My Approach to AI as a Product Marketing Consultant

17 Jul, 2025

In talks with other product marketing consultants, I’ve learned that there’s not a lot of early, open communication with clients about how we use AI to enhance our work, or their expectations around privacy and quality control. No one I've spoken with has seen AI use clauses in their client engagement paperwork, either.

There's an opportunity to get ahead of these conversations about AI. This post is my initial draft of a light-weight AI policy. It’s not intended as a legal document, just an overview of my approach: the tools I use, how I use them, and what I do to protect client confidentiality in this brave new world.

My main goal with this post is to demonstrate to potential clients (hello!) that I’m using AI responsibility, reasonably, and in a way that adds value to our engagement.

My secondary goal is to encourage other consultants (hi!) to articulate their approaches to using AI in client work. Eventually, our clients will catch up and dictate the rules of engagement. We’ll start seeing AI clauses in contracts. Until that happens, they’ll be looking to us to set the standard.

My friend Amber Rhodes has already shared her stance on AI, and I want to see yours. Share it with me and I’ll link it in this post.

General standards

• Typically, as a Product Marketing Manager, I’m not handling information that’s particularly scary (ex. PII, health data). But if I’m ever uncertain of how to handle specific information, I’ll ask.

• That said, sometimes I do have access to data that’s on the edge, such as strategy documents, product roadmaps, or positioning for features that haven’t launched yet. In these cases, I use AI tools that run locally to avoid any potential leaks. Call it an abundance of caution.

• If I’m working as an embedded resource with access to a client's internal systems (e.g. Slack, G Suite), I’ll use any internal tools they have available over my personal tools.

• I never outsource strategic thinking or deep analysis. Clients never get anything from AI that I haven’t gut-checked or massaged first.

• I keep client work in a standalone file inside my AI instances (ex. a dedicated Project in Claude) and delete it after the engagement ends. Any material generated for their work—poof!—disappears.

• If for any reason a client would prefer I not use any of the tools below to support our project together, I encourage them to speak up.

AI tools usage overview

Fathom – Note taker

• Privacy policy summary
  • Collects personal data (name, email of meeting attendees) and meeting content (video, transcript) as needed to render their services
  • Does not use content to train their AI
• How I use
  • Record prospect conversations to create an accurate statement of work
  • Record discovery conversations to create a transcript or summary of key points from the call
• Acceptable input
  • Anything recorded in the conversation (I let the client know I'm using an AI note taker
• Unacceptable input
  • Anything the client wants to leave out of the recording (I will pause the recording if they ask)

Perplexity (Pro, Web) – Researcher

• Privacy policy summary
  • Collects queries and conversation history to improve services
  • Pro users can opt out of having conversations used for model training (I’m opted out)
  • Processes data through third-party AI providers (OpenAI, Anthropic, etc.)
• How I use
  • Research
  • Fact-checking
• Acceptable input
  • Generic questions about the client’s industry, customers, and competitors - Examples:
    • What are the top three benefits for IT managers using a compliance automation tool? What are the top three benefits for IT managers using a compliance automation tool?
    • What are the top sales platforms CROs at Fortune 100 platforms consider? What makes them strong or weak against one another?
  • Hypotheses about a client’s specific business or industry - Example:
    • It seems as though small business owners aren’t interested in advanced reporting for their candidate pipeline. Is that true?
• Unacceptable input
  • Content that explicitly mentions the client business by name
• Notes
  • While I don’t read through all the links that Perplexity links as part of a query, I do spot check references to make sure they’re not content farm slop (which happens)

Claude (Pro, Web) – Chatbot and assistant

• Privacy policy summary
  • Collects conversation data to improve services
  • Does not train on Pro user conversations without explicit consent (I’m opted out)
  • May collect information from users who opt into feature previews (I’m opted out)
  • Conversations may be reviewed for safety purposes
• How I use
  • Marketing my business
  • Relationship management
  • Problem solving partner
  • Rough drafts and copy edits for external-facing content about existing, public-facing features or campaigns
• Acceptable input
  • First draft social posts, email content, and blog content from notes
  • First draft statements of work
  • Drafts of client emails or project updates (Ex. Project timelines)
  • Drafts of external-facing content
  • General descriptions of a challenge or scenario for problem-solving - Example:
    • My client has a new CMO who has paused our project. What’s are some options for moving forward?
    • My client is pivoting from a bottoms-up sales approach to a top-down sales approach. What are some case studies of businesses who have done this successfully? What might need to change in my product marketing strategy to support this move?
• Unacceptable input
  • Client business or full contact names
  • Launch dates or other confidential timelines
  • Unreleased features or details

Llama (Local) – LLMs for language processing tasks

• Privacy Policy Summary
  • Collects anonymous telemetry data (usage, product crashes)
  • Never collects personal data (chat content, documents, etc.)
  • All processing happens locally on my device
• How I use
  • Information synthesis of call notes, shared documentation, and existing content shared by the client
  • Rough drafts and copyedits for internal-facing content
  • Rough drafts and copyedits for external-facing content about features or campaigns that aren’t yet public
• Acceptable input
  • Shared documentation (ex. Internal-facing strategy documents and decks)
• Unacceptable input
  • Client customer PII or similarly sensitive documentation

Example scenario

Let’s imagine a client approaches me about creating sales assets for a new audience.

They’ve started offering a small business tier of their SaaS product to get customers into their ecosystem at a lower price point and continue to be their product of choice as they grow.

Here are some ways I’d use AI tools throughout the project lifecycle.

Pre-Project

• Perplexity: Shares relevant context on the business ahead of my first call
• Fathom: Turns conversation transcripts into next steps, draft deliverables list for statement of work
• LLaVA Llama 3B (Local): Supports content creation for statement of work based on documents I've used in the past

Discovery

• Perplexity: Shares additional context on the differences between small business and enterprise challenges in the markets
• Fathom: Turns conversation transcripts into summaries with takeaways relevant to the project
• Llama (Local): Supports outlining the one-pager based on Perplexity research inputs, Fathom conversation transcripts, and documentation shared during the discovery process

Draft

• Perplexity: Acts as a thought-partner to validate or reject specific hypotheses I may have about the new audience as I work through the one-pager content
• Fathom: Discovery conversations can be queried if I want to revisit an idea or point that was made
• Llama (Local): Supports making headlines punchier, improving content flow, ensuring grammar/spelling is correct

Final deliverable

• Perplexity: Provides extra context on a client comment before I follow up with them for more info
• Llama (Local): Supports implementing client feedback on content tweaks or copy edits
• Claude: Supports streamlining my long list of notes to share with the client via email with final delivery

Starting the conversation

This is the jumping-off point for how I use AI in my work. I’ll continue to update this post based on feedback from customers and other consultants.

Have questions or feedback? I'm all ears—uh, eyes. Send me an email.