These are the 30 AI prompts I really use in my consulting business in 2026, grouped by use case, with the exact wording plus the reasoning behind each one. Not 100 generic write-a-thing prompts, and not vague advice to give clear instructions. The actual prompts I run, ready to copy and adapt.
In this blog post I'm going to give you the 30 AI prompts I really use in my consulting business in 2026, grouped by use case, with the exact wording I use plus the reasoning behind each one. The version with the actual prompts. Not the version with "give the AI clear instructions" advice.
Every "AI prompts for marketers" article I've seen falls into one of two failure modes. Either it gives you 100 generic prompts that all amount to "write a [thing] about [topic]," or it gives you 10 prompts so abstract they're useless. Neither helps you produce better work.
This article is different. These are real prompts I run in real client work, copied verbatim from my prompt library. If they help you, brilliant. If a few don't fit your context, ignore them. None of this is theoretical.
I've been a marketing consultant for twenty-one years. I went all in on AI in 2024. I run roughly 200 AI prompts per day across content, research, sales, and operations work. The 30 below are the ones I run weekly or more often. They've been iterated dozens of times.
By the end of this blog you'll have the prompts, the structure that makes them work, and the framework for writing your own.
TL;DR
The structure that makes prompts work:
- Role (who the AI is)
- Task (what to do)
- Inputs (what context you're providing)
- Constraints (what NOT to do)
- Output format (exactly what you want back)
Most prompts you'll find online skip 4 and 5. That's why they produce generic output.
Content prompts (1-8)
1. Blog post structure from a transcript
> You are an experienced content editor for a marketing blog with a strong opinionated voice. I'll paste a 60-minute interview transcript below. Extract the 7 most useful insights for a working marketing consultant audience. For each, write a one-line takeaway and the supporting quote (verbatim). Then propose a blog post structure using the strongest 5 insights, with H2 headings and a one-line note on what each section covers. Constraint: do not invent any insight that isn't in the transcript.
Why this works: Role + specific count + verbatim requirement + explicit "do not invent" constraint. Catches AI's tendency to fabricate plausible-sounding quotes.
2. Headline variants
> Generate 15 blog post headlines for the following piece. Format: numbered list, one per line. Constraints: include a specific number where possible, avoid the words 'transform,' 'open up,' 'use,' 'big deal.' Vary the headline patterns , at least one how-to, one list, one contrarian, one specificity-led, one curiosity-led. Topic: [paste]. Audience: [paste].
Why this works: Explicit count, banned word list (catches AI tells), explicit pattern variety. Prevents the 15-identical-headlines problem.
3. Anti-fluff editor
> You are a brutal editor. Read the draft below and flag every sentence that doesn't add information, makes a claim without supporting it, or sounds like AI wrote it. For each flagged sentence, paste the original and rewrite it to be sharper or note that it should be cut entirely. Constraint: be honest, not encouraging.
Why this works: Role explicitly invites criticism, instruction to be honest rather than encouraging counters AI's default sycophancy.
4. SEO meta from finished post
> You're optimising a blog post for SEO. Read the post below and produce: (1) Yoast title under 60 characters including the focus keyphrase '[paste]'. (2) Meta description under 155 characters that includes the focus keyphrase and gives a concrete reason to click. (3) URL slug (lowercase, hyphens, under 60 characters, includes focus keyphrase). (4) Three internal link suggestions with anchor text. Output format: labelled bullets.
Why this works: Specific character limits, named focus keyphrase, named output format. Eliminates rework.
5. Article outline from a search query
> A user has searched Google for '[paste query]'. Based on the query, write the outline for the best possible blog post answering it. Outline format: H1, intro hook (1 line), TL;DR (3 bullets), 6-8 H2 sections with one-line summaries each, FAQ section with 5 likely questions, closing CTA placeholder. Constraint: assume the reader is a working marketer, not a beginner.
Why this works: Starts from search intent, not from a topic. Forces the outline to match what the user really wants.
6. Repurposing a blog into a LinkedIn carousel
> Take the blog post below and convert it into a 10-slide LinkedIn carousel. Slide 1 = hook. Slides 2-9 = one insight per slide with one supporting line. Slide 10 = CTA to read the full article. Each slide max 30 words. Tone: confident, anti-fluff, written for a marketing professional. Use my voice patterns from the blog itself.
Why this works: Word cap per slide, role instruction to mirror existing voice. Produces carousels that match your brand.
7. Email subject line testing
> Generate 10 email subject lines for the body copy below. Half should be curiosity-driven, half should be specificity-driven. Each under 50 characters. None should include the words 'don't miss,' 'urgent,' 'exclusive,' or use ALL CAPS. Format: numbered list, label each as [curiosity] or [specificity].
Why this works: Explicit length, explicit banned phrases, explicit labelling for testing.
8. Comparison table generator
> Build a comparison table for [topic]. Columns: feature, [option A], [option B], [option C]. Rows: 8 features most relevant to a [target audience] decision. For each cell, one line maximum. At the end, write a 3-sentence verdict on which option fits which type of buyer. Source: my evaluation criteria below.
Why this works: Explicit structure, length constraint, verdict requirement. Produces usable comparison content.
Research prompts (9-15)
9. Competitor messaging audit
> Read the homepage copy below from three competitors. For each, extract: (1) the main promise they make, (2) the proof they offer, (3) the implicit audience they're written for, (4) the words they use that signal positioning. Then write a 3-sentence summary of where the gaps and overlaps are. Constraint: quote verbatim where possible, don't paraphrase.
Why this works: Specific extraction targets, verbatim requirement, gap analysis. Far more useful than "compare these companies."
10. Audience interview synthesis
> I've pasted 5 audience interview transcripts below. Extract: (1) the language they use for the problem (verbatim phrases), (2) the language they use for the solution, (3) the objections they raised (with verbatim quotes), (4) the buying triggers mentioned. Format: four sections, verbatim quotes only.
Why this works: Forces verbatim language, which is the actual value of interviews. AI paraphrasing destroys this.
11. Pricing benchmark
> Find publicly available pricing for these 8 [category] businesses: [list]. For each, note: stated price, what's included at that price, what's excluded, the implied positioning (premium/mid/budget). Source URL for each. Format as a markdown table. If pricing isn't public, note 'not disclosed' rather than guessing.
Why this works: Explicit "don't guess" instruction. Catches AI's hallucination tendency on pricing.
12. Industry trend extraction
> Read the 5 industry reports linked below. Extract: the 3 trends mentioned in all 5 reports (consensus), the 3 trends mentioned in only one report (outlier), the 3 specific data points cited most often. For each, name the source. Format: three sections.
Why this works: Cross-source synthesis with consensus vs outlier distinction. Filters signal from noise.
13. Sales call objection analysis
> Read the 10 sales call transcripts below. List every objection raised, deduplicated, sorted by frequency. For each objection, paste the verbatim phrasing the prospect used most often. Constraint: do not interpret or rephrase , quote verbatim.
Why this works: Verbatim requirement is critical. Marketing teams need to know how prospects really talk, not how AI thinks they talk.
14. ICP language extraction
> From the 20 customer testimonials below, extract: (1) the 10 most common phrases customers use to describe the problem we solve, (2) the 10 most common phrases for the outcome they got, (3) the 5 most common phrases for what made us different. Format: three lists, exact phrasing only.
Why this works: Verbatim phrase extraction. This is what should populate your homepage copy, not your team's guess at what customers say.
15. SEO opportunity from keyword list
> From the keyword list below, identify the top 10 that meet ALL these criteria: (1) commercial intent (buying signal in the query), (2) monthly search volume above [X], (3) keyword difficulty below [Y], (4) we don't currently rank on page 1. Format: table with keyword, volume, difficulty, current rank, why it's a fit.
Why this works: Explicit multi-criteria filter. Produces an actionable shortlist not a brainstorm.
Sales prompts (16-22)
16. Cold email personalisation
> Write a cold email to [prospect name] at [company]. Context: I sell [offering] to [audience type]. The prospect's company [recent specific event from their website/LinkedIn]. Email constraints: under 100 words, no opening compliment, one specific observation about their business, one specific offer of help, one specific call to action. Tone: confident, not desperate. Output: subject line + body.
Why this works: Word cap, explicit anti-patterns (no opening compliment), specific call-to-action requirement. Produces cold emails that don't look like cold emails.
17. Discovery call follow-up
> Read the discovery call transcript below. Write a 6-sentence follow-up email that: (1) restates the prospect's primary problem in their own words, (2) confirms the timeline they mentioned, (3) restates the success criteria they articulated, (4) confirms the next step we agreed to. Format: email subject + body. No closing pleasantries.
Why this works: Mirrors back the prospect's own words, which builds rapport better than generic recap.
18. Proposal draft from notes
> Build a proposal for [prospect] based on the discovery call transcript below. Structure: (1) restate their problem in 2 sentences, (2) propose specific deliverables (3-5 max, each with scope), (3) propose timeline, (4) propose investment, (5) one paragraph on why I'm the right fit. Tone: confident, specific, no buzzwords. Length: under 1 page.
Why this works: Structure constraints + length cap + tone instruction. Avoids the 12-page proposal trap.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
19. Objection handling script
> The prospect raised this objection in our last call: '[paste objection verbatim]'. Write 3 different response strategies: (1) acknowledge-and-redirect, (2) reframe, (3) provide evidence. Each under 80 words. Note which strategy fits which type of prospect (analytical, emotional, time-pressed).
Why this works: Three strategies with explicit categorisation. Lets you choose based on the actual prospect rather than guessing.
20. Case study extraction
> Read the client interview transcript below. Build a case study with: (1) one-line client identifier, (2) the situation (3 sentences), (3) what we did (3-5 bullet points), (4) the result (with at least 2 specific numbers), (5) one client quote (verbatim). Total length under 350 words.
Why this works: Specific numbers requirement + verbatim quote + length cap. Produces case studies people really read.
21. Sales call prep brief
> Build a 5-minute prep brief for an upcoming sales call with [prospect]. Sources: their LinkedIn profile, their company website, their recent activity below. Output: (1) what they likely want from this call, (2) the 3 questions I should ask first, (3) the 2 things I should not say, (4) one likely objection I should be ready for. Format: 4 sections, max 100 words each.
Why this works: Forces the AI to do the prep work you'd otherwise do manually. The 'things I should not say' section is uniquely valuable.
Operations prompts (23-30)
23. Weekly business review summary
> Read the metrics dashboard below from the past week. Produce: (1) the 3 most important changes vs last week, (2) the 3 metrics most worth investigating further, (3) one recommendation for action this week, (4) one question I should put to the team. Format: 4 sections, plain English, no jargon.
Why this works: Focuses attention on what matters. Most dashboards are too much data to read; this produces the read-out.
24. Inbox triage
> Read the 30 emails below. Categorise each as: ACT (needs response from me today), DELEGATE (can be handled by someone else), DEFER (can wait this week), IGNORE (no action needed). For each ACT and DELEGATE, write a one-line note on what's needed. Format: 4 sections, email subject lines only.
Why this works: Forces a decision on every email. Reduces 30 emails to a 1-page action list.
25. Meeting agenda generator
> Based on the previous meeting notes below and the project status, propose an agenda for the next meeting. Format: (1) 3 items to discuss with time estimates (max 60 min total), (2) one decision needed by the end, (3) one action coming out of the meeting. Constraint: no 'updates' or 'check-ins' , every item must have a decision attached.
Why this works: Forces decisions, not status updates. Eliminates 50% of meeting time.
26. Standard operating procedure draft
> I'll describe a workflow I do regularly. Build an SOP for it. Sections: (1) when this SOP is triggered, (2) who owns it, (3) inputs needed, (4) steps in order (each one action), (5) outputs produced, (6) common failure modes and fixes. Format: numbered sections, written so a new hire could follow it on day one.
Why this works: Forces consideration of failure modes, which most SOPs skip.
27. Status update from chat history
> Read the Slack thread below from the past week. Produce a 5-bullet status update suitable for a client: (1) what's been done, (2) what's in progress, (3) what's blocked and what we need from the client, (4) decisions needed, (5) next milestone with date. Tone: confident, factual, no apologies.
Why this works: "No apologies" constraint catches the over-apologetic default tone AI produces in client comms.
28. Risk register from project notes
> Read the project notes below. Identify the 5 highest risks. For each, name: the risk, the probability (low/med/high), the impact (low/med/high), the trigger that would tell you it's materialising, the mitigation if it does. Format: risk register table.
Why this works: Forces specific triggers and mitigations. Produces a usable risk register, not a list of vague concerns.
29. Decision log entry
> I just made the following decision: [paste]. Write a decision log entry capturing: (1) the decision in one sentence, (2) the date, (3) what I decided NOT to do (alternatives considered), (4) the reasoning, (5) the criteria I'll use to review whether the decision was correct. Format: 5 sections, brief.
Why this works: The "what I decided NOT to do" section captures rationale that's normally lost. Saves arguments in 6 months.
30. Process audit
> Read the workflow below. Identify: (1) the 3 steps most likely to fail, (2) the 2 steps that add the least value, (3) the 1 step where automation would have the biggest impact. For each, name a specific improvement. Format: 3 sections, named improvements only , no generic recommendations.
Why this works: Forces specific improvements per step. Produces an actual improvement list, not a general critique.
How to use these in your own business
Three rules:
Rule 1: Adapt the language. Copy the structure verbatim, but swap the language to match your domain. The prompts above use marketing-consulting language. If you sell something else, the verbs and nouns should reflect your work.
Rule 2: Test before you trust. Run each prompt 3 times on the same input. If you get wildly different outputs, the prompt is under-specified. Add constraints.
Rule 3: Iterate the prompt, not the output. When an output isn't right, fix the prompt and re-run. Don't fix the output by hand , that defeats the purpose. The prompt should produce work you can ship.
What's NOT in this list
Some prompt categories I deliberately don't use:
"Write me a viral LinkedIn post." Viral can't be prompted into existence. The right prompt is "write a LinkedIn post that captures this insight clearly," and you handle distribution separately.
"Pretend you're a [famous person]." Imitating named individuals is a copyright and brand minefield. Use voice patterns, not impersonation.
"Be creative." AI's idea of creative is averaging. Specificity beats creativity instructions every time.
"Make it more engaging." Vague. Replace with specific constraints (sentence length, word choice, structure).
Frequently asked questions
Should I use these prompts in ChatGPT, Claude, or another tool? They work across major LLMs. The structure is what matters, not the platform. I use Claude for long-form work and ChatGPT for quick tasks.
Should I build a custom GPT or assistant for each prompt? Only if you'll run it 20+ times per month. Below that, the maintenance cost outweighs the convenience.
How do I write my own prompts that work? Follow the role-task-inputs-constraints-output structure from the TL;DR. Skip any element and the quality drops.
Should I share my prompts with my team? Yes. Prompt libraries are competitive assets. Document the ones that work, share them internally, iterate together.
Should I share my prompts publicly? That's a positioning decision. I share mine because I sell consulting, not prompts. If your product IS the prompts, keep them.
How often do you iterate on a prompt? Most of my "good" prompts have been iterated 10-20 times. The first version of a prompt is rarely the final version.
Want my full prompt library?
The 30 above are the most-used. I have roughly 200 in total covering content, sales, ops, research, recruiting, and personal productivity. Clients get access to the relevant subset as part of any engagement.
[Book a prompt library session →](/contact)
I'm Lilach Bullock. I've been a marketing consultant for twenty-one years. I went all in on AI in 2024. I work with founders and marketing leaders who want AI to really move their numbers, not just their tool stack.
Related: the best content repurposing tools
