> ## Content Index > Fetch the complete content index at: https://www.productmarketingalliance.com/llms.txt > Use this file to discover other available public pages before exploring further. # My AI prompting workflow for positioning - URL: https://www.productmarketingalliance.com/my-ai-prompting-workflow-for-positioning/ - Published: 2026-10-12T12:00:57.000Z - Updated: 2026-10-12T12:00:56.000Z - Description: A PMM's AI prompt workflow for positioning: the inputs to gather, 6 prompts to run, and where human judgment still beats the model. - Author: Swati Singh - Tags: AI, Positioning & messaging, Articles, #AI A few months ago I watched someone prompt an AI chatbot with a list of product features and a rough idea of who used the product, and ask it to "write a [positioning statement](https://www.productmarketingalliance.com/positioning-statement-template-framework/)." What came back was a two-line, fill-in-the-blanks sentence, the kind you could paste onto almost any [SaaS product](https://www.productmarketingalliance.com/developer-marketing/what-are-saas-integration-platforms-and-why-are-they-important/) and it would still sort of fit. It's not the only time I've seen this. I once sat across from the founder of a company doing eight figures in revenue, a founder who considered himself the product marketing expert for his own company, and he genuinely couldn't tell me why he'd need a dedicated PMM for this kind of work. And that's the problem that makes positioning so hard to crack. Here's the thing: "AI is changing marketing" is a headline, not an insight. Similarly, using AI for positioning is a workflow, not a **fill-in-the-blank sentence**. If you come from a technical or engineering background, as I do, I suspect you jumped on AI workflows the moment you got real access to them, because engineers are trained to think in systems, not one-off tasks. And over the last three years, I have been building an AI prompting workflow to refine positioning, and before sharing this, I want you to remember that it is always a work in progress when you are using an AI assistant to build it. > ***"Workflow means building the flow of the work into a system, not typing one clever sentence into a chatbot."*** ## The old way – and why it stayed broken Before any of this, [positioning](https://www.productmarketingalliance.com/your-guide-to-positioning/) work followed a fairly standard, industry-wide pattern: dig through whatever customer and product data your company had, look at how customers were actually perceiving your product's value, then go find the best positioning templates and frameworks out there, study how competitors and category leaders were positioning themselves, and try to reverse-engineer what was working for them. This is, roughly, the tradition that [April Dunford's](https://www.aprildunford.com/) widely-read *Obviously Awesome* tried to formalize, a structured approach meant to help businesses find and communicate their advantages on their own terms, built around naming competitive alternatives, unique attributes, and the right market category rather than guessing at a tagline. It's a genuinely useful framework that I used for years, and a lot of the input categories I use below still map onto it. But in practice, even with a good framework in hand, the process stayed slow and hard to defend. Teams pulled web traffic reports to see which keywords and headlines were actually performing, ran round after round of [A/B tests](https://www.productmarketingalliance.com/is-a-b-testing-still-relevant-in-b2b-marketing/) before committing to a single positioning line, and still spent most of their time not on the writing but on collecting the data in the first place. Getting [real alignment from stakeholders](https://www.productmarketingalliance.com/internal-stakeholders-and-senior-leadership-are-we-all-rowing-in-the-right-direction-across-product-marketing-and-sales/) was a constant fight, because everyone had an opinion and nobody had the same data in front of them at the same time. And the moment a new product entered the market, or the category itself shifted, the whole thing had to be redone, which made [repositioning](https://www.productmarketingalliance.com/monzos-repositioning-strategy/) feel like a genuine ordeal rather than a normal part of the job. And to be honest, AI didn't remove any of that work. What it changed is how fast I can move through it, and how many angles I can actually test before I commit to one. And this is what I am going to talk about today. ## Before you prompt anything: get your inputs in order Here's where most people go wrong with AI and positioning – they skip straight to asking AI to write the positioning. My rule is the opposite: your goal at this stage isn't to build positioning yet; it's to understand how your customers are actually perceiving the product. I want you to collect pieces of relevant information already synthesized *before* you write a single positioning prompt. Without this, you don't have positioning – you have an opinion about positioning. Solid positioning is built on inputs across five categories: | Category | What to gather | Where to find it | | ------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------- | | Market size & trends | TAM/SAM/SOM, category maturity, growth trajectory, regulatory or tech shifts, analyst perspective (Gartner, Forrester) | Product strategy decks, executive decks, investor pitch decks (startups almost always have this already, find whoever pitched to investors) | | Customer & buyer data | Win/loss interview themes, the customer's own language (not yours), segmentation, jobs-to-be-done, churn reasons | Win/loss calls, support tickets, sales call transcripts, community/forum chatter, surveys | | Competitive intelligence | Competitors' own positioning and messaging, win/loss patterns against them, pricing/packaging, feature gaps weighted by what buyers actually care about | Competitor sites, decks, ads, battlecards | | Product data | Current capabilities and near-term roadmap, usage data on what drives activation/retention, defensible differentiators, proof points | Roadmap docs, usage analytics, case studies | | ICP & persona | Firmographics, technographics, triggering events, budget authority, plus role, KPIs, pain points, buying role, objections, and the buyer's own emotional drivers | CRM, best-retention account patterns, persona interviews, current users profile analysis | Once this is pulled together, update anything stale, benchmark it against where the product and the market actually stand today, and then start feeding it to your AI assistant. In practice, here's what I connect before I run what I think of as the positioning engine: ### The stack behind it | Input type | Tools | | ------------------------ | ----------------------------------------------------- | | Structured data / decks | Excel, Power BI, Google Docs, PowerPoint | | Competitive intelligence | Klue, G2, Crayon | | Call intelligence | Gong, Chorus | | Product & UX data | Confluence, Miro, Notion, Pendo, Qualtrics, Amplitude | All of it goes in before AI writes a single word about positioning. ## The actual prompt chain Once the inputs are synthesized, this is the sequence I run. Think of it as one continuous thread, not six separate conversations, each prompt builds on what came before it. **The shape of it:** > Collect raw data → generate → tighten → usable framework → stress-test → adapt → MVT test the finalists → finalize positioning. ### Step 1: Load the context *Run this first, or save it as a Project/custom instructions so you're not re-pasting it every time.* You are acting as a senior product marketing lead. I'm going to give you structured research about ****\[COMPANY NAME\]** so you have full context before we work on positioning. Read and hold all of this – don't generate anything yet, just confirm you've absorbed it and summarize back the 3 biggest tensions or opportunities you notice across the data. ****MARKET DATA:** \- Category: \[...\] \- TAM/SAM/SOM: \[...\] \- Market trends/shifts: \[...\] \- Analyst perspective: \[...\] ****CUSTOMER & BUYER DATA:** \- Win/loss themes: \[...\] \- Voice of customer (their exact language): \[...\] \- Segmentation: \[...\]Jobs-to-be-done: \[...\] \- Churn reasons: \[...\] ****COMPETITIVE INTELLIGENCE:** \- Direct competitors + their positioning: \[...\] \- Indirect alternatives (including "do nothing"): \[...\] \- Where we win / lose competitively: \[...\] \- Pricing/packaging comparison: \[...\] ****PRODUCT DATA:** \- Core capabilities today: \[...\] \- Near-term roadmap: \[...\] \- Defensible differentiators: \[...\] \- Proof points: \[...\] ****ICP & PERSONA:** \- Firmographics / technographics / triggering events: \[...\] \- Role, KPIs, pain points, buying role, objections, language used unprompted: \[...\] ### Step 2: Generate the angles Using everything I just gave you about \[COMPANY NAME\], generate 6 distinct positioning statement angles. Each one should lead with a different value driver (cost, speed, risk reduction, ease of adoption, status/prestige, outcome/ROI). Rules: \- Use the customer's own language from the voice-of-customer data wherever possible – flag which phrase you pulled and from where. \- Do not use generic SaaS phrasing ("empowers teams to," "seamless," "unlock," "next-generation," "robust platform"). \- Each angle: (1) a one-sentence positioning statement, (2) 2-3 sentences of reasoning tied to the specific data I gave you. \- Note which ICP/persona each angle is strongest for. \- At the end, flag the 1-2 weakest angles and why – give me your honest, critical read, not just six polished options. **What I look for:** at least 2-3 angles that use a phrase I actually recognize from a real interview. That's my signal the AI absorbed the input instead of generating generic category filler. ### Step 3: Tighten it, then try to break it Once I've picked one or two angles I actually want to pursue: ****PROMPT 3: Tighten the chosen angle(s)** I'm going with Angle \[#\] as our positioning direction. Now refine it: 1\. Tighten the one-sentence statement to under 20 words – cut anything that isn't load-bearing. 2\. Rewrite it 3 ways at 3 registers: (a) an internal strategy doc, (b) the homepage hero, (c) how a rep would say it out loud on a discovery call in one breath. 3\. Tell me what got lost or changed between registers, and whether that's a problem. ****PROMPT 4: Stress-test against competitors and objections** Now pressure-test this positioning against the competitive and objection data I gave you earlier: 1\. For each of our top 3 competitors: does this hold up, or could they credibly claim the same thing? Be blunt. 2\. Take our top 3 buyer objections – does this positioning defuse them, ignore them, or make them worse? Play devil's advocate: argue the strongest case AGAINST this positioning – why might it fail or not survive a skeptical buyer? ### Step 4: Turn the survivor into a usable framework Once a statement has actually held up under pressure: ****PROMPT 5: Translate into a messaging framework** Turn this refined positioning into a working messaging framework: 1\. Core positioning statement (final version) 2\. 3 supporting message pillars — each with the pillar statement, the proof point(s) behind it, and which persona it serves 3\. A one-line "so what" per pillar, in the buyer's language 4\. A 30-second elevator pitch, spoken register ****PROMPT 6: Adapt across personas** Take this messaging framework and adapt the emphasis (not the core truth, just what's foregrounded) for \[persona 1\], \[persona 2\], \[persona 3\]. For each: which pillar leads, which gets cut for space, and why — based on what that persona is measured on and worried about. With AI, there's no cost to doing this over and over until it actually clicks. And if a pressure test surfaces a better angle than the one you started with, run it through the same test; nothing says you have to stop at your first pass. ## Where AI gets it wrong, and where I don't hand it the pen AI can do the analysis, surface insights faster, and synthesize bulk data in a way that used to take me months. That part is real, and it's the part that gets all the attention. What it still misses is the human factor, specifically, judgment about *value*. To an AI, every pain point in a transcript reads as roughly the same weight: a frustrating problem is a frustrating problem. A human who's actually talked to these customers knows the difference between a [pain point](https://www.productmarketingalliance.com/understanding-user-pain-points/) someone will pay to fix and one that's a nice-to-have sitting on a roadmap forever. That distinction doesn't show up in the data the same way it shows up in the room. So my rule as a PMM is simple: bring your final decision-making and strategic judgment to this as a leader who has actually talked to humans throughout the process, not just fed them into a model. A few things I hold myself to on every pass: - **Always sanity-check the output.** Criticize AI's work the way you'd criticize a junior teammate's first draft, find the flaws, and trust your gut when it's backed by data, not instead of the data. - **Never finalize before testing.** I've watched too many PMMs commit to a positioning statement without running any real test on it first. Build your top 2-3 candidates from the research and the AI work, then pressure-test them with real customers before you finalize anything. ## The principles, if your workflow looks different from mine 1. **Every positioning statement is built on data — more or less doesn't matter, but it's never built on a guess.** It should always trace back to something your users actually said or did. 2. **Pressure-test in the real world before you finalize — don't call it done until you've sold something with it.** [Multivariate testing (MVT)](https://www.kameleoon.com/blog/multivariate-testing) is the fastest way to do this. Private assets (pitch decks, emails, gated ebooks) are cheap and fast to test; public assets (LinkedIn, your website, paid ads) are expensive to redesign once they're live, so choose where you test carefully. 3. **Know your product, your customers, and your industry better than anyone else in the room.** Don't hesitate to go further than the job technically requires to actually become the expert. Treat the category like you own it. Be the leader, not just the person running the prompts. *(If you've found a principle I'm missing, I'd genuinely like to hear it.)* ## Where I'm taking this next AI workflows are changing faster than almost anything else in our toolkit right now, faster than I can keep this article updated. So here's my actual ask: tell me which parts of this you have already tried, and which parts you'd change. And if you've got all this data sitting in your company already but still can't land on a positioning that holds up, I'd genuinely like to talk it through. Drop it in the comments, or find me on [LinkedIn](https://www.linkedin.com/in/swatisingh25/) for a conversation.