Have You Read What Your AI Is Sending? Why AI Is Confidently Wrong and How to Fix Its Missing Context

Introduction

Kelly Hopping, Chief Marketing Officer at 6sense and a multi-time CMO with prior stops at Demandbase, HYCU, and Gartner Digital Markets, has a blunt take on AI in go-to-market: it is not making most teams more pipeline, it is making them faster at being wrong. Give an AI the wrong signal, or the right signal without the history behind it, and it executes the wrong decision with total confidence. Here is where the context gap shows up, and what marketing leaders can do about it.

n

Confidently Wrong Is Not the Same as Hallucinating

Everyone worries about AI making things up. Hopping argues the bigger risk is AI taking completely accurate data and drawing a reasonable-sounding conclusion that is wrong. An executive changes jobs, and the AI flags her as a hot new prospect. In reality, she is a former customer who churned unhappily. An account surges in intent, but it sits in an industry you do not sell to. As Hopping puts it, "The data isn't wrong, it's just the interpretation of how to use it is." Context is what tells the machine what a signal actually means: identity, relationship history, and business rules.

n

More Activity Does Not Equal More Pipeline

Every CMO is getting the same board questions every couple of weeks: where is more pipeline, and how are you using AI to cut costs? That pressure pushes teams to automate outbound, enrichment, scheduling, and event follow-up. Hopping's honest read is that almost no company outside the AI vendors themselves can say AI has increased pipeline. Activity is up, volume is up, and the needle is not moving, because the work has become about speed instead of accuracy. My takeaway: if AI is generating more sends but not more engagement, you are treading water. The measure of email was never volume. It was always the response it earns.

n

A Signal Is Not Intent

Website traffic comes from buyers, but also from competitors, researchers, employees, and job seekers. One signal ca ot tell you motivation. Intent comes from reconciling many signals into a picture of why the activity is happening, which is hard, and which is why feeding a flat contact file into an LLM and expecting intent is going to miss. The bigger shift is that buying research has moved off your website entirely. 6sense's own buyer journey research found that more than 90 percent of buyers build their vendor shortlist before engaging with any vendor, and the first name on that list wins over 80 percent of the time. That means owned content, press releases, blogs, and podcasts are no longer nice-to-haves. They are how you get represented in the LLMs where the shortlist gets built.

n

Build the Context Layer Before You Buy More Data

Customers tell 6sense all the time that they just need to throw more data into the lake and figure it out later. Hopping's counter is that aggregating data is not the same as reconciling it. What is missing in most stacks is a context engine that sits between data, intelligence, and activation, so outreach can say "I know you changed companies, and I know you were a customer advisory board member at your last one" instead of introducing your company cold to someone who has worked with you for ten years. That one bad email gets marked as spam, and the relationship is dead. Agents need the same thing. In my experience, it took a year of training my own AI chief of staff on what the numbers mean and why emails are written a certain way before it got good. Hopping is doing the same at 6sense, adding governance so Claude-generated decks stay on brand and on the latest messaging version, not version three.

n

When Every Dashboard Is Green and Nothing Works

Clicks and registrations are still observable, but they are no longer business impact on their own. Hopping's fix is multi-touch attribution with a six-month lookback that includes LLM referrals, SEO, review sites, and organic social. The result was telling: 6sense's top four pipeline drivers are organic search including LLMs, organic social, owned content, and top-of-fu el paid display. All four are brand assets, not bottom-of-fu el capture. Last-touch models had been crediting gifting for meetings people took for the tacos. Now gifting is reserved for deal acceleration, moving a stalled stage three to stage four. She is also replacing the four-touchpoint demo gauntlet with self-serve product tours, and measuring sales cycle length, average selling price, and win rate instead of click-throughs.

n

Conclusion

The pattern is consistent. AI executes whatever you feed it, so the leverage is in context, not volume. Audit what your automation is actually sending. Stop treating individual signals as intent. Invest in the owned media that shapes shortlists you ca ot track. Build a context layer and train your agents like you would a new hire. Then measure the lagging indicators that prove any of it worked. As always, my advice is to just focus on keeping your customers happy.