipulation, misleading user guidance, unsubstantiated claims, incorrect recommendations, unauthorized rights, accounts or customer data, mass outreach, non-existent features, hidden follow-up costs, discrimination, and unacceptable physical, legal, security, or reputational risk.

5. Evaluate transparently

Agree on criteria and weights situationally: target group evidence, goal/phase fit, offering maturity, distribution access, learning value, proof/creative maturity, time to signal, effort/maintenance, budget risk, data protection/legal/platform/security/accessibility/reputation, measurability, reversibility, and confidence. Show assumptions and sensitivity; missing data lowers confidence.

6. Recommend a portfolio

Three to five ideas only if this makes the decision easier: if applicable, a low-cost learning test, a reusable asset, a distribution/relationship test, and optionally a clearly marked higher stake. If a longlist is requested, group by mechanism and make prerequisites as well as low confidence visible.

Experiment brief

For each shortlisted idea:

Idea and mechanism: expected behavior and rationale.
Fit: target group, phase, goal, and evidence.
Hypothesis: a falsifiable statement instead of a promise.
Minimal test: the smallest ethical implementation that answers the question.
Resources: responsibility, competencies, time span, budget cap, dependencies, tools.
Required evidence: facts, rights, creative material, product maturity.
Risks/checkpoints: data protection, legal, platform, rights, security, accessibility, reputation.
Measurement: numerator, denominator, cohort, source, period, quality and harm thresholds.
Decision rule: continue, change, stop, or investigate.
Approval boundary: exact external action that is not yet authorized.

Use ranges only when there is a basis. Describe the expected outcome as an observable signal, never as guaranteed business success.

Channel-specific guardrails

Content/search/social/community: standalone value, sources, moderation standards, rights, accessibility, and privacy; do not scrape private groups or post automatically.
Paid advertising/retargeting: check current platform rules, audience restrictions, consent, claims, landing page, budget, billing, attribution, contact frequency, and stop criteria before launch.
Email/lifecycle: purpose and recipients appropriate; separate transactional and marketing; respect unsubscribe, blocking, and withdrawal of consent.
Referral/affiliate/promotion: plan benefit, incentive, disclosure, abuse protection, terms, taxes, eligibility, caps, and support.
Data/reports: authorized, minimized and sufficiently aggregated; document sample, method, uncertainty, purpose, retention and re-identification risk.
Product/Developer: Specification, permissions, test environment, rollback, monitoring, support and real QA.
Events/PR: Consent, recording/location rights, approval, accessibility, disclosure, moderation and crisis plan.

Measurement principles

Define metrics before objectives. Include quality, costs, complaints, cancellations, support load, accessibility errors, fraud and negative feedback as guardrail metrics. Distinguish correlation from incrementality and contribution from attribution. Impressions, clicks, followers, registrations, rankings or mentions alone are not success. If possible, use a comparison baseline, holdout, geo split or robust time-series design. Stop if data quality is poor, legal basis is missing, there is overload, or user harm occurs.

German/DACH context

For a German-language assignment, Germany is the marked default assumption. Assess Austria and Switzerland separately; do not transfer permissibility, channels, expectations, language, taxes and costs wholesale.

Define user/target language and Sie/Du consistently per audience and channel.
Budget and prices in the appropriate currency; label EUR and net/gross/VAT appropriately, treat CH separately.
Localize dates and deadlines unambiguously; retain technical ISO dates.
Mark UWG, GDPR and TDDDG, consent, direct marketing, sweepstakes, price information, platform rules, trademark law and accessibility as current review areas.
Do not provide legal advice or legal guarantees; arrange for review by the competent specialist before implementation.
DACH localization is a hypothesis as long as local target audience and channel evidence is lacking.

Output format

Selection list

Rank	Idea	Mechanism	Evidence	Minimal test	Time/cost framework	Main risk	Metric/guardrail	Confidence

Then provide the experiment brief and all necessary approvals for the top candidate. For a large collection of ideas, group by mechanism and show for each idea the concept, fit, prerequisite, evidence class, effort, budget risk, review and next validation step. No false precision.

Checklist

Objective, target group, baseline value, threshold and follow-up decision are clear.
Observation, estimate, hypothesis and unknown are separated.
Ideas were checked for safety before scoring.
No idea relies on tracking, spam, manipulation, false proof or unauthorized data/rights.
Costs include implementation, media budget, tools, work, support, maintenance and compliance.
Statements, rights, platform, data protection, security, accessibility and reputation are visible.
Measurement includes quality and defect limits.
Experiment has responsibility, budget cap, stop rule, and approval limit.
Germany/AT/CH, language, salutation, currency, tax, and date are clarified.
No legal guarantee and no external action without approval.

Scope

product-marketing provides positioning and evidence, customer research validates needs, content-strategy plans content portfolios and offering strategy structures services. Channel, legal, product, research, and operations workflows govern implementation. This skill develops hypotheses and does not implement them tacitly.

Origin and deviations

German/DACH adaptation of the MIT-licensed original skill marketing-ideas by Corey Haines based on the upstream commit 7868cb9251fad80a73d26e488a5ad5f6c4a9f335 (homepage in the frontmatter). Deviation categories: Translation, Restructuring, DACHlocalization and safety adaptation through evidence separation, risk review, and approval of external actions. Details are in the repository path docs/UPSTREAM-AENDERUNGEN.md.