Short answer
Programmatic advertising buys impressions across thousands of websites in automated auctions. It targets people by inferred signals: IP addresses, cookies, device IDs and third-party intent data. Conversational Intent Advertising (CIA) shows one ad under an AI answer, matched to the problem a buyer has just described. Programmatic gives you cheap reach on a list. CIA costs more per impression but appears at the decision. Use programmatic to stay known and CIA to be chosen.
How programmatic works in B2B
A demand-side platform (DSP) bids in real time for ad space on websites and apps. In B2B, it usually decides who to bid on in one of three ways:
- IP-to-account matching. Company IP ranges are matched to your account list. This is the basis of programmatic ABM.
- Device and email graphs. Hashed contact lists are matched to cookies or device IDs.
- Third-party intent data. Accounts that read more about a topic across publisher networks show a "surge" and get prioritised.
The ad then appears next to whatever the person is reading: news, a recipe, a sports score.
How Conversational Intent Advertising works
CIA doesn't buy space on pages. It appears in one place: under an AI assistant's answer, when the conversation matches the problem you advertise on. In ChatGPT ads you describe those conversations with context hints. A relevance-weighted auction picks the ad, which is labelled as sponsored. After the click, the page continues the conversation and the visiting company is identified. What is Conversational Intent Advertising?
Side by side
| Programmatic / programmatic ABM | Conversational Intent Advertising | |
|---|---|---|
| Where the ad appears | Thousands of websites and apps | Under an AI assistant's answer |
| How the buyer is chosen | IP ranges, devices, cookies, intent scores | The problem described in the conversation |
| Intent signal | Inferred from browsing and reading | Declared in the buyer's own words |
| Context around the ad | Unrelated content | An answer about your category |
| Reported CPM | About $5–20 on the open web, plus platform fees | About $25–60, averaging around $45 |
| Frequency | Planned at 15–20 impressions per person a month for awareness | Usually once, at the moment of the question |
| Account matching | IP accuracy of about 70–90% for companies over 500 staff, 20–40% under 50 | None before the click; companies named after it |
| Privacy model | Cookies, device IDs and third-party data | Advertisers never see conversations |
| Retargeting | Yes | Not inside ChatGPT ads |
| Best for | Awareness and frequency across a known list | Being chosen at the decision |
Programmatic figures come from a 2026 vendor guide. CIA cost figures come from agency reports. Treat both as directions, not benchmarks.
Inferred intent vs declared intent
This is the real difference, and the reason we treat CIA as its own discipline.
Inferred intent is a guess built from behaviour. Someone at an account read three articles about reconciliation this week, so the account "surges" on the topic. It might be a buyer. It might be an analyst writing a report, or a student.
Declared intent is a statement. "We need reconciliation software for a 400-person company on NetSuite, ideally live before year-end" leaves little to guess. Advertisers don't see the sentence itself. But an ad group built for that question only appears when a question like it is asked, and each click tells you which question brought the buyer.
Intent data still has a job: it covers accounts that haven't asked anything yet. Use it to choose which questions to advertise on. More in intent data vs AI buyer questions.
Why CPM comparisons mislead
On a CPM basis, CIA looks several times more expensive than open-web display. That's the wrong comparison. An illustrative month with $3,000 of media:
| Illustrative, $3,000 a month | Programmatic ABM | Conversational Intent Advertising |
|---|---|---|
| CPM | $12 (mid-point of $5–20) | $45 (reported average) |
| Impressions | About 250,000 | About 66,700 |
| What each impression is | A banner beside unrelated content, shown 15–20 times a month to the same person | An ad under an answer to a buying question, shown once |
| People reached | About 12,500–16,700 at awareness frequency | Up to about 66,700 conversations |
| Of those, actively looking now | Unknown | All: each asked a question in your category |
These are illustrative calculations, not results. Judge both on the same outcomes: target accounts engaged, companies that fit your profile, meetings and pipeline.
The remote-work problem
Programmatic ABM depends on IP matching, and remote work broke much of it. Home broadband and mobile networks carry no company signal. One vendor guide puts IP accuracy at 70–90% for companies over 500 people but only 20–40% for companies under 50. The limits of IP-based ABM.
CIA targets the question, so it reaches the buyer wherever they work. To be honest about it, the same limit returns after the click: company identification also uses IP data, and it names office-network visits more easily than home ones. We report matched and unmatched visits side by side, so no demand goes uncounted.
Where programmatic still wins
- Reach and frequency. Nothing beats it for keeping your name in front of a list of accounts every week.
- Formats. Video, connected TV, native and audio.
- Retargeting. Following up with visitors and engaged accounts across the web.
- Control. You decide which accounts see the ad and how often.
- Cost per impression. Cheap awareness, when awareness is the goal.
Using both: the CIA-to-programmatic workflow
- CIA identifies in-market accounts. Companies that clicked on vendor and decision questions are named and scored.
- They become a priority tier. Fitting accounts go to your DSP or ABM platform as a separate tier with higher frequency.
- The creative continues the question. Programmatic ads for that tier pick up the topic the account asked about.
- Intent data feeds CIA back. Topics that surge across your list become new questions to advertise on.
- Measure lift. Compare meetings and pipeline from accounts in the priority tier with the rest of the list.
Programmatic was built for a web of pages. Conversational Intent Advertising is built for a web of answers. B2B buyers now use both, so the workflows we're building connect them: CIA tells programmatic where to concentrate, and programmatic keeps those accounts warm until sales gets the meeting.
Questions
Can we buy ChatGPT ads programmatically?
Not on the open exchange. ChatGPT ads are bought through OpenAI Ads Manager or OpenAI's partners, which include some ad tech companies. See our guide to buying partners.
Is CIA more expensive than programmatic?
Per 1,000 impressions, yes. Reported CPMs are about $25–60, against about $5–20 for open-web display. Per in-market account it can be cheaper, because every impression appears on a buying question.
Should we cut programmatic to fund CIA?
Usually not all at once. Keep programmatic for awareness on your list, run CIA on decision questions for 90 days, and move budget based on cost per fitting account and per meeting.
Does CIA use cookies?
No. Ads are matched to the conversation inside the platform, and advertisers never see chats. Conversion tracking and company identification on your website follow your own consent setup.
Facts checked against the sources below on 14 October 2026. AI ad platforms change quickly: if something here no longer matches what you see, tell us and we will update it.
