Your AI doesn’t have a point of view.

Most AI marketing workflows start with a prompt. This often creates polished content, but the content can be generic. The problem is not always the model. The problem can be missing context and a clear point of view. I found this problem while I evaluated AI use cases for product marketing. I built PMK to address it.

PMK is an agentic workflow that starts with the marketer's point of view. It uses this point of view to anchor the narrative. AI then helps research, test, develop, and apply the narrative. It does not create the point of view for the marketer.

In this session, I will show how I built the workflow. I will explain how context moves through it, where human judgment is necessary, and what I learned from building it. I will also share a framework to help marketers decide which AI use cases to automate, augment, or keep human-led. Attendees will get a practical approach to build AI workflows that support human thinking instead of replacing it.

Transcript

1. ALL DAY AI · OCTOBER 22, 2026 Your AI doesn’t have a point of view Building an AI workflow that does Mac Gupta · Precision GTM
2. linkedin.com/in/mac-gupta Mac Gupta I started as a software engineer and kept getting pulled into go-to-market decisions. Eventually I stopped resisting and ran them full time. Since then I’ve worked across product marketing, partner GTM, and sales enablement, mostly in complex, multi-product portfolios sold into enterprise accounts. What I found again and again: the GTM system was broken upstream of where people were looking. The product was fine. The team worked hard. But the ICP had drifted, the narrative wasn’t landing with the economic buyer, or the partner motion ran on relationships instead of a repeatable play. 15+ years of GTM leadership MBA, Chicago Booth MS, Computer Science Work experience: IBM · Software AG · Pitney Bowes · WealthEngine · Rocket Software Your AI doesn’t have a point of view · All Day AI 2026 Newsletter
3. AI is producing content faster than anyone can review it We gave everyone a printing press. Nobody hired an editor. What an enterprise product marketing team told me: Everyone is off creating their own content. There’s a frenzy of new agents, and they aren’t using approved messaging. People think they know marketing better than the marketers. So much AI slop is being created that product marketers can’t keep up. The approved messaging gets refreshed on a cadence, about once a quarter. Your AI doesn’t have a point of view · All Day AI 2026
4. Nobody owns these new jobs yet PROFOUND CLAY Shows you what AI engines say about you. Writes hundreds of personalized emails before lunch. Someone has to decide which of it should Someone has to check they all describe the change your story. same company. Job 1: keep the story current Your AI doesn’t have a point of view · All Day AI 2026 Job 2: keep the story consistent
5. The fix: an approved story with an owner Right now the story plays a game of telephone. Every hop retells it. PMM “SOC 2 Type II certified” → Enablement deck “Enterprise-grade security” → Field “Bank-grade security” Then a competitor asks to see the certificates. Give every person and every agent the same source. → AI agent “Certified for ISO 27001 and GDPR” → Customer hears “Reduces our audit risk”
6. PMM + AI agents run the loop The story learns from what happens when it meets the market. GOVERNANCE · keeps the story consistent GTM Context Auditor + Writer In the field the approved story check and fix outbound Clay, field reps and mktg brought in by a approved changes FEEDBACK · keeps the story current person today PMM decides Analyst Market signals confirm, update or retire flags where they diverge Profound, calls, analysts Company changes launches, M&A, review dates A constitution: agents propose amendments. PMM ratifies them. Your AI doesn’t have a point of view · All Day AI 2026
7. Meet Paylark It checks every payment before it reaches the bank. A fictional accounts payable control platform for mid-market finance teams. Its GTM Context holds six BOUNDARIES TO WATCH approved documents: SAP S/4HANA connector Planned for later Bank-detail verification Beta Free trial Pilot on their own data Certifications SOC 2 Type II only Category POV Positioning ICP and personas Competitors Capabilities and boundaries Brand voice Your AI doesn’t have a point of view · All Day AI 2026
8. Let’s break some rules. 1 2 3 4 The foundation Governance: Clay emails Feedback: a Profound Same rules everywhere export
9. The fix kept the rep’s voice BEFORE · CLAY ROW 3, TAILSPIN LOGISTICS AFTER · THE WRITER’S FIX IT Subject: Your AP team Subject: Your AP team Noticed Tailspin is hiring two AP clerks. Noticed Tailspin is hiring two AP clerks. Before you hire, take a look at Paylark. Our touchless AP replaces Before you hire, take a look at Paylark. Paylark checks every line of a your AP team with AI that reads invoices 10x faster than a person, so payment run against vendor history before it goes to the bank, you can skip those hires entirely. It’s a seamless switch. Hold · 5 / 100 · 6 contradictions and routes the exceptions to your team, it doesn’t replace them. If it’s useful, we can run your last six months of payments through show what it finds. Ready it toand send · 95you / 100 Underlined: flagged by the Auditor. Bold: changed by the Writer. Real rehearsal output, October 17, verbatim. Your AI doesn’t have a point of view · All Day AI 2026
10. LESSON 1 · WHAT FELL APART A true signal led an agent to a false promise WHY PMM SAID NO ANALYST OUTPUT · REHEARSAL RUN, OCT 17 “…name vendor impersonation or payment fraud explicitly as a category-level threat the new way is built to stop…” The evidence was real: buyers do ask AI engines about vendor Bank-detail verification is still Beta Our boundaries: never promise to stop fraud Our voice: no fear-based fraud pitches Agents catch false facts. They struggle with restraint. fraud. The forecast said rain. The agent wanted to sell umbrellas we haven’t built yet. Your AI doesn’t have a point of view · All Day AI 2026
11. LESSON 2 · WHAT FELL APART Some rules shouldn't be retrieved Search finds what’s similar. “Never say seamless” is similar to nothing. Retrieved when relevant Always on, in every call Positioning Banned words Competitive counters Boundaries and non-goals Persona detail The whole brand voice You don’t look up “wear a seatbelt” when the road gets interesting. THE BUG THE NEAR MISS Approving a document also indexed it, voice guide included. The Two products, one unlabeled rule list: Product A’s boundaries would voice guide showed up in chat as a cited source. have governed Product B. Fixed before launch. Your AI doesn’t have a point of view · All Day AI 2026
12. LESSON 3 · WHAT HELD UP Signals stay out of the answers Great kitchens read every suggestion card, and keep them in the kitchen. WHAT WE HEAR WHAT WE SAY Market signals GTM Context Profound reports, call notes, analyst takes. Unverified. Separate table, separate search Sellers can’t see signals Make the wrong path impossible to take. Your AI doesn’t have a point of view · All Day AI 2026 PMM approves Approved. The only thing sellers and agents can quote. The answer code can’t read the signals table
13. What it costs 1 Someone has to own the story 2 Approval is a queue that needs work 3 The Auditor is strict, sometimes pedantic Every document needs an owner and a review date. Writing it and keeping it current is the real work. Every suggested edit waits for PMM, and the Analyst takes about 35 seconds a run. In rehearsal it called a ski resort’s “seasonal vendors” opener off-profile. Someone still needs to judge the Review pile. Worth it when the alternative is a story nobody owns. Your AI doesn’t have a point of view · All Day AI 2026
14. Steal these, whatever tools you use 1 2 3 and give it one owner. it. Make PMM approve them. 20 outbound emails against Write down the approved story, Let agents propose changes to On Monday, check last week’s your positioning doc. Count the contradictions. Same loop on any stack. Only the tools change: SharePoint, Copilot and Teams in a Microsoft shop; Google Drive, ChatGPT or Claude, and Zoom elsewhere.
15. Let’s connect. If you’re working through this in your organization, I’d love to show you what we’re building. Mac Gupta · Precision GTM mac@precisiongotomarket.com precisiongotomarket.com linkedin.com/in/mac-gupta Questions? Precision GTM newsletter
16. Each kind of knowledge has its own place Think of a restaurant: the menu, the house rules, and the suggestion box. What’s in it How the AI uses it Who can see it The menu: approved content Positioning, competitors, personas Retrieved when relevant Everyone The house rules: always-on rules Banned words, boundaries, voice Injected into every call Everyone The suggestion box: market signals Profound exports, call notes, analyst Read only by the Analyst and Scout Product marketing only takes One GTM Context, everywhere: web app · Slack · Claude via MCP · a brand kit for Writer, Jasper and AirOps Your AI doesn’t have a point of view · All Day AI 2026
17. A product marketer and a team of agents One product marketer runs the team. Sales and marketing use what it produces. Analyst Auditor Writer Concierge Librarian Flags “we keep Checks content Rewrites only what Answers sellers’ Keeps the GTM hearing X, but our against the GTM was flagged questions, with Context from going story says Y” Rehearsal: a 90-row Context Rehearsal: 6 Clay Rehearsal: two sentences changed, sources Same answers on the stale Lists documents past Profound export, 5 or 6 emails in about 30 score 5 to 95 web, in Slack and in their review date or suggested edits seconds, 3 held Claude still in draft Your AI doesn’t have a point of view · All Day AI 2026
18. Live today vs. planned LIVE TODAY PLANNED • GTM Context with owners, review dates and approvals • A live Clay column (API keys, REST) • Concierge answers on the web, in Slack and in Claude (MCP) • Pulling Profound data automatically • Auditor: single texts, call transcripts, CSV batches • A CRM connector • Fix it: rewrite what was flagged, then re-audit • Sign-in for claude.ai and ChatGPT • Analyst: Profound and other AI visibility exports • Audit trends • Brand kit export for Writer, Jasper and AirOps Your AI doesn’t have a point of view · All Day AI 2026 connectors
19. The audit that loved a robots.txt 100 robots.txt We audited a robots.txt file. It scored 100. Silence read as agreement. It never said anything wrong, so it was “perfectly on-message.” Small talk on sales calls would have scored 100 too. The fix: the Auditor first asks whether there is anything here to check. If not, it reports “nothing to check.” Your AI doesn’t have a point of view · All Day AI 2026
20. How the CSV audit picks columns CHECKED: THE MESSAGE Subject Body Opener SKIPPED: THE PROSPECT Personalized line P.S. Company description LinkedIn summary Job title Email address A Clay export is mostly about the prospect. Auditing their company description against our story would grade the prospect’s own words. So the audit reads columns named like part of an email, and you get the results back as a CSV with a status column for Clay. Your AI doesn’t have a point of view · All Day AI 2026