AI prompts for marketers

Build a topic cluster.

CORE TOPIC: [THE SUBJECT]
MY SITE: [WHAT IT DOES, WHO IT SERVES]
MY AUTHORITY LEVEL: [new site / some rankings / established in this niche]
KEYWORDS I HAVE: [PASTE YOUR LIST, or 'none - suggest them']
WHAT I SELL OR WANT PEOPLE TO DO: [CONVERSION GOAL]

Produce:

1. THE CLUSTER MAP - group keywords by SEARCH INTENT, not by word similarity. Keywords that want the same answer belong on one page, even if they look different. Keywords that look similar but want different answers need separate pages.
   For each group: the page it becomes, the primary keyword, the secondary keywords it can rank for, and the intent (informational / commercial / transactional / navigational).

2. THE PILLAR PAGE - the one broad page everything links to, and what it must cover.

3. THE SUPPORTING PAGES - each with: working title, URL slug, primary keyword, intent, and the specific question it answers.

4. THE CANNIBALISATION CHECK - any two pages in my plan that would compete for the same query. Merge them and say so. This is the most common self-inflicted SEO problem.

5. PRIORITY ORDER - given my stated authority level, which pages to write first. For a new site, start with the lowest-competition, most specific pages, not the head term. Explain the reasoning.

6. THE COMMERCIAL PATH - which pages sit closest to my conversion goal, and how the informational pages should link toward them.

7. WHAT NOT TO WRITE - keywords in my list that are not worth a page: too competitive for my authority, wrong intent for my business, or better served as a section of an existing page.

Rules:
- Do not recommend one page per keyword variation. That is the scaled-content pattern Google penalises.
- Be honest about competitiveness. A new site will not rank for a head term, and saying otherwise wastes months.

Plan a page to rank for this keyword.

TARGET KEYWORD: [KEYWORD]
WHAT CURRENTLY RANKS: [PASTE THE TOP 5 TITLES AND WHAT EACH IS, or describe them]
MY SITE: [WHAT IT IS, AUTHORITY LEVEL]
WHAT I CAN OFFER THAT THEY CANNOT: [YOUR ADVANTAGE - data, experience, a tool, access]
MY CONVERSION GOAL: [IF ANY]

Produce:

1. THE INTENT READING - from what currently ranks, what does the searcher actually want? The SERP is Google telling you what it thinks the intent is; match it or do not rank. Classify: are the results guides, listicles, tools, product pages, comparisons, forum threads?

2. THE FORMAT REQUIRED - if the top results are all tools and I write an article, I will not rank regardless of quality. State plainly what format this keyword demands.

3. WHAT ALL THE TOP RESULTS COVER - the table stakes. Omit any of these and the page looks incomplete.

4. WHAT NONE OF THEM COVER - the gap. This is where the page can win.

5. THE OUTLINE - sections in the order the searcher needs them. Lead with the answer; searchers arriving from Google want it immediately, not after 600 words of context.

6. THE ANSWER BLOCK - the 40-60 word direct answer to the query, to sit immediately under the H1. Written to stand alone if quoted.

7. TITLE AND META - three title options under 60 characters, and a meta description under 155 that describes the page honestly rather than teasing.

8. WHAT WOULD MAKE THIS BETTER THAN THE TOP RESULT - specifically, using my stated advantage. If I have no advantage over the current results, say so plainly - that is a reason to pick a different keyword rather than to write a page that will not rank.

9. INTERNAL LINKS - what should link to this page, and where it should link.

Do not recommend writing a longer version of what already ranks. Length is not a differentiator.

Optimise this content to be cited by AI answer engines.

CONTENT:
"""
[PASTE]
"""

THE QUESTIONS IT SHOULD ANSWER: [LIST]
MY EXPERTISE ON THIS: [WHY YOU CAN SPEAK TO IT]
ANY ORIGINAL DATA OR EXPERIENCE: [WHAT YOU HAVE THAT OTHERS DO NOT]

First, the honest framing: there is no special markup or file that gets you into AI answers. Google has stated there is no required schema and that files like llms.txt neither help nor harm. What gets cited is content that is indexed, crawlable, specific, and says something quotable that is not available everywhere else. Work from that.

Produce:

1. THE ANSWER BLOCKS - for each question I listed, a 40-70 word standalone answer to sit directly under a heading phrased as that question. Each must make complete sense quoted with no surrounding context. This is the single highest-impact change.

2. THE QUOTABILITY AUDIT - quote every sentence in my content that could be lifted as a standalone claim. If there are few, the content is too hedged or too vague to cite. Say so.

3. SPECIFICITY UPGRADES - vague claims that should be specific. 'Many businesses find this helpful' cannot be cited; 'in our test of 40 accounts, 31 saw the issue within two weeks' can. Where I have the specifics, write the replacement; where I do not, mark [NEED SPECIFIC].

4. THE ORIGINAL CONTRIBUTION - from what I told you, what does this content say that is not on twenty other pages? Answer engines cite sources that add something. If the answer is nothing, no optimisation fixes that; say so plainly.

5. STRUCTURE - descriptive question-phrased headings, one idea per section, definitions stated plainly, and the conclusion near the top rather than the bottom.

6. EVIDENCE AND DATING - claims with a date, a method, a sample size, or a named source are far more citable. Add where possible; flag where missing.

7. AUTHOR AND CREDIBILITY SIGNALS - visible author, credentials relevant to this topic, last-updated date. These are checked by both quality raters and retrieval systems.

8. WHAT NOT TO DO - do not create an llms.txt, do not stuff schema that does not describe visible content, do not create a page per question variant. Those are wasted effort or actively harmful.

Write social ads.

PRODUCT: [WHAT IT IS]
PLATFORM: [Meta / Instagram / LinkedIn / TikTok / X]
AUDIENCE: [WHO YOU ARE TARGETING AND WHAT THEY ARE DOING WHEN THEY SEE THIS]
THE PROBLEM IT SOLVES: [PAIN POINT]
AWARENESS LEVEL: [they don't know they have the problem / know the problem, not the solution / comparing solutions / know you]
CREATIVE FORMAT: [image / video / carousel]
OFFER: [WHAT THEY GET]
LANDING PAGE: [WHERE THEY GO]

Produce:

1. THE AWARENESS-MATCHED APPROACH - this is the most important decision. Someone who does not know they have the problem needs a completely different ad from someone comparing vendors. State what this awareness level requires and what it rules out.

2. FIVE ADS, each with: the hook (first line), the body, the call to action, and a creative direction note.
   Vary the angle: the problem stated in their words, the specific outcome, the objection handled upfront, the contrast between before and after, and the unexpected observation.

3. THE FIRST LINE ANALYSIS - for each, why it would stop a scroll. The first line is 90% of the job; everything after it is read by people who already decided to stop.

4. PLATFORM FIT - what this platform's audience and format require. LinkedIn tolerates length and professional framing; TikTok needs the hook in the visual; Meta truncates after a couple of lines.

5. THE TRUTH CHECK - anything in these ads that overstates. Social ads that overpromise convert badly and generate refunds and complaints.

6. WHAT NOT TO USE - the tired patterns in this category that audiences have learned to ignore.

Rules:
- No engagement bait, no 'comment YES below', no fake questions
- No manufactured urgency or scarcity that is not real
- No before/after claims you cannot substantiate
- Write like a person, not a brand
- The hook must be specific. 'Struggling with X?' is the most ignored opening in advertising.

Help me design an A/B test.

WHAT I WANT TO TEST: [YOUR IDEA]
THE PAGE OR AD: [DESCRIBE OR PASTE]
CURRENT PERFORMANCE: [CONVERSION RATE AND VOLUME]
TRAFFIC: [VISITORS OR IMPRESSIONS PER WEEK]
WHAT I THINK WILL HAPPEN: [YOUR HYPOTHESIS]

Produce:

1. THE SAMPLE SIZE REALITY CHECK - first, before anything else. Given my conversion rate and traffic, how long would it take to detect a meaningful difference? Do the arithmetic and show it. If the answer is months, say so plainly - testing a button colour on 200 visitors a week is not testing anything, and most small-site A/B tests are statistically meaningless. This may be the whole answer.

2. IS THIS WORTH TESTING - small changes need enormous traffic to detect. Big changes can be detected with modest traffic. Assess my idea: is the expected effect large enough to measure at my volume? If not, recommend testing something bigger instead.

3. THE HYPOTHESIS - restated properly: 'Changing X to Y will increase Z because [reason]'. A test without a reason teaches you nothing even when it wins.

4. WHAT TO TEST INSTEAD, if my idea is too small: the headline, the offer, the price presentation, the form length, or the page structure. These move numbers; button colours do not.

5. THE TEST DESIGN - variant A, variant B, what changes (exactly one thing, or a coherent set if testing a whole approach), the primary metric, and the guardrail metrics that must not get worse.

6. THE RULES - decided in advance: minimum runtime (at least one full business cycle, usually two weeks, to avoid day-of-week effects), the sample size to reach, and the commitment not to stop early when it looks good. Stopping early on a favourable result is the most common way A/B tests produce false conclusions.

7. WHAT COULD INVALIDATE IT - traffic source changes, seasonality, a promotion running concurrently, or another test overlapping.

8. WHAT TO DO WITH THE RESULT - including the likely outcome of no significant difference, which is the most common result and is still information.

Be honest about statistics. Most A/B tests run by small sites are underpowered and their results are noise.

Write a welcome email sequence.

WHAT PEOPLE SUBSCRIBED TO: [WHAT YOU PROMISED AT SIGN-UP]
WHERE THEY SIGNED UP: [THE PAGE OR OFFER]
WHAT I ACTUALLY SEND: [YOUR REGULAR CONTENT AND FREQUENCY]
WHAT I SELL: [IF ANYTHING]
NUMBER OF EMAILS: [COUNT]
OVER WHAT PERIOD: [TIMEFRAME]

The welcome sequence is the most-opened email you will ever send. Design it for engagement, not for selling.

For each email produce: timing, subject line, preview text, body, and the one action.

Email 1 (immediately) must:
- Deliver what they signed up for, first thing. If there is a lead magnet, it is at the top, not after three paragraphs.
- Set expectations: what you send, how often, and what it is about. This single paragraph prevents a large share of later unsubscribes.
- Give them something useful immediately
- Ask for one small thing that improves deliverability: reply, or move to the primary inbox. Framed as helping them, because it does.
- Not sell

The rest of the sequence should:
- Establish what you are for, with your best existing content
- Show what you do, through a specific example or result rather than a description
- Introduce who you are, briefly, and only where it builds trust
- Only then, if at all, make an offer — and only if it fits what they signed up for

Then produce:
- THE EXPECTATION-SETTING PARAGRAPH, isolated. It is the highest-value 40 words in the whole sequence.
- THE SEGMENTATION OPPORTUNITY - a question or click in the sequence that would tell me what this subscriber wants, with minimal friction
- THE EXIT - what happens if they take the action early
- THE HONEST CHECK - if my sequence would sell before it delivers value, say so

Rules:
- No 'Welcome to the family'
- No asking them to follow you on six platforms
- Do not sell in email 1

Write a value proposition.

WHAT WE DO: [DESCRIPTION]
WHO IT IS FOR: [SPECIFIC CUSTOMER]
THE PROBLEM: [WHAT THEY STRUGGLE WITH]
WHAT THEY DO TODAY INSTEAD: [THE ALTERNATIVE - a competitor, a manual process, or nothing]
WHAT WE DO DIFFERENTLY: [YOUR CLAIM]
WHY THAT IS TRUE: [THE STRUCTURAL REASON - what makes it hard to copy]
PROOF: [EVIDENCE]

Produce:

1. FIVE VALUE PROPOSITIONS, each a different angle:
   - The outcome achieved
   - The problem removed
   - The comparison to the current alternative
   - The specific mechanism that makes it work
   - The customer's own words

   Each under 20 words.

2. THE SUBSTITUTION TEST - for each, substitute your three closest competitors' names. Does it still read as true? If yes, it is a category description, not a value proposition. Mark each PASS or FAIL.

3. THE OPPOSITE TEST - would any competitor claim the opposite? 'Easy to use' fails; 'takes a week to learn and then you are faster than anyone' passes.

4. THE SPECIFICITY UPGRADE - for each surviving option, make it more specific. Numbers, timeframes, named situations. 'Save time' becomes 'closes the month in two days instead of nine'.

5. THE PROOF CHECK - can you substantiate the differentiated claim? An unsupported value proposition is a liability; the first customer who tests it and finds it false becomes a detractor.

6. THE SUPPORTING STRUCTURE - the headline value proposition, then three supporting points that make it credible, then the proof for each.

7. THE RECOMMENDATION - one, with reasoning, and where to use it: homepage, ads, sales calls, the elevator answer.

8. WHAT THIS EXCLUDES - a value proposition that appeals to everyone appeals to no one. Name who this turns away, and confirm that is acceptable.

Define a brand voice.

WHAT THE BUSINESS DOES: [CONTEXT]
WHO IT SERVES: [AUDIENCE]
HOW YOU WANT TO BE PERCEIVED: [ADJECTIVES OR DESCRIPTION]
EXISTING WRITING: [PASTE SAMPLES, or 'none']
COMPETITORS' TONE: [HOW THEY SOUND]
WHO WILL WRITE USING THIS: [TEAM SIZE AND EXPERIENCE]

Produce:

1. THE VOICE TRAITS - three or four. For each:
   - The trait
   - What it means concretely in writing
   - A DO example and a DON'T example, both real sentences
   - THE DIAL: how far to go, because every trait has a failure mode. Direct becomes blunt; warm becomes saccharine; playful becomes unprofessional; expert becomes condescending. Name the failure mode.

2. THE DISTINCTIVENESS TEST - for each trait, could a competitor claim the opposite? 'Trustworthy' and 'helpful' fail; nobody claims to be untrustworthy. A voice built from traits everyone claims is not a voice. Mark each PASS or FAIL and replace the failures.

3. AGAINST THE COMPETITION - given how competitors sound, where does this voice actually differentiate? If everyone in the category is friendly and approachable, being friendly and approachable is invisible.

4. THE MECHANICS - the decisions that make voice concrete and consistent: person, contractions, sentence length, formality, humour, jargon, how you address the reader, how you deliver bad news, and how you apologise.

5. VOICE ACROSS CONTEXTS - a table: Context | How the voice shifts. Cover: marketing, support, error messages, bad news, legal, and social. A single voice applied identically everywhere fails; an error message should not be playful.

6. THE WORDS - terms you use, terms you avoid, and how you refer to your own things.

7. BEFORE AND AFTER - three real sentences rewritten from off-voice to on-voice.

8. THE ONE-PAGE VERSION - what a new writer needs. If the guide is longer than a page, nobody will use it.

Derive from my samples where I gave them. Do not produce generic brand-voice adjectives.

Analyse this customer research.

INTERVIEW NOTES OR TRANSCRIPTS:
"""
[PASTE ALL, LABELLED BY PARTICIPANT]
"""

NUMBER OF PARTICIPANTS: [COUNT]
HOW THEY WERE RECRUITED: [SOURCE]
WHAT I WAS TRYING TO LEARN: [RESEARCH QUESTION]
MY PRIOR ASSUMPTIONS: [WHAT YOU EXPECTED]

Produce:

1. THE SAMPLE CAVEAT FIRST - who these participants are and who they are not. Recruitment source determines what you can conclude; people recruited from your own customer list cannot tell you why people do not buy. State plainly what this sample can and cannot support.

2. THE PATTERNS - themes appearing across participants, with the count. Distinguish rigorously:
   - Said by most participants
   - Said by several
   - Said by one, vividly
   The third category is where research goes wrong; a compelling quote from one person is not a finding.

3. WHAT SURPRISED YOU - things that contradict my stated assumptions. Lead with these; confirmation is less valuable than correction.

4. BEHAVIOUR VS OPINION - separate what people said they did from what they said they think or would do. Weight the first far more heavily.

5. THE LANGUAGE - the exact words participants used for the problem, the solution and the outcome. This is your marketing copy and your keyword research, free.

6. THE CONTRADICTIONS - where participants disagreed, or where one participant contradicted themselves. Do not resolve these into a consensus. Disagreement usually means there are two segments.

7. THE UNPROMPTED MENTIONS - things raised without being asked. These carry more weight than answers to your questions.

8. WHAT NOBODY MENTIONED that you expected them to. Silence is data.

9. THE FINDINGS - ranked by confidence. For each: what it is, the evidence, how many participants, and how confident you are.

10. WHAT TO DO NEXT - and specifically, what would need more research before acting.

Do not over-conclude. With a small sample, most of this is directional. Say so.

Build a social media content plan.

PLATFORMS: [WHERE YOU POST]
WHAT MY BUSINESS DOES: [CONTEXT]
AUDIENCE: [WHO AND WHAT THEY WANT]
TIME AVAILABLE PER WEEK: [HOURS]
WHAT I ALREADY PRODUCE: [BLOG POSTS, PROJECTS, CUSTOMER WORK, ANYTHING]
GOAL: [awareness / leads / community / recruitment]
WHY PREVIOUS ATTEMPTS FAILED: [IF APPLICABLE]

Produce:

1. FREQUENCY REALITY CHECK - given my hours, what cadence can I sustain for a year? Be conservative. Most social plans fail in week five from over-commitment, not from bad content. Recommend fewer posts than I want.

2. PLATFORM CHOICE - should I be on all the platforms I listed? Concentrating on one platform properly beats being mediocre on four. Say which to drop.

3. CONTENT PILLARS - 3-4 recurring themes, derived from my business and audience. Each with: what it covers, why the audience cares, and roughly what proportion of posts.

4. REPEATABLE FORMATS - the actual point of a content plan. 5-6 post formats I can refill indefinitely without inventing a new idea each time. For example: a question customers ask, something that went wrong and what changed, a short breakdown of a piece of work, a contrarian take on a common practice, a useful thing found elsewhere with commentary.
   For each: the structure, roughly how long it takes to make, and where the raw material comes from.

5. THE SOURCE PIPELINE - where content comes from in my existing work. A plan requiring separate content creation is a plan that stops. Describe the capture habit.

6. THE CALENDAR - four weeks, mapped to formats and pillars, with specific topics for week one.

7. THE BATCHING PLAN - how to produce a week's posts in one session.

8. WHAT TO CUT from my ambitions - anything that costs more than it returns: daily posting, video if I have no capacity, being on every platform, or designed graphics.

9. THE MEASUREMENT - what actually indicates this is working for my stated goal. Followers usually is not.

Turn this into multiple social posts.

THE SOURCE:
"""
[PASTE ARTICLE, TALK, REPORT, PROJECT DESCRIPTION, OR CASE STUDY]
"""

PLATFORMS: [WHERE YOU POST]
MY AUDIENCE: [WHO]
MY TONE: [PASTE A SAMPLE POST, or describe]
HOW MANY POSTS I WANT: [NUMBER]
OVER WHAT PERIOD: [TIMEFRAME]

Produce:

1. THE IDEA EXTRACTION - every distinct idea in the source that could stand alone as a post. There are usually more than expected: a claim, a counter-claim, a method, a mistake, a number, a story, an objection handled, a definition, something you changed your mind about.
   List them all before writing any.

2. RANK THEM - by how interesting each is on its own, to my audience. The most interesting idea is often buried in the middle of the source, not the headline.

3. THE POSTS - for the top [NUMBER], each:
   - Written natively for its platform
   - Delivering value without requiring the reader to click through
   - With the specific detail that makes it worth reading
   - Never opening with 'I wrote about this recently'

4. THE SCHEDULE - spread over my timeframe, ordered so that the strongest goes first and related ideas are not adjacent.

5. THE VARIETY CHECK - if all the posts are the same format, they will read as a campaign. Vary: a claim, a story, a practical tip, a question, a short observation.

6. WHICH ONES LINK BACK - not all of them. Posts that always drive to a link train the audience to scroll past. Recommend which two or three should carry a link.

7. WHAT IS LEFT OVER - ideas from the extraction that did not make this batch, banked for later.

Rules:
- Each post must be useful to someone who never sees the source
- Do not summarise the source in each post. Extract one idea and develop it.
- No 'new post on the blog' announcements

This draft is too generic. Diagnose it and tell me what to add.

DRAFT:
"""
[PASTE]
"""

WHAT I ACTUALLY KNOW ABOUT THIS: [YOUR EXPERIENCE, EXAMPLES, DATA, STORIES]
READER: [WHO]

Produce:

1. THE GENERIC SENTENCES - quote every sentence that could appear in any article on this topic by any author. These are the ones to fix.

2. THE ABSTRACTION LADDER - for each, what the concrete version would be. 'Businesses struggle with communication' → what business, what communication, what happened. Where I have given you the specific, write the replacement. Where I have not, write [NEED: the specific thing required].

3. THE MISSING PROOF - claims stated without evidence. For each: what would make it credible - a number, an example, a named case, a quote, or a description of a specific instance.

4. WHAT YOU KNOW THAT IS NOT IN THE DRAFT - from what I told you, the specifics I have that did not make it in. This is usually where the article's value is hiding.

5. THE ONE PARAGRAPH TO REWRITE - if I only fix one, which, and the rewritten version.

6. THE STRUCTURAL PROBLEM - if the draft is generic because it is trying to cover too much, say so. Breadth is the most common cause of vagueness, and no amount of sentence editing fixes it. Propose the narrower piece.

Do not rewrite the whole draft. Tell me what is missing, because I am the one who has the specifics.