GUIDE 1.2 • FOUNDATIONS
Prompting 101
Most AI output is generic because the input is generic. This guide covers the structure of a prompt that works: how to set context, assign a persona, specify format, and iterate without starting over. Includes 15 copy-paste templates for common B2B marketing tasks.
10 min read|Guide 1.2 of 4|15 copy-paste templates
The anatomy of a prompt that works
AI prompting for B2B marketing is a learnable, repeatable skill — and the teams that build it produce better briefs, faster research, and stronger copy than those using default inputs. This guide covers structure, persona framing, format instructions, and iteration, with 15 copy-paste templates built for common B2B tasks. According to HubSpot’s State of Marketing report, marketers who use AI with structured prompting save an average of 2.5 hours per week. Adjust these prompts for your target market using the buyer context by market guide.
A prompt is a brief. The model will follow your instructions as precisely as you write them. If you give it a vague brief, you will get a vague output. If you give it a specific brief with clear constraints, you will get something usable. The difference between a good prompt and a bad one is not technical skill. It is the same clarity you would apply to briefing a junior copywriter.
Every effective prompt has four components. You do not always need all four. But when output is not landing, one of these is usually missing.
1
Context
Tell the model who you are, what the product is, and who you are writing for. Without this it defaults to generic assumptions.
I am a B2B marketer at a Bangalore-based SaaS company that sells marketing automation software to mid-market US tech companies with 200-500 employees.
2
Persona
Tell it who to write as or write for. A persona shapes tone, vocabulary, and assumed knowledge level significantly.
Write as a senior B2B copywriter who specialises in SaaS demand generation. Write for a VP of Marketing at a US tech company who is time-pressed and data-driven.
3
Task and constraints
State exactly what you want and what the limits are. Length, format, tone, what to include, what to avoid.
Write a cold email subject line. Maximum 8 words. No questions. No clickbait. Lead with a specific outcome, not a feature. Do not use the word "revolutionary".
4
Format
Tell it how to structure the output. Bullet list, numbered list, paragraph, table, JSON. Unspecified format means whatever the model defaults to.
Output as three options, each on a new line. No explanations. No numbering. Just the subject lines.
Weak prompt vs strong prompt
The same task. Two different prompts. The difference in output quality is significant.
WEAK PROMPT
Write a cold email for our product.
TYPICAL OUTPUT
Generic. Probably starts with "I hope this email finds you well." No specific audience. No specific outcome. Could be for any product in any industry. Requires significant rewriting before it is usable.
STRONG PROMPT
I am a B2B marketer at a Bangalore-based SaaS company. Our product helps marketing ops teams reduce campaign setup time by 60%. Write a cold email for a Head of Marketing Operations at a 300-person US SaaS company. The reader is busy and data-driven. Get to the outcome in the first sentence. No opener like "I hope this email finds you well." Maximum 100 words. End with one low-commitment CTA: a 20-minute call.
TYPICAL OUTPUT
Specific, on-tone, right length. Leads with a metric. Addresses the correct buyer. Usable with minor edits or no edits at all.
Iteration: the step most people skip
The first output is a draft. Almost no experienced AI user accepts the first response unchanged. The model has given you a starting point. Your job is to tell it what to adjust.
Iteration is faster than rewriting from scratch. Instead of rejecting the output and starting again, tell the model specifically what is wrong and what to change. Three or four exchanges will almost always produce something significantly better than the first attempt.
1
Send the prompt
Use your four-component prompt. Do not start with a vague question.
2
Read critically
What is wrong? Too long? Wrong tone? Missing a point? Too general?
3
Give specific feedback
Tell it exactly what to fix. Not "make it better." Say "shorten to 80 words and remove the second paragraph."
4
Repeat
Two or three rounds is usually enough to get from draft to usable.
USEFUL ITERATION INSTRUCTIONS
Shorten this to 80 words. Keep the first sentence exactly as is.
The tone is too casual. Rewrite in a more formal register without being stiff.
Remove everything after the second paragraph. End with just the CTA.
The opening is too generic. Start with the specific metric instead.
Give me three alternative subject lines. No questions. All under 8 words.
This reads like marketing copy. Rewrite it to sound like a practitioner wrote it.
The CTA is too high-commitment. Change it to asking for a 15-minute call.
Add a specific example from the SaaS industry to the second paragraph.
15 copy-paste prompt templates
Replace the bracketed fields with your specifics. Use as starting points, not finished briefs. Each template is built on the four-component structure from Section 1.
Content and copy
CONTENTLinkedIn thought leadership post
I am a [job title] at [company], a [description] SaaS company based in India selling to [target market]. Write a LinkedIn post sharing a specific insight about [topic] that would be useful to [target audience]. Write in first person. Avoid generic advice. Lead with a specific observation, not a question. Maximum 200 words. No hashtags in the body. Three line breaks maximum.
Why this works: First-person framing plus the instruction to lead with an observation prevents the generic "Have you ever wondered..." opener that most AI defaults to.
CONTENTBlog post introduction
I am writing a blog post titled "[title]" for [company], targeting [ICP]. Write the introduction: three paragraphs, maximum 150 words total. The first paragraph must state the problem clearly without hedging. The second paragraph must establish why this matters now. The third must preview what the reader will learn. Tone: direct, practitioner-level. No "In today's rapidly changing world" openers.
Why this works: Specifying the three-paragraph structure and banning the cliche opener forces the model away from its most common defaults.
CONTENTEmail newsletter section
Write one section for a B2B marketing newsletter. Topic: [topic]. Reader: [audience description]. Structure: one-sentence hook, three to four sentences of substance, one concrete takeaway or action. Total length: 100-120 words. Tone: peer-to-peer, not instructor-to-student. Do not start with "In this section" or "Today we will explore".
Why this works: The peer-to-peer tone instruction is the most important constraint here. It prevents the didactic tone that AI defaults to for educational content.
Demand generation
DEMAND GENCold outreach email
I am a B2B marketer at [company], which helps [ICP] achieve [outcome] by [mechanism]. Write a cold email for a [job title] at a [company size] [industry] company in [market]. The reader is [personality/behavior description]. Get to the business outcome in the first sentence. No opener like "I hope this finds you well." Maximum [X] words. End with one CTA: [specific CTA]. Do not mention features.
Why this works: Explicitly banning the weak opener and limiting to one CTA forces a cleaner structure. "Do not mention features" keeps the focus on buyer outcomes.
DEMAND GENLinkedIn ad copy
Write LinkedIn ad copy for [company] targeting [job title] at [company size] [industry] companies. The ad promotes [offer: webinar / guide / demo / trial]. Introductory text: maximum 150 characters. Headline: maximum 70 characters. Lead with the benefit to the reader, not what the company does. No exclamation marks. No "Free" unless the offer is genuinely free.
Why this works: Character limits force concision. The benefit-first instruction and ban on exclamation marks prevent the default ad-copy register.
DEMAND GENWebinar invitation email
Write an invitation email for a webinar titled "[title]" on [date]. Target audience: [description]. Three key things the attendee will learn: [1], [2], [3]. Speaker: [name and one-line credibility statement]. Structure: hook, three learning outcomes as bullet points, speaker line, CTA. Maximum 150 words. Subject line: maximum 8 words, no questions.
Why this works: Providing the three learning outcomes means the model does not have to invent them, which eliminates hallucination risk on the most important content in the email.
Product marketing
PRODUCT MKTPositioning statement
Help me write a positioning statement for [product]. Target customer: [ICP description]. Category: [what category does it compete in]. Key differentiation: [what makes it different from alternatives]. Proof point: [one specific and verifiable claim]. Format: For [target customer] who [need or problem], [product] is the [category] that [key benefit] unlike [main alternative] which [limitation]. Write three variations. Do not add marketing language. Keep it factual.
Why this works: Providing the positioning statement template as the output format forces the model to work within a proven structure rather than inventing its own.
PRODUCT MKTFeature to benefit translation
I will give you a list of product features. For each feature, write the buyer benefit in plain language. Buyer: [ICP description]. Write the benefit from the buyer's perspective, not the vendor's. Format: Feature | Benefit. One line per feature. No marketing language. Features: [paste your feature list]
Why this works: The "from the buyer's perspective" instruction and the table format prevent verbose explanations and keep each benefit to one usable line.
PRODUCT MKTCompetitive comparison section
Write a comparison section for our website comparing [product] to [competitor]. Our strengths: [list]. Their known limitations: [list]. Do not make claims we cannot verify. Do not be disparaging. Format: a table with rows for [criteria 1], [criteria 2], [criteria 3]. Columns: Us / [Competitor]. One-sentence entry per cell. Tone: factual, confident, not aggressive.
Why this works: Providing both strengths and limitations means you control the facts. The tone instruction prevents the combative register that competitive pages often slip into.
Research and analysis
RESEARCHICP interview synthesis
I am going to paste notes from [number] customer interviews. Extract: the three most common pain points (with direct quotes if available), the words and phrases customers use to describe the problem, the alternatives they considered before choosing us, and the outcome they most wanted. Format each as a bullet list. Do not paraphrase quotes. If something appears in only one interview, flag it as a single data point. [Paste interview notes]
Why this works: Asking it to flag single data points prevents the model from presenting one person's view as a pattern. The quote preservation instruction catches the model before it smooths out the most useful raw material.
RESEARCHCompetitor content audit
I will paste the homepage copy and one blog post from [competitor]. Analyse: their primary ICP based on the language used, the positioning claim they are making, the content topics they prioritise, three gaps or angles they are not covering. Output as four labelled sections. Be specific. Do not describe the content, analyse it. [Paste competitor content]
Why this works: The instruction to analyse rather than describe prevents summary output. Asking for gaps specifically produces something you can act on.
RESEARCHMarket trend summary for leadership
I will paste [number] articles about [topic or trend]. Write a 200-word summary for a senior marketing leader. Include: what is changing, what the implication is for B2B marketing teams, and one specific action worth considering. Cite the source for any specific claim using [Source Name] in brackets. Do not include statistics you cannot attribute to a specific source in the pasted text. [Paste articles]
Why this works: The citation instruction and the ban on unsourced statistics prevent hallucination on the highest-risk content type: factual claims about market trends.
Revenue ops and reporting
REVENUE OPSPipeline commentary for leadership
Write a pipeline commentary for a weekly leadership update. Data: MQLs this week: [number], SQLs: [number], pipeline added: [amount], pipeline lost: [amount], reasons for loss: [list]. Write three paragraphs: what the numbers show, what changed versus last week, and one recommendation. Maximum 150 words. Tone: direct and analytical. No positive spin on negative numbers.
Why this works: Providing all the numbers directly prevents invention. The instruction to avoid positive spin on negative numbers is the most important constraint: AI defaults to framing everything diplomatically.
REVENUE OPSCampaign performance summary
Write a campaign performance summary. Campaign: [name]. Goal: [goal]. Period: [dates]. Results: [paste metrics]. Benchmark or target: [target]. Write: one sentence on whether the campaign hit its goal, two sentences on what drove the result, one sentence on what to do differently next time. Do not describe what the campaign was. Jump straight to performance.
Why this works: The instruction to skip campaign description forces the model into analysis rather than recap. The four-sentence structure keeps it concise and actionable.
REVENUE OPSAttribution explanation for sales
Write a plain-English explanation of [attribution model] for a sales team that has no marketing background. Explain: what it measures, why marketing uses it, and one limitation they should be aware of when reading pipeline reports. Maximum 120 words. No jargon. If you need to use a technical term, define it immediately after.
Why this works: The "define it immediately after" instruction catches the most common failure in cross-functional communication: assuming shared vocabulary.
Six prompting mistakes to stop making
✕
Asking without context
"Write a blog post about AI in marketing." The model has no idea who you are, who you are writing for, or what angle you want. Always give context first.
✕
Accepting the first output
The first response is a draft. Treating it as a finished product leads to generic content. Expect to iterate at least twice.
✕
Vague feedback in iteration
"Make it better" tells the model nothing. "Shorten the second paragraph to two sentences and remove the industry jargon" gives it something to work with.
✕
Asking it to invent facts
"Add some statistics about email open rates." It will. They may not be real. Always supply the facts and ask it to write around them.
✕
One prompt for everything
Do not ask it to research, outline, draft, and format in a single prompt. Break complex tasks into steps. Each step produces better output.
✕
Not saving prompts that work
When a prompt produces good output, save it. A reusable prompt library built over weeks is one of the highest-value assets a marketing team can have.
What to read next
NEXT IN FOUNDATIONS
AI literacy quiz
Five questions that tell you where you actually are with AI in marketing and give you a personalised reading path.
Take the quiz →
ALSO IN FOUNDATIONS
Buyer context by market
How to adjust your prompts for NA, European, and APAC buyers. Applies directly to the demand gen and content templates in this guide.
Read guide 1.4 →
TOOLS PILLAR
The full prompt library
40+ prompts across all marketing functions, filterable by use case. The expanded version of the 15 templates in this guide.
Browse the library →
Get the Digital Uncovered newsletter
B2B marketing intelligence for India. No padding. No filler. Straight to the point, twice a week.