The intent data black hole: why your signal has gone quiet

Intent data has always relied on one thing: buyers leave a trail. Publishers saw it, intent providers packaged it and vendors paid for it. In the traditional sense, a large part of that trail has gone dark for all three, because buyers now get their answers directly inside AI tools.

I call this the intent data black hole - here's why:

I've spent around 15 years selling B2B intent data and building intent products, at Spiceworks, Dennis and Autovia, and for clients since. Despite what some may think, a name and a job title aren't intent, they're a contact record. Real intent is an account moving: several people from one buying group, reading decision-stage topics, more often than last month. Every part of that is now harder to see.

For most of those 15 years, intent largely came from content. Declared data existed, but it never scaled the way reading did. You baselined what each company read on each micro-topic and watched for spikes. The most valuable spikes came from evaluative reading: comparisons, buyer's guides and shortlists, read on publishers' sites, while vendors watched the same accounts turn up on their own comparison and pricing pages.

That reading has moved. According to G2, 51% of B2B software buyers start their research with an AI chatbot more often than with Google, and Responsive found that 80% of technology buyers use generative AI at least as much as search when researching vendors. The comparing and checking increasingly happens inside the answer, where no publisher, intent provider or vendor can see it. The providers say as much themselves: 6sense concedes that a ChatGPT answer sends no signal to any vendor it names, and Bombora warns that intent built on one or a few sources will struggle as traffic gets bottlenecked at AI summaries. That's the first property of the black hole: things go in and don't come out.

ChatGPT shows some of its working. In 36 of 59 answers we tracked, its very first search already named vendors. The candidates come from what the model already knows: everything it absorbed in training, long before anyone asked the question. Nobody outside the AI companies can see inside that, which means nobody can tell which of your pages, articles or reviews put a name there. What we can see is the check.

Across two sectors in early October, ChatGPT ran 310 searches while answering buying questions. Not one named a trade or tech media title. More than half its searches went to check its candidates on the vendors' own websites.

That's the second property of the black hole: you only see it by its effect. If you're a publisher, you barely feature in the model's research. If you're a vendor, it's your own site that gets checked. Either way, what the model searches for and cites is the only part of the research inside the answer that anyone outside can see. It's a category signal: it tells you what the model checks, but never who is checking or why.

This is the third property, and the one that costs most. What moves inside the AI answer is the comparing and checking, the stage closest to the B2B buying decision. For a publisher, that's the reading you charge most for: a reader comparing suppliers is worth more to the vendors paying for your audience than one browsing the news. For a vendor, it's the most valuable signal there is: an account weighing you against its alternatives.

There's a fair counter-argument. Bombora's co-op shows specialty B2B publishers' traffic holding up, and argues buyers still head to destination sites nearer the decision. But a steady page count doesn't show where the comparing happened. When the model does it, it checks the vendors' pages, not the publishers'.

The pull isn't total. The point where the buyer reliably steps back out is in person: the event, the meeting, the reference call. The trail comes back into view there, and it's worth more than it used to be: the buying group in the room beats any page view, and for sales teams it's often the first point at which they can see who is really in the deal.

For B2B media and events, the exposure is direct and, for some, existential. For vendors, it hits the account targeting a lot of pipeline was built on.

The reading-based signals go first: surges on comparison topics, co-op and page-view intent on buyer's guides, and the gated downloads that depended on that traffic.

What survives is what AI can't do on the buyer's behalf. An AI tool can read, compare and shortlist for a buyer. It can't fill in a registration form with their budget, walk into an event or a meeting for them, or pull their colleagues into the decision. Those things still need the buyer, which is why they're still visible. So the signals that last are the ones the buyer gives you directly. Declared data, where they state their project and budget on a registration, event sign-up, survey or RFP. Attendance, when people from the buying group turn up at an event, a demo or a meeting, so you can see who came and what they asked about. And buying-group depth - several people from one account engaging at once - which was always the real signal anyway, and is counted at the registration desk and in the room.

Here's the side change. Intent used to come from what buyers read. More and more of it now comes from what the AI model checks on their behalf, and its searches and citations tell you about demand in a category, even if they can't tell you which buyer asked. You can't target an account with it, because the model does the fetching and the buyer rarely shows up in anyone's logs, but you can measure it.

The test is the one I've always used: does it turn into pipeline in the customer's CRM?

A complete view of intent has two halves. One is the signals buyers still give you directly. The other is a measure of what happens inside the AI answer: whether you're named, whether you're cited, and what the model checks before it decides. Neither half works on its own.

If you sell intent data, package what survives - declared data, attendance and buying-group depth - and add the half you can't currently see: whether your titles are named or cited when buyers in your sector ask AI to compare suppliers. The Source Layer measures exactly that, and it tells you whether you're still part of the comparison your customers are paying to reach.

If you buy intent data, run the CRM test, and ask your provider how much of the signal comes from evaluative reading on third-party sites, because that's the part thinning fastest. Then measure the other half: whether AI includes, positions and prefers you when buyers compare, and what it finds when it checks your own pages. The Selection Layer is where that happens, and where the shortlist now gets built.

The trail hasn't gone. Part of it has crossed into the black hole, and the part outside still needs the buyer to show up. Measure both, or you're making decisions on half the picture.

This is where the exposure sits: when an AI model researches a purchase on a buyer's behalf, it goes to vendors, analysts and standards bodies, and trade media isn't in the trail. Yet most of the AI money I see B2B media boards approve goes on tools and internal adoption, doing today's work faster or protecting content from being copied. Very little goes on finding out where buyers now research, or whether its media brands are still part of it. That exposure is what anyone buying or backing a B2B media business will price. More on this in a future edition.

The question for the board: What share of our sector's buyer research now happens where we can't see it, and is any of our AI budget measuring it?

If you market a vendor and want to know what AI checks before it puts you on a buyer's shortlist, and what it finds when it does, that's what the AI Shortlist Scan measures.

Edition One of the Selection Index publishes on Monday 16 November. It lands in your inbox that morning.

John
Decoding Growth