Intent data is collected five main ways: first-party website tracking, bidstream ad exhaust, publisher co-ops, review-site signals, and public-web change detection. Vendors happily tell you WHAT their intent data predicts. Very few tell you where it comes from. I learned that in 2019, at my first SaaS marketing job in Hamburg. I bought a three-month intent-topic trial. Then I sequenced the 412 accounts the dashboard marked as “surging.” I booked 3 meetings. And when I asked where the surge data came from? A diagram with a cloud labeled “data network.” So let’s open the box.
📌 TL;DR: Intent data comes from five sources. First-party: your own site analytics. Bidstream: ad-auction exhaust. Publisher co-ops: pooled reader behavior from B2B sites. Review-site signals: category research on G2-style platforms. Change detection: public company footprints diffed for real events. The verdict: the closer the data sits to something you can verify, the more you can trust it.
How Is Intent Data Collected? The Five Sources at a Glance
Intent data gets collected from five sources: your site, ad auctions, publisher networks, review platforms, and the public web. The mix is the product. Two vendors can both say “we detect buying intent” while gathering totally different raw material. And totally different consent stories.
Quick definition first. Intent data is behavioral evidence that a company or person may be moving toward a purchase. That’s the what. This page is the how, because collection method decides accuracy, legality, and price. For the type-by-type map, I’ve broken down the types of intent data separately.
| Source | What is actually captured | Who owns the raw data | Consent posture |
|---|---|---|---|
| First-party tracking | Page views, form fills, product usage on YOUR properties | You | Your own consent banner and privacy policy |
| Bidstream | Page URL, IP, device details from ad-auction bid requests | Ad exchanges and auction participants | Contested; regulators keep challenging it |
| Publisher co-ops | Reader behavior across a network of cooperating B2B sites | The co-op vendor, via member sites | Gathered at member-site level |
| Review-site signals | Category and comparison research on review platforms | The review platform | Logged-in users, platform terms |
| Public-web change detection | Observable changes to company pages, jobs, funding records | Public information, snapshotted | Public-footprint data, no tracking of readers |
And the flavor barely changes by team. How is intent data collected in B2B sales? These five sources, packaged as account alerts. In marketing? Same sources, packaged as audience segments. The plumbing doesn’t change.
First-Party Intent: The Data You Collect Yourself
First-party intent data comes from your own website analytics, forms, and product usage, captured with your own scripts and consent banner. Someone visits your pricing page three times. Someone’s trial usage spikes. You saw it happen on property you own.
The mechanics are simple: a script records visits against a cookie or a logged-in identity. Form fills turn anonymous visitors into named people. And reverse-IP matching maps anonymous traffic to a company. So “someone at a 200-person logistics firm read the API docs” becomes usable without a name.
Where it wins: trust. The consent chain is yours, and nobody can sell your competitors the same signal. The honest limit? Reach. First-party data only sees buyers who already found YOU. That gap is why the third-party industry exists, and why the first-party vs third-party intent data split matters.
What Is Bidstream Intent Data?
Bidstream intent data is the exhaust of programmatic ad auctions, harvested from billions of bid requests. Here’s the mechanic. When a page with ad slots loads, a bid request goes out through real-time bidding. That request broadcasts the page URL, the reader’s IP, and device details to hundreds of bidders in milliseconds. The protocol behind that broadcast is OpenRTB, documented openly by the IAB Tech Lab.
Now the part vendors say quietly. You don’t need to WIN the auction to see the request. Some participants keep that exhaust and resell it. Map the IP to a company and the URL to a topic. Suddenly “Acme is reading about payroll software” is a sellable signal. That’s bidstream: enormous, cheap, and collected from people who never knowingly opted in.
So let’s be honest about the criticisms, because they’re substantial:
- Consent. The UK’s Information Commissioner’s Office has publicly questioned whether RTB’s data sharing is lawful at its current scale. Its online-tracking work keeps adtech under active scrutiny. And in 2022 the Belgian Data Protection Authority ruled that the IAB’s consent framework infringed the GDPR. That framework is the pop-up machinery behind much of this.
- Accuracy. Topic inference from a URL is crude. Office IPs are shared, VPNs lie, and bot traffic pollutes everything. The IAPP’s look at where RTB fits in the risk landscape is a sober read.
- Disclosure. Many intent vendors simply don’t say which share of their signal is bidstream-derived. Bombora, to its credit, publicly positions its co-op AGAINST bidstream collection. That alone tells you how contested the practice is inside the industry.
🔍 Did You Know? In 2022 the Belgian Data Protection Authority ruled that the IAB's Transparency and Consent Framework, the consent pop-up system behind much of programmatic advertising, infringed the GDPR. A big slice of bidstream collection ran through exactly that machinery.
Balance, though. Where bidstream wins: breadth. Nothing else sees reading behavior across the open web at that scale, for that price. If you understand what you’re buying (explainers like Publift’s lay the mechanics out well), it can still earn a place. Think wide, noisy early-warning layer.
How Do Publisher Co-ops Collect Intent Data?
A publisher co-op pools reader-behavior data from a network of B2B publisher sites that agree to share it. Member sites carry the co-op’s tag, and reading events flow into one pool. The vendor baselines each company’s normal reading level per topic and flags surges above it. Hence the name “topic surge” score.
Bombora’s Company Surge is the canonical example. Per its public materials, thousands of cooperating B2B sites feed the pool. Consent is gathered at the member-site level, not scraped from auctions. A genuinely cleaner story than bidstream. It’s also the de facto standard: when another vendor “includes intent,” a co-op feed is often under the hood.
But co-ops have honest limits too. The data is company-level, not person-level, so you know Acme is reading, not who at Acme. Topic taxonomies are broad. And you can’t audit which sites are in the pool. A co-op is only as good as a membership list you will never see.
Review-Site and Search Signals: Intent You Buy Downstream
Review platforms like G2, TrustRadius, and Capterra sell intent signals collected from buyers researching categories on their own properties. Someone opened your category, read your profile, compared you against a competitor. The platform saw it all, first-party for THEM. Then it sells the signal downstream as “this account is evaluating.”
The same downstream logic covers two cousins. Content syndication: a sponsor places a whitepaper, and every download is declared, form-filled intent. The catch is follow-up fatigue, because that reader gets called by everyone. And search signals: some vendors run content networks or license search data to catch category queries.
Honest read: per signal, this is the highest-intent third-party data you can buy; comparison research is nearly a shortlist. The limits are narrowness and timing. The platform only sees its own visitors. And by the time an account compares vendors on a review site, the deal is late-stage and crowded.
How Does ZoomInfo Get Intent Data?
ZoomInfo says its intent engine analyzes B2B content consumption across a large, sensor-style data network. Per its public materials, the mix includes network signals, data partnerships, and its own properties, aggregated into company-level topic scores.
What can you verify from outside? Honestly, the output more than the pipeline. Industry analyses of third-party intent, like the IAPP’s RTB work above, place auction-derived signals in many large vendors’ mixes. Most publish the sensor metaphor, not a source-by-source breakdown. That’s not an accusation. It’s a purchasing reality: the burden of asking “collected where?” sits with you.
How Does 6sense Get Intent Data?
6sense combines your first-party signals with its proprietary intent network and licensed third-party sources. Per its public materials, a predictive AI layer fuses those streams into account scores and buying-stage predictions. That prediction layer is the product.
And here’s the criticism buyers voice, the reason this question fills Google’s People Also Ask box: the model is a black box. You see the score. You rarely see the evidence behind it for one specific account. Black-box predictions can still be useful, but only when you test them against your own closed-won history.
Firmographic-Change Detection: Signals Collected From the Public Web
Change detection crawls public company footprints and emits a signal when two snapshots differ. No reader tracking, no auctions. Just observable events: a funding round appeared, a VP of Sales joined, twelve jobs opened. This is the method we build on at CUFinder. So I can show you the actual pipeline instead of a cloud diagram.
Five steps, end to end:
- Crawl and snapshot company pages, job postings, employee profiles, and funding records on a recurring schedule.
- Diff each new crawl against the previous one, field by field: employee count, job count, name, description, locations, funding rounds.
- Evaluate a deterministic trigger condition for each signal. No vibes, a rule.
- Sort quantitative changes into magnitude buckets: low is a 1-5% change, moderate 5-15%, high 15-30%, hyper 30% or more.
- Store metadata with the event (the function hired for, the country expanded into). The signal arrives ready to act on.
The triggers come in three types, with real thresholds. Single-change triggers fire on any tracked field change, like employee growth at plus one. Baseline triggers compare against a rolling average. A followers spike fires at 3x the previous six-crawl average. A jobs-open spike fires at 2x the previous four-crawl average. A layoff signal fires at a 10%-plus single-interval drop. And windowed triggers watch patterns over time. Office consolidation waits for three closures in 90 days. An acquisition signal needs a new parent plus a name or description change within 60 days.
📌 Example: A company had an average of 5 open roles across its last four crawls. Today's crawl finds 12. That is 2.4x the baseline. The jobs-open spike signal fires at high magnitude the same week, before any press release.
Now the honest scope paragraph, because our method has limits too. A change signal is an observable event, not inferred attention. It tells you budget or urgency appeared. It does NOT tell you someone read three comparison articles yesterday. And it can lag the event by a crawl cycle, since profiles updated late get detected late. The two are complements, not substitutes. I’ve written up buying signals vs intent data if you want that argument in depth: signals are events, intent is attention.
“A static lead list tells you who a company is. It never tells you when to call them.”
CUFinder buying signals documentation
In practice, CUFinder tracks 99 signal types across 10 categories this way, refreshed daily against 1B+ people profiles and 85M+ company profiles, and every trigger is published. So every buying signal traces back to a rule you can read. The categories span growth, decline, funding, and people moves. They sit on top of the same firmographic data a static list gives you, plus the one thing a list never has: timing. It’s the collection layer under all our B2B buying signals work.
Using it takes five clicks more than reading about it:
- Open the Buying Signals filter inside the Prospect Engine on CUFinder’s buying signals page.
- Stack the signals that match your motion, say funding rounds plus sales-leader hires.
- Set a minimum strength, low to hyper.
- Pick a time frame, 1 week back to 6 months.
- Apply. Matching companies and contacts update live, with verified emails and phones attached.
And if you’re the engineer in the room, the same events are one POST away through the Company Signals API. Send a signal name, a time frame, and a bucket to /v2/csa. You get back the companies where that signal fired.
Which Intent Data Collection Method Should You Trust?
Trust rises with verifiability: weight first-party behavior and observable public events highest, inferred third-party attention lowest. First-party is what happened on your property. Change detection is what verifiably happened in the world. Co-ops are what an unauditable network noticed. Bidstream is what leaked out of an auction.
In practice, match the method to your motion. Then interrogate the vendor. Three questions do the work:
- What share of your signals is bidstream-derived? A serious vendor answers in one sentence.
- What is your consent basis under the GDPR? Regulators don’t care how useful the surge was.
- Can I see the evidence behind one score for one account? If no, you’re buying my 2019 cloud diagram.
💡 Pro Tip: One question kills more bad intent-data deals than any feature comparison: "Where was this collected?" I sequenced 412 surging accounts in 2019 and booked 3 meetings because I never asked it. Ask it first.
Two housekeeping notes. First, cookie-based collection did not die. Google announced it is keeping third-party cookies in Chrome after years of deprecation plans. So the cookie pools behind many intent products live on, even as regulators tighten. Second, collection is only half the job. Turning the data into pipeline is its own craft, and I’ve covered using intent data for sales tactic by tactic. There are also honest free intent data sources: job boards, funding announcements, tech lookups. Fine manual versions while budget is tight. That’s the heart of signal-based selling: pick verifiable inputs, then act fast.
FAQ: How Intent Data Is Collected
How does ZoomInfo get intent data?
ZoomInfo says its intent comes from a B2B data network analyzing content consumption at scale. Per its public materials: network signals, partnerships, and its own properties, aggregated into company-level topic surges. Like most large vendors, it publishes the model, not a source breakdown, so ask about the mix.
How does 6sense get intent data?
6sense blends your first-party data with its proprietary intent network and licensed third-party sources. A predictive layer turns those streams into account scores and buying-stage labels. The caveat: it is a black box, so validate predictions against your own closed-won accounts.
What is bidstream intent data?
Bidstream intent data is reading behavior harvested from programmatic ad-auction bid requests. Each request leaks a page URL, IP, and device details to hundreds of bidders, and some participants resell that exhaust. Broad and cheap, but regulators like the UK ICO keep challenging its consent basis.
How is intent data collected in B2B sales?
B2B intent data is collected from five sources: your own site, bidstream auctions, publisher co-ops, review platforms, and public-web change detection. Sales teams mostly buy the last three as account alerts, because those arrive with company names attached.
What does intent data mean?
Intent data means behavioral evidence that a company or person is moving toward a purchase. First-party intent is behavior you observe on your own properties. Third-party intent is behavior someone else observed and sold to you, which is why collection method matters.
Who offers the best intent data?
Nobody wins every use case; the best source depends on the collection method you need. Co-ops lead on topic surges, review platforms on bottom-funnel comparisons, change-signal platforms on verifiable timing. I keep an honest comparison of buyer intent data providers for shortlisting.
It’s Time to Ask Where Your Intent Data Comes From
You now know more about intent-data collection than most people selling it. Seriously. Five sources, one sorting question: “Where was this collected?”
A colleague in Hamburg once told me, “data you can’t trace is rumor with a dashboard.” I’ve bought the rumor. Trace your data, weight the verifiable stuff, and let the rest be a tiebreaker.
Which of the five sources surprised you most? Tell me in the comments. And if a vendor ever answers with a cloud diagram, you know what to do.
Bad
Okay
Good
Amazing
