Free Starter Pack

3 Production-Ready
AI Workflow Templates

Full prompt chains. Exact text for every step. Input/output specs.
Ready to run on Claude, GPT, or Gemini — today.

Content Pipeline — 6 steps
Data Analysis — 7 steps
Customer Support — 6 steps
🔓

Unlock All Steps — Free

Enter your email to unlock every prompt, every step, and every customization tip in all three workflows. No credit card. No catch.

By submitting you agree to receive occasional emails from PromptForge. Unsubscribe anytime.

✓  Access granted. All steps unlocked below.
3 Workflow Templates
✍
Content Pipeline Workflow
Takes a raw topic and moves it through research, outline, draft, edit, and publish-ready formatting — with each step's output feeding directly into the next.
▪ Content & Writing
01
Topic Research & Angle Discovery
"You are a research analyst. Given the topic [TOPIC], identify the 5 most compelling angles an audience of [TARGET AUDIENCE] would find valuable today. For each angle, note what makes it timely, who it helps, and what the key tension or insight is. Return as a ranked list with a one-sentence summary per angle."
InTopic keyword or phrase
Out5 ranked angles with summaries
Replace [TARGET AUDIENCE] with a specific descriptor — "early-stage founders," "B2B marketers," "HR directors at 50-500 person companies." Specificity here shapes every downstream step.
02
Outline Construction
"You are a senior editor. Using angle #[N] from the research above, create a tight content outline. Include: a working headline, 4-6 main sections with one-sentence descriptions, a logical flow that builds tension and resolves it, and a closing call-to-action. Do not write body copy yet — structure only."
InSelected angle from Step 1
OutStructured outline, headline, CTA
Run this step with your top 2 angles and pick the stronger outline before writing. Saves revision cycles on the full draft.
03
First Draft Generation
"You are a skilled content writer with expertise in [DOMAIN]. Using the outline above, write the full first draft. Aim for [WORD COUNT] words. Use [TONE: conversational / authoritative / technical]. Every section must deliver on its outline promise. End each section with a natural bridge to the next. Do not add filler — if a section runs short, it should run short."
InOutline from Step 2 + tone/length specs
OutFull first draft, structured by outline
Set [WORD COUNT] 20% higher than your target. Editing down is faster than padding up, and you'll want material to cut during the edit step.
Unlock to view
04
Structural & Copy Edit
"You are a copy editor. Review this draft critically. Flag: (1) sections that underdeliver on their promise, (2) sentences over 25 words that can be split, (3) passive constructions to rewrite, (4) any factual claims that need a source citation. Return the full revised draft with all edits applied inline — not as a list of suggestions."
InFirst draft from Step 3
OutEdited draft with inline revisions
Pass this step twice for high-stakes content. First pass for structure, second pass for language. The model improves significantly on the second pass when it can review its own prior edits.
Unlock to view
05
SEO & Metadata Layer
"You are an SEO specialist. Given this article, produce: (1) an optimized title tag under 60 characters, (2) a meta description under 155 characters, (3) 3 H2 alternatives for the weakest headline in the draft, (4) 5 semantic keyword phrases to work into the body naturally. Do not stuff — suggest only phrases that add meaning."
InEdited draft from Step 4
OutTitle, meta, H2 variants, keyword list
Run this step after editing, not before. SEO optimizing an unfinished draft wastes iterations — the structure changes too much during editing to preserve keyword placement.
Unlock to view
06
Publish-Ready Formatting
"You are a content publisher. Format this article for [PLATFORM: blog / LinkedIn / newsletter / Twitter thread]. Apply platform-appropriate structure: headers, bullet points, line breaks, and CTA placement. For LinkedIn: add white space after every 2-3 sentences. For newsletters: add a one-line hook at the very top before the headline. Output the final, publication-ready version."
InEdited draft + target platform
OutPlatform-formatted, publish-ready content
Save platform-specific formatting instructions as a reusable template. A "LinkedIn format" prompt block you paste once saves 20 minutes per post.
Unlock to view
Unlock Remaining Steps →
📊
Data Analysis Workflow
Takes raw data through cleaning, pattern identification, insight generation, visualization planning, and executive summary — producing analyst-grade output from a single data input.
▪ Data & Analytics
01
Data Intake & Validation
"You are a data analyst. I am providing you with a dataset in [FORMAT: CSV / JSON / table]. Before any analysis, audit the data: identify missing values, duplicates, obvious outliers, inconsistent formatting, and fields that need type conversion. Return a structured data quality report with counts for each issue type and a recommended cleaning plan."
InRaw dataset + format description
OutData quality report + cleaning plan
Always run validation first, even on "clean" data you received from a trusted source. Silent data issues — like dates formatted as strings — corrupt downstream analysis without throwing errors.
02
Cleaning & Normalization
"Using the cleaning plan from the validation step, clean the dataset. Apply these rules: (1) impute or flag missing values using [STRATEGY: median / mode / forward-fill / drop row], (2) remove exact duplicate rows, (3) standardize date fields to ISO 8601, (4) normalize text fields to lowercase with leading/trailing whitespace removed. Return the cleaned dataset and a change log of every modification made."
InRaw data + validation report from Step 1
OutCleaned dataset + change log
Keep the change log. When stakeholders question a number, you can show exactly what transformations were applied and in what order — this is the difference between "trust me" and "here's the audit trail."
03
Exploratory Pattern Detection
"You are a senior data analyst. Perform exploratory analysis on the cleaned dataset. Identify: (1) distributions and outliers in numeric columns, (2) correlations between key variables (Pearson if continuous, Cramer's V if categorical), (3) time-based trends if date fields exist, (4) any clusters or segments that emerge naturally from the data. Flag the top 5 most analytically interesting findings, ranked by potential business impact."
InCleaned dataset from Step 2
OutTop 5 findings ranked by impact
Ask the model to rank findings by business impact, not statistical significance. Significant p-values on trivially small effects waste stakeholder attention.
Unlock to view
04
Hypothesis Testing & Validation
"Given the patterns identified in Step 3, formulate 3 testable hypotheses. For each: state the hypothesis in plain English, identify the appropriate statistical test (t-test, chi-square, ANOVA, etc.), run the test on the data, report the result with p-value and effect size, and translate the finding into a one-sentence business implication. Flag which hypotheses are confirmed vs. rejected at p < 0.05."
InTop findings from Step 3
Out3 tested hypotheses with p-values + business translations
Always request effect size alongside p-value. A result can be statistically significant but practically meaningless. Effect size tells you whether the finding is worth acting on.
Unlock to view
05
Visualization Specification
"For each confirmed finding from Step 4, recommend the single best visualization type. Specify: chart type and justification, X/Y axis variables and labels, whether to include confidence intervals or error bars, color encoding for any categorical splits, and the key takeaway text to display as a chart annotation. Output as a structured spec a developer or BI tool can implement directly."
InConfirmed findings from Step 4
OutChart specs ready for Tableau / Python / Looker
Request chart specs in a structured format your BI tool accepts. For Python/matplotlib: ask for the exact plt.plot() call. For Tableau: ask for the field pill configuration. Specificity cuts implementation time by 80%.
Unlock to view
06
Executive Summary Generation
"You are a data storyteller presenting to a non-technical executive audience. Write a 300-word executive summary of this analysis. Structure: one-paragraph situation (what data we analyzed and why), one-paragraph top findings (3 bullets, each anchored to a specific number), one-paragraph recommendations (what to do, in what order, what to watch). No jargon. No caveats buried in footnotes — surface them inline."
InAll prior steps + audience context
Out300-word exec summary ready for board/stakeholder
Specify "no jargon" explicitly. Models default to technical language when summarizing technical work. Name the audience ("CFO with no data background") for better calibration.
Unlock to view
07
Next-Step Recommendations & Monitoring Plan
"Based on the complete analysis, identify: (1) the single highest-leverage action the business should take based on the data, (2) two hypotheses that require more data to test — specify exactly what data is needed and how to collect it, (3) a monitoring plan: which 3 metrics should be tracked going forward, at what cadence, and what threshold change would trigger a re-analysis."
InComplete analysis from all prior steps
OutAction item + data gaps + monitoring plan
The monitoring plan is the most overlooked step in data analysis. Most teams produce a one-time report and move on. Building the monitoring plan in the same session closes the loop and creates a recurring decision-support system.
Unlock to view
💬
Customer Support Workflow
Takes an incoming customer inquiry through intent classification, knowledge retrieval, response drafting, tone and accuracy review, and escalation decision — handling Tier 1 cases autonomously.
▪ Customer Support
01
Intent Classification & Priority Triage
"You are a customer support triage system. Classify the following inquiry: (1) primary intent — one of [billing / account access / product bug / feature request / refund / general question / complaint], (2) sentiment score 1-5 (1 = very negative, 5 = neutral/positive), (3) urgency level — [P1: service down / P2: blocking workflow / P3: non-urgent], (4) is this a Tier 1 resolvable issue or does it require escalation? Return as structured JSON."
InRaw customer message
OutJSON: intent, sentiment, priority, escalation flag
Customize the intent list to match your actual ticket categories. Generic classifications produce generic routing. A list tuned to your product's specific failure modes reduces misrouting by 40-60%.
02
Knowledge Base Retrieval & Gap Detection
"Given this classified intent, search the knowledge base for the most relevant documentation. Return: (1) the top 3 most relevant articles ranked by match quality, with a one-sentence summary of each, (2) the specific passage from each article that most directly addresses the inquiry, (3) a confidence score 0-100 for whether the knowledge base fully covers this issue. If confidence is below 70, flag it as a knowledge gap and describe what information is missing."
InClassification from Step 1 + KB documents
OutTop 3 articles + relevant passages + confidence score
The knowledge gap detection is the most valuable part of this step over time. Log every low-confidence result. After 30 days you'll have a prioritized list of exactly which docs need to be written to reduce escalation rates.
03
Response Drafting
"You are a senior customer support specialist at [COMPANY NAME]. Draft a response to the customer inquiry using the retrieved knowledge above. Requirements: (1) open by acknowledging the specific issue — do not use generic openers like 'Thank you for reaching out,' (2) address the root cause, not just the symptom, (3) provide step-by-step resolution instructions where applicable, (4) close with a clear next step for the customer. Tone: [TONE: professional / friendly-professional / empathetic]. Max 200 words."
InKB passages from Step 2 + company/tone context
OutDraft response under 200 words
The "acknowledge the specific issue" instruction is critical. Generic openers are the number-one customer satisfaction killer in support emails. Naming the exact problem in sentence one raises CSAT scores measurably.
Unlock to view
04
Tone & Accuracy Review
"Review the draft response above against these criteria: (1) Does the tone match the customer's sentiment score [SCORE]? A score of 1-2 requires a more empathetic opening, (2) Are all factual claims in the response supported by the retrieved knowledge base passages? Flag any unsupported claims as UNVERIFIED, (3) Does the response avoid these prohibited phrases: [LIST YOUR BANNED PHRASES], (4) Is the response under the 200-word limit? Return the approved response or a revised version with all issues corrected."
InDraft from Step 3 + sentiment score + KB sources
OutApproved or revised response, verified against KB
Maintain a banned phrases list. "We apologize for any inconvenience," "as per my previous email," "I understand your frustration" (without context) — these phrases consistently score lowest in CSAT. Build your list from your worst-rated tickets.
Unlock to view
05
Escalation Decision & Handoff Brief
"Based on the complete context — original inquiry, classification, KB confidence score, and reviewed response — make a final escalation decision. If escalating: write a 3-sentence handoff brief for the Tier 2 agent covering (1) what the customer wants, (2) what you've already tried or confirmed, (3) why Tier 1 cannot resolve this. Include the customer's sentiment score and priority level. If not escalating: confirm the response is ready to send."
InAll prior context + reviewed response
OutSend confirmation OR Tier 2 handoff brief
The 3-sentence handoff format is non-negotiable. Tier 2 agents who receive full context resolve tickets 35% faster on average. The brief format forces Tier 1 to synthesize — which also catches cases where escalation was actually unnecessary.
Unlock to view
06
Quality Log & Pattern Capture
"Log this resolved ticket to the quality database. Extract: (1) intent category and sub-classification, (2) KB articles used — which ones were most and least helpful, (3) whether a knowledge gap was detected and what information was missing, (4) final resolution: resolved by AI / escalated to Tier 2 / required human rewrite, (5) one suggested improvement to either the KB or the response template based on this case. Return as structured JSON for ingestion."
InComplete case context from all prior steps
OutStructured JSON log entry for quality tracking
This step makes the entire workflow self-improving. Run a monthly analysis on the quality log to find the top 10 most common KB gaps and the top 5 prompt improvements. After 90 days, you'll have a workflow that outperforms what you started with by a significant margin.
Unlock to view
19
workflow steps
in this pack
3
production-ready
templates
$0
cost to unlock
right now

Ready for more?

Upgrade to PromptSharp Pro

Unlock the full library — advanced multi-step chains, multi-model strategies, and new workflow packs added every week.

Monthly — $19/mo Annual — $97/yr save 57%