ChatGPT Ads are live. Here is what the first real B2B data means for security marketers

There is a new advertising channel worth paying attention to, and for once the security industry does not have to wait three years to find out whether it works.

OpenAI opened ChatGPT Ads to self-service advertisers this year, and the ads run underneath relevant conversations for free-tier users. There are no keywords to bid on, no way to buy the phrase "access control system," and no device or demographic targeting. Instead you describe your buyer and the problems they are trying to solve, and the model decides when your ad belongs in the conversation.

Nobody in security has published numbers on it yet, so at Maize we went looking for the closest thing available, and found it in a 30-day pilot run for industrial manufacturers and published in full. That audience is not ours, but it is close enough to be useful, because industrial buyers behave a lot like security buyers: technical, deliberate, and working through long cycles inside small qualified universes. If you sell video management software to hospital systems, or chase multi-site retail as an integrator, the read-across is fair. Here is what the data shows and what we think it means for anyone marketing security products.

01The numbers

The pilot ran two campaigns with three ad groups each and four ads per group, launching June 22 on a $25 daily minimum per campaign for $1,470 in total spend.

Read that table twice, because the interesting number is not the zero at the bottom of it.

02The clicks are cheap right now

They set a maximum CPC bid of $20, which is a realistic ceiling for competitive industrial search terms, and the actual cost came in between $5 and $6, roughly a third of what the same click costs on Google Search.

Anyone who has pulled a Google Ads report for "cloud access control" or "video surveillance system for schools" knows those clicks run $12 to $25 and climb every year, bid up by the large security manufacturers and by every regional integrator with a $3,000 monthly budget. Almost none of them are running on ChatGPT Ads yet, and that gap is the entire opportunity.

Early channel pricing never lasts. It did not last on Facebook and it did not last on LinkedIn, so the advantage will go to whoever has a working account before the auction fills up.

03There is real inventory for niche B2B audiences

Nearly 40,000 impressions in a month is meaningful volume, particularly on tightly defined industrial targeting and on a platform that currently serves ads only to free-tier users in select countries.

The security market is smaller than industrial manufacturing, so scale your expectations down accordingly. Even so, the worry that a niche B2B audience simply would not be there in enough volume to matter did not hold up, and the quality held as well. Half the sessions were engaged at an average of a minute and a half, which puts the traffic in Display and LinkedIn territory rather than anywhere near the bot range.

04Contextual targeting is a different question than keyword targeting

This is the part worth sitting with, because it changes the question you are asking. Google Ads asks what your buyer types into a search box, while ChatGPT Ads asks what conversation your buyer is having before they ever open a search box.

Those are not the same question, and in security the gap between them is enormous. A facilities director is not searching "mobile credential migration." What they are actually doing is opening ChatGPT and typing something closer to this: "We have 400 employees across six buildings on an old prox card system and we keep losing badges. What are our options?"

That example is not unusual, either. Our market generates versions of these every day:

• "How do I write an RFP for a video management system?"
• "What does NDAA Section 889 compliance actually require for cameras?"
• "Cloud access control versus on-premise, what is the real five-year cost?"
• "How many cameras do I need to cover a 200,000 square foot warehouse?"
• "Our alarm monitoring contract is up. What should I be asking vendors?"

Every one of those is a buying conversation wearing the clothes of a research question, which is exactly what makes the inventory underneath it worth buying.

05Zero conversions, and why that is the expected result

The pilot produced no form fills, that number was published alongside all the others, and the honest read is that this is what a first controlled test on a new platform looks like.

There are a few reasons it should not scare you off. The platform optimizes for clicks and conversion optimization is not yet available, so the algorithm was never pointed at the outcome that matters. The budget was $1,470, which is a rounding error rather than a media test. And the buying cycle in security is what it has always been, in that a video upgrade for a school district takes nine to eighteen months and moves through a consultant, a specification, a bid and a board vote, so nobody was going to see an ad under a ChatGPT answer in June and sign in July.

The practical risk is that you run this channel, grade it on month-one MQLs and kill it before it has a chance to work. Grade it instead on the indicators that move first, which are engaged sessions, time on page, assisted conversions, branded search lift, and whether the companies showing up in your traffic data look like the companies you actually sell to.

06What a security-specific test looks like

If you decide to run one, the real work is not the ad itself but the Context Hints, which are the plain-language description of who you sell to and what keeps them up at night. Most of what you need is already sitting in your account:

• Your winning Google search terms, meaning the queries that actually converted last year rather than the ones with the most volume.
• Your ICP written with real specificity. Not "security professionals," but something closer to security and facilities directors at multi-site healthcare systems with 5 to 40 locations, running legacy panels past end of life while IT pushes them toward a cloud platform it can manage centrally.
• The questions your sales team answers on every first call, written down verbatim, because those are the conversations you are trying to appear inside of.
• Your competitive displacement story, meaning what customers are switching away from and what finally makes them start looking.

From there, keep the test honest by holding Context Hints constant within each segment and varying only the creative and the landing page. Give it 60 to 90 days rather than 30, and budget enough that the results mean something, which for most manufacturers lands somewhere between $3,000 and $5,000 across the test window. We lay out how we structure paid programs like this on our services page.

What the first published data tells us is that a new channel has opened with cheap clicks, real inventory and almost no competition from our industry, and that nobody has figured out how to convert it yet. Both of those things are true at the same time, and both of them argue for starting now rather than in eighteen months, when the CPCs have tripled and the learning curve belongs to someone else.

The companies that own Google Ads in this industry today are the ones who were buying clicks back when they were cheap and nobody could prove it worked. That window has opened again.