Your AI Content Looks Like AI Content (And That’s Costing You)

Part 4 of a 5-part series on building AI-GTM infrastructure that actually works.

The customer-facing layer where your AI infrastructure finally meets the market. Part 4 of a 5-part series on building AI-GTM infrastructure that actually works.

If you’re just joining us, hiii. Welcome. You’re jumping in just in time to hear me finish out this rant on building an AI-based GTM infrastructure that’s rooted in reality, not that insane “I just made an agent that replaced my 9-person marketing team” nonsense.

A quick recap:
- Part 1: Your data is clean.
- Part 2: Your AI knows what qualified means.
- Part 3: Your workflows are automated.

Now comes the part where most B2B marketing teams completely blow it: actually talking to customers.

I reviewed a demand gen campaign last month where everything behind the scenes was perfect. Clean CRM, smart intent tracking, automated workflows triggering at exactly the right moments. They’d done Parts 1-3 of this series flawlessly.

Then I read the emails their AI was generating.

Every single one started with “I hope this email finds you well.” Every value prop was a generic restating of their website copy. Every call-to-action was “Let’s schedule a time to chat.” The personalization was limited to {{FirstName}} and {{CompanyName}} tokens.

Their VP of Marketing said: “But we’re using AI to personalize at scale.”

They were personalizing nothing at scale.

This is where the rubber meets the road. And honestly y’all, it is hard. I can’t lie. All that infrastructure work from Parts 1-3 culminates in the moments where your brand actually talks to customers. And most companies are using AI to create perfectly mediocre content faster than ever before.

---

👋 Hiii, it’s Kaylee Edmondson and welcome to Looped In, the newsletter exploring demand gen and growth frameworks in B2B SaaS. I write this newsletter every Sunday, and wildly, a few thousand of you read it each week. I’m grateful. If there’s ever anything in particular I can help explore, research, and write about, hit me up! I’d love to chat.

---

Building a Content Production Pipeline

Let me show you what content production looks like when you have proper AI infrastructure versus when you don’t.

Without infrastructure (what most companies are doing):

• Someone manually briefs AI: “Write an email about our new feature”

• AI generates generic copy

• Maybe someone edits it

• It goes out to everyone on a list

• Results are meh 🤷🏻‍♀️

With infrastructure (what actually works):

• Your CRM knows which accounts showed interest in specific pain points

• Your messaging hierarchy tells AI which value props matter to this segment

• Your engagement history shows which content formats this persona responds to

• AI generates variations based on account tier, signals, and past engagement patterns

• Human reviews outputs against brand standards and positioning

• Content goes out with genuine contextual relevance

The difference between these approaches is everything we covered in Parts 1-3. The AI tools are the same. The outputs are completely different because one has proper inputs and the other is just prompting ChatGPT.

Let me break down how this works across different content types:

Email Copy That Doesn’t Sound Like a Robot Wrote It

I worked with a Series B SaaS company that was sending 50,000-ish emails per month. All AI-generated. All terrible. Sidebar: They were going through so many domains they literally couldn’t warm up new ones fast enough. Wild times.

The problem was simple: they were treating AI like a copywriter instead of like an assembly line worker.

Here was the shift to rebuild their email production:

Step 1: Create modular content blocks

Instead of asking AI to write full emails, we built a library of components:

• 12 different opening hooks (based on trigger signals)

• 8 value proposition statements (mapped to their messaging hierarchy)

• 15 proof point variations (customer stories, data points, feature highlights)

• 6 closing CTAs (based on account stage and persona)

Each component was written by their content team, reviewed for brand voice, and approved for use.

Step 2: Build decision trees for assembly

AI doesn’t write emails. AI selects and assembles components based on:

• What signal triggered the email (pricing page visit vs. competitor comparison)

• What persona we’re targeting (technical vs. business buyer)

• What stage of engagement (aware vs. considering vs. evaluating)

• What previous content they’ve engaged with

Step 3: Layer in dynamic personalization

Only after assembly does AI add the personal touches:

• Referencing specific pages they viewed

• Mentioning their tech stack (from enrichment data)

• Calling out recent company milestones (from Clay signals)

• Noting mutual connections or customers in their industry (a combo of Clay enrichment and Crossbeam)

Step 4: Human QA on strategic accounts

Tier 1 accounts get human review before send. AI suggests, humans refine. For Tier 2-3 accounts at scale, AI sends but logs outputs for random spot-checks.

The result: Their email engagement rates went from industry average to 2.3x average. More importantly, sales started telling marketing the leads were better qualified.

The framework here applies to any AI content production:

1. Humans create the components and set the quality bar

2. AI assembles based on structured data

3. Humans QA the strategic outputs

4. System learns from results

Video and Creative That Actually Connects

Text is the easy part. Video is where most teams hit a wall.

A demand gen director told me last quarter: “We want AI to help with video content, but we don’t know where to start.”

Fair question. Video is resource-intensive, which is why most B2B companies produce way less of it than they should. AI can help, but not in the ways most people think.

Here’s what works:

Script development: Feed AI your messaging hierarchy, customer interview transcripts, and winning sales call recordings. Have it draft scripts that follow proven narrative structures. Your video team then adds personality and polish.

Variation creation: You record one video explaining your core value prop. AI tools like Synthesia or HeyGen let you create dozens of variations with different hooks, use cases, or industry examples without re-filming.

One client wanted AI to “generate demo videos automatically.” I watched their sales team’s best demos and found the secret wasn’t the feature walkthrough, it was how the AE adapted the story based on customer questions and reactions.

We couldn’t automate that (at least not yet). But we could use AI to help AEs prep better demos by:

• Analyzing which features matter most to each industry segment

• Surfacing relevant case studies based on the prospect’s tech stack

• Suggesting questions to ask based on intent signals

• Creating personalized follow-up videos recapping the demo

AI became the prep work, not the performance.

The Personalization Stack That Scales

Every marketer wants personalization. Most think personalization means inserting someone’s first name, or trying to make a reference based on the college they attended.

Real personalization is contextual relevance based on everything your infrastructure knows about an account, and the contact.

Here’s how the layers stack up:

Layer 1: Firmographic personalization

Industry-specific messaging, company size considerations, geographic relevance. This is table stakes and should happen automatically based on your CRM data.

Layer 2: Behavioral personalization

Content recommendations based on past engagement, feature callouts based on similar customers, proof points relevant to their use case. Requires your behavioral ICP and intent signal tracking from Part 2.

Layer 3: Temporal personalization

Messaging that matches where they are in the buying journey, urgency that reflects their timeline signals, offers that align with their decision-making stage. This is where account stages and signal recency matter.

Layer 4: Contextual personalization

References to their tech stack, acknowledgment of recent company changes, connection to current market dynamics affecting their industry. This combines enrichment data with external signals.

The magic happens when these layers work together.

When You Should Absolutely Not Use AI

Truthfully, AI makes a lot of content worse, not better.

So, let’s give each other permission to not use AI for:

High-stakes strategic content: Your annual customer summit keynote, your Series B announcement post, your CEO’s thought leadership on LinkedIn. These need authentic voice and strategic positioning that AI can’t replicate.

Crisis communications: When something breaks, your response needs genuine empathy and careful consideration. Not AI-generated templates.

Deeply personal outreach: When reaching out to a champion who’s changing jobs, or following up after a thoughtful customer conversation, write it yourself. People can tell.

Brand-defining content: Your website positioning, your category POV, your differentiation strategy. These require human strategic thinking and market understanding that AI doesn’t have.

Complex negotiations or objection handling: When a deal is on the line and the prospect has specific concerns, you need human judgment and adaptability.

The pattern: use AI for scale and efficiency on operational content. Reserve human creativity and judgment for strategic and relationship-critical moments.

One CMO told me: “We use AI for 70% of our content production, but the 30% we don’t use it for drives 80% of our pipeline.”

The framework to decide: Will this content be more effective if it’s perfect-but-generic or good-enough-but-genuinely-personal? If the answer is the latter, keep AI out of it.

Building Your Content Layer Stack

If you’ve done the work in Parts 1-3, here’s what your customer-facing AI stack might look like:

Content production:

• AI writing assistant (Claude, ChatGPT, Jasper)

• Modular content library (in your CMS or knowledge base)

• Quality control workflows (approval chains in your project management tool)

Personalization engine:

• Dynamic content platform (Mutiny, Marketo, HubSpot)

• Signal-based triggering (from your orchestration layer in Part 3)

• Testing framework (built into your martech stack)

Creative production:

• AI image generation (Midjourney, DALL-E, or Canva AI)

• Design variation tools (Bannerflow, Celtra, or native platform features)

• Brand compliance system (Frontify, Brandfolder with approval workflows)

Conversation tools:

• Email sequencing (Outreach, Salesloft, Apollo)

• Video messaging (Vidyard, Loom)

• Sales intelligence (Gong, Chorus)

Copilot capabilities:

• Research automation (Clay, Apollo, ZoomInfo)

• Conversation intelligence (Gong, Clari, Momentum)

• Customer insights (Gainsight, ChurnZero, Vitally)

Pick tools that integrate with your core system from Part 1. Every tool should feed data back into your source of truth.

—

Next week we close this series with the feedback loop: how to measure what’s working, iterate on what’s not, and scale your AI-GTM engine (hopefully without) breaking it.

We’ll cover the metrics that matter, the testing frameworks that work, and how to enable your team to keep getting better.

See ya next week,
Kaylee ✌