THE STORY
I'd been running the workflow for three months without a problem. Drafts came back at maybe 60% quality, I'd edit them up to 90, ship. The workflow knew the brand voice well enough.
When I went back through the last six weeks of drafts looking for similar patterns, I found three more fabricated claims. One invented a competitor pricing tier that didn't exist. One credited a strategy decision to a team member who'd never worked on it. One cited a stat from "a 2025 industry report" that traced to nothing.
Four fabricated claims in six weeks, in a workflow I trusted. The model wasn't doing anything wrong. It was filling context gaps with plausible-sounding content. The gaps were mine to close.
Here's what bothered me. Every workflow I had running was vulnerable to the same gap. Every team running AI workflows without a grounded source-of-truth doc is operating with the same blind spot. The fix isn't more editing. It's less context drift.
Why this is going to bite a lot of teams in 2026
AI fabrication is going to become the editing tax marketing teams pay for the next 24 months. Two reasons.
First, AI workflow adoption is accelerating faster than guardrail adoption.
Marketing teams ship one workflow, see it save five hours a week, and ship four more before the first one's outputs have been audited. The compounding cost of fabricated claims across five workflows is invisible until something breaks publicly.
Second, the breaks are quiet.
A fabricated customer quote in a blog post might never get caught. A made-up stat in a comparison page can rank for years. A misattributed strategy decision in a sales deck can survive an entire account expansion cycle. None of these trip an alarm. They erode credibility slowly, in a way that doesn't show up in any dashboard.
The single document that prevents most of this (the master marketing document) is the editing-tax fix. Not because it changes the model. Because it changes the context the model receives.
The 7-section framework
Seven sections. Version 1 is fine. Start here.
Brand identity: Positioning, ICP with named examples, anti-positioning.
Voice + tone: Voice characteristics, register per surface, banned phrases and AI-tells.
Product or service truth: What it actually does, pricing, common misconceptions.
Evidence library: Real customer quotes with attribution, data points with sources and dates, named case studies, stats from 2025 or later.
Competitive context: Brands buyers compare against, where competitors win honestly.
Customer language samples: Pain points in buyer words, qualification call questions, the gap between team jargon and buyer phrasing.
Channels + cadence: Where you publish, how often, who owns each surface.
Each section answers a question every downstream workflow will ask. Skip any section and that's the section your AI workflow will fabricate from.
The build is faster than people expect. Pull up your homepage, three case studies, your last five sales call transcripts. Work through the seven sections. Trim what isn't true. Ship the v1. The polishing happens in week two, and once it's live, you load it into every Claude project, every employee and contractor's onboarding email. The editing tax starts to compress immediately.
Deeper reads
Three posts on propsaasgrowth.com/blog that go deeper on parts of this issue:
I ran my AEO framework on my own brand. I was invisible. Five archetype prompts, two AI engines, zero brand presence. The dogfood diagnostic that catches gaps in your own infrastructure before AI engines do.
How to write 'X vs Y' pages that AI engines actually cite The evidence library section of the master doc, applied. Honest comparison content wins citations; biased "we win every row" pages get skipped.
Buyer intent keywords: how to find the queries that actually drive SaaS revenue Where the "customer language samples" section of the master doc actually comes from — sales calls, AI engine queries, review sites.
What's coming next
Cadence: The Compound is monthly. Issue #3 lands mid-July.
Next issue: The workflows that ground in the doc.
Once your master marketing document is live, the next question is where to plug it in. I'll walk through the five Claude project setups every marketing team should run first: brief generator, draft drafter, comparison page builder, refresh queue, and citation tracker. Each one grounded in the master doc, each one cutting hours off a recurring task.
If you'd rather skip the doc-building part entirely
AirOps already bakes this structure in. Brand kit handles voice + tone + positioning. Knowledge base handles evidence library, customer language, and named case studies. Every workflow you build on top of it grounds in those two surfaces automatically, no doc-engineering required.
I have clients shipping at 80% draft quality in their first week on AirOps. The reason is structural: the brand kit + KB does the same job the master marketing document does, and it does it inside the tool where the workflows already live. If you're going to run AI workflows for marketing in 2026, AirOps is the fastest path I know from "we should ground our AI" to "drafts come back ready to ship."
Try AirOps with my affiliate link - Small kickback for me ;) Get the brand kit and KB populated this week and you'll have the master doc structure live in two surfaces by next.
Until next time,
Gemma