The key points:
- Most PMMs don’t distrust AI because they’re anti-AI. They distrust outputs that lack traceability, evidence, and grounding in real buyer reality.
- The core problem in many AI workflows is not generation quality — it’s trust architecture. PMMs are trained to question unsupported claims, weak methodologies, and polished outputs that can’t be verified.
- “AI slop” is often less about factual inaccuracy and more about the absence of conviction, specificity, contextual thinking, and real strategic judgment.
- Citation and provenance fundamentally change adoption. When AI outputs are tied to buyer interviews, call transcripts, win/loss data, and validated research, users stop asking “Can I trust this?” and start asking “Is this useful in this context?”
- The next phase of AI in GTM will be won by systems that improve confidence and decision-making, not just systems that generate content faster.
The Real Problem Starts After the Output Appears
There’s a moment every product marketer (myself included) has had with AI by now.
You paste in a transcript, maybe a pile of calls from something like Gong, CRM data, or some customer interviews, and thirty seconds later the machine hands back something that looks surprisingly complete. The formatting is clean. The summaries sound intelligent. The recommendations seem thoughtful enough at first glance.
But then a different reaction kicks in: “Wait… where did this actually come from? And what happens when someone starts asking questions about it?”
That hesitation is important because it reveals something a lot of AI companies still misunderstand about adoption of AI tech inside GTM teams: Most PMMs are not anti-AI. In fact, many product marketers were some of the earliest adopters of these tools because the surface-level value proposition is obvious: Faster synthesis. Faster drafts. Faster analysis. Faster execution.
What PMM wouldn’t love that? When I first tried ChatGPT at the urging of a former boss, I was positive I had just stumbled upon magic.
The problem is not that the outputs exist, it’s what happens after the output is generated — when a human has to decide whether they trust it enough to act on it.
Most AI Failures in GTM Are Actually Trust Failures
Many AI failures inside marketing organizations are not failures of models, but failures of trust architecture. People don’t act on information they don’t trust, even when the information itself is technically correct.
Behavioral Psychology has proven this for years. People judge credibility before usefulness. If something feels unsupported, unclear, or disconnected from reality, they hesitate. They double-check it themselves. Or they stop using the system entirely after one bad experience.
One of the more interesting things I learned as an organizational psychologist was how closely trust impacts adoption. Research in areas like the Technology Acceptance Model and source credibility theory consistently show that people are far less likely to adopt systems they perceive as opaque, even when those systems produce objectively useful outputs.
Usefulness alone doesn't drive behavior. People also need confidence in the reliability and interpretability of the system itself. If users can’t understand how conclusions were formed or verify the underlying evidence, skepticism increases and adoption slows.
That framework is relevant to the current state of AI in GTM. PMMs are being asked to operationalize outputs that often appear polished and authoritative without enough visibility into the evidence chain behind them. In high-stakes strategic environments, that lack of interpretability becomes a trust problem long before it becomes a capability problem.
PMMs are particularly sensitive to this because they’re trained evaluators of information quality. Experienced PMMs can immediately spot cherry-picked data, weak methodologies, biased research, and inflated claims disguised as “market insight.” These days they recognize that most so-called “buyer insight” tools are leaning heavily on aggregating public internet language and repackaging it as useful strategic intelligence.
So when an AI system confidently produces recommendations without showing the reasoning or evidence behind them, skepticism is a fair response.
Why “AI Slop” Feels Wrong Even When It’s Technically Correct
That’s part of why the phrase “AI slop” has spread so quickly across marketing teams. What’s interesting is that people often use the term even when the content itself isn’t factually wrong.
AI output feels flattened, averaged, and opinionless. It sounds like something engineered to offend nobody rather than persuade somebody. There’s a kind of eeriness to it. In Mark Fisher’s 2017 Essay Collection The Weird and the Eerie, Fisher describes “the eerie” as the sensation produced by “a failure of absence or by a failure of presence.” That framing feels relevant to modern AI output. The content might contain all the structural components you expect, but still feel like something important is missing beneath the surface.
Experienced PMMs and their colleagues can typically sense AI-generated content before they can formally disprove it because the output lacks conviction and anything interesting. There’s too much balance, too many caveats, and too much “on one hand, on the other hand.” There’s no actual stance, real tension, or evidence that someone made a deliberate judgment call.
And strategy requires judgment, not just synthesis.
The “Looks Finished” Problem
The problem gets more complicated because AI-generated work already looks finished. It comes packaged with clean formatting, summaries, recommendations, bullets, and executive-ready language. The output creates the impression that the thinking has already been done.
And in most organizations, that appearance matters more than people want to admit.
PMMs are now being asked to produce more than ever before. AI has dramatically increased the expected pace of output. But the pressure to produce often moves faster than the pressure to deeply understand whether the work is actually grounded in real buyer behavior or meaningful strategic insight.
That creates a dangerous dynamic. Because the outputs look polished, teams can start mistaking production for progress. The workflow quietly becomes:
Generate. Refine. Ship. Repeat.
And once that happens, it becomes much harder to stop and ask uncomfortable questions like:
- Is this actually true?
- Is this differentiated?
- Would a real buyer care about this?
- Did this come from genuine insight?
The people who feel skeptical about the output are often the same people expected to move quickly and keep content flowing, and the problem is usually difficult to prove in the moment. Nothing is obviously broken, but the content just feels generic, shallow, or disconnected from reality in a way that’s hard to articulate in fast-moving organizations focused on speed and scale.
The PMMs who value the quality of the output may compensate quietly. They manually validate claims. They rerun prompts. They reach out to customers asking the same questions directly because they don’t fully trust the interpretation layer sitting between them and the market, and they hope learning from existing customers will act as an appropriate proxy to understand future buyers.
In many cases, the AI didn’t remove cognitive effort. Instead it shifted it onto the people responsible for making strategic decisions while simultaneously increasing the pressure to produce more artifacts than ever before.
Evidence and Citation is the Product.
Citation is becoming far more important than most AI vendors initially expected. Citation isn’t just a feature enhancement or UX improvement. In many cases, citation is the product. Once outputs become traceable to real buyer interviews, win/loss calls, call recorder transcripts, or some other form of high-quality validated research, the user’s mental posture changes. The question shifts from “Can I trust this?” to “Is this insight useful in this situation?” That’s a fundamentally different thought process.
When people can see where information originated, they stop evaluating whether the AI is lying to them and start evaluating evidence the way strategists and researchers naturally do.
Humans calibrate trust through traceability. We trust direct buyer language, contextual evidence, primary research, and visible sourcing. We distrust unsupported summaries, black-box recommendations, and decontextualized aggregation. Yet most current AI tooling still behaves as though confidence and polish alone should be persuasive.
High-Stakes GTM Work Requires Verifiable Thinking
But confidence without evidence creates fragility, especially in high-stakes GTM environments where there is a high cost of getting messaging wrong. A PMM can survive moving a little slower, but often can’t survive confidently shipping the wrong narrative into the market. That’s why so many AI workflows lose momentum after the initial excitement wears off. Initially the workflow feels impressive, but the day-to-day reality of validating and defending the outputs creates friction that teams eventually abandon. The generation feels magical, but the verification process becomes exhausting. Eventually, the effort required to verify the output cancels out the speed benefit of generating it.
The companies that ultimately win AI adoption inside GTM organizations may not be the ones building the most impressive generation systems. They may be the ones building the best confidence calibration systems. These systems are designed around interpretability, evidence and source visibility, contextual grounding, and decision support instead of simply producing outputs faster.
The next phase of AI in GTM is not about making models sound smarter, but about designing systems that help humans make impactful decisions with confidence. Product marketers already understand the cost of shallow research and false certainty. AI tools that want to win the game of long-term adoption need to respect that reality instead of assuming polished outputs automatically deserve trust.