What marketer hasn’t agonized over the perfect adjective, the right message, the quintessential tagline? When you think you finally nailed the positioning and messaging, you can’t wait to get the campaign launched and see the leads roll in.
You’re wasting your time.
Or at least, you’re wasting your time if you don’t think about how that message will get summarized, distilled, reduced, and distorted a hundred different ways before your audience sees it. You’re in a game of telephone — but instead of a chain of people between you and the ultimate recipient, it’s a bunch of machines and AI agents.
I’ve been manually testing this on a small scale with a handful of brands and products, and the results are a little unsettling. In today’s enterprise buying process, your carefully crafted story is increasingly getting scraped, compressed, categorized, and scored by machines long before a decision-maker ever sees it.
Procurement portals normalize ROI claims. Analyst copilots digest your deck into three-line notes. Buyers lean on copilots to compare vendors and prep internal briefs. In many cases, the version of your story your buyer sees isn’t yours — it’s a machine’s.
Try It Yourself
In my early experiments, I’ve taken messaging from a few vendors, fed their decks, datasheets, and websites into multiple AI models, and prompted them the way a buyer would:
“You’re a CFO at a Fortune 500 company. Summarize the top vendors in this category and explain why they are contenders.”
Then I compared what the models returned to what the vendors actually claimed about themselves.
Here’s what’s happening:
- Unique features often get bucketed into generic categories
- ROI stories get rounded to safe averages
- Claims drop out entirely, especially ones buried deep in supporting material
- In some cases, less than half of the vendor’s core differentiators survive intact
This isn’t just anecdotal. Peer-reviewed summarization research shows that 30% of key claims are typically dropped or altered when machines compress content, and in short-form summaries, 70% include distorted information as a result of aggregation.
The takeaway: your buyers are hearing a version of your story — but it may not be the one you intended.
What Happens
Enterprise decisions hinge on trust, proof, and nuance — but machines aren’t great at nuance. This is not to say AI is bad; it just doesn’t care about preserving your message when trying to address a query. Its intent is to compress and normalize aggressively:
- A six-week time-to-value claim might turn into “typical ROI: six to twelve months”
- A specialized workflow engine becomes an “automation platform” alongside five competitors you’ve out-innovated for years
- If your category touches compliance or security, copilots often overweight risk signals while underweighting mitigation
The result: vendors start to sound the same in internal summaries, which collapses differentiation and drives price pressure. When buyers can’t hear the sharp edges of your value, you become “Vendor Generic.”
Measuring What Survives
I’ve been experimenting with a simple way to quantify this: a Signal-to-Whisper Ratio (SWR).
Pick your top five positioning claims — the things you’d fight to keep intact at all costs. Feed your content into multiple AI copilots using buyer-style prompts. Score how many claims show up accurately on the other side.
It’s not a perfect measure yet, but even small tests show huge variance. I’ve seen SWRs as high as 70% — and as low as 30%, where most differentiation is lost before a human sees the message.
Where This Is Headed
This isn’t just about AI summarization. It’s about a shift in the sequence of enterprise buying:
- Machines are the first audience for your story
- Humans make the decision later
- The risk lives in the handoff between the two
The goal is to design for that handoff — strategies to enable your brand to survive M2M (machine-to-machine) compression and get you to H2H (human-to-human) conversation.
Because if the machines can’t carry your story forward faithfully, the humans may never hear it at all.
