ChatGPT Ads for B2B Software: What Ships Today, and Who You Cannot Reach
Most articles on ChatGPT ads are still written as if the channel is coming soon. It is not. It shipped, it is self-serve, and you can have a campaign running this afternoon.
The question worth asking is narrower and more uncomfortable: can this format reach a B2B software buyer at all? For a large share of B2B SaaS companies, the honest answer is that it reaches a fraction of your buying committee, and OpenAI has published exactly which fraction. That single constraint should drive your decision more than anything else written about this channel so far.
Here is what actually ships, what it costs, who is excluded, and what a defensible test looks like.
What Is Live Right Now
OpenAI began testing ads in ChatGPT on February 9, 2026, and the platform has moved quickly since. The self-serve Ads Manager opened in May 2026, removing the spend minimums that had gated the pilot to large managed accounts. Product feed campaigns for retail catalogs followed in June. Advertiser docs, a feed specification, and a public Advertiser API now live at developers.openai.com/ads.
The mechanics, per OpenAI's own advertiser documentation:
Placement. Ads appear below a ChatGPT response, clearly labeled as sponsored and visually separated from the answer. They do not appear inside the answer text. OpenAI has been explicit that ads run on separate systems from the model and that advertisers cannot shape, rank, or alter what ChatGPT says. If you have read that ChatGPT ads appear woven into the recommendation itself, that is wrong, and it is wrong in the direction that flatters the channel.
Format. A single unit containing advertiser name, favicon, headline, description, image, and landing page. Short. Closer to a sponsored card than a search ad.
Targeting. There are no exact-match keywords. You supply context hints at the ad group level describing the conversations and topics where your product is relevant. The system weighs those hints alongside conversation intent, your landing page, and your ad copy. Hints guide matching. They do not guarantee delivery.
Pricing. CPM buying under a Reach objective, or CPC under a Clicks objective. You set a maximum bid at the ad group level. OpenAI recommends starting CPC campaigns at a $3 to $5 maximum bid. Selection runs through a relevance-weighted second-price auction.
Reporting. Impressions, clicks, spend, CTR, average CPC, average CPM, and conversions. Conversion measurement runs on a pixel plus a Conversions API, the same pattern as Meta CAPI or GA4 Measurement Protocol. UTM parameters persist on ad clicks.
That is a functioning performance channel. It is not a preview.
The Paid Tier Problem, Which Is The Whole B2B Story
Here is the line that should determine your budget decision, straight from OpenAI's advertiser documentation: ads are not shown to users on Plus, Pro, or any Business plan. The user-facing FAQ extends that to Enterprise and Edu accounts. Ads serve to logged-in adults on the Free and Go tiers only.
Now map that against a B2B software buying committee.
The people you are trying to reach are a VP of Engineering, a Head of RevOps, a procurement lead, a CTO at a Series B company. These are exactly the people whose employer bought them a ChatGPT Business seat, or who expensed Plus eighteen months ago, or who are on Pro because they use the thing eight hours a day. Software buyers are among the heaviest and earliest AI adopters on the planet. Heavy adoption means paid plans. Paid plans mean no ads.
The more senior and more technical your buyer, the less likely they are reachable through this channel. That is an inverse relationship between ad reach and deal value, and it is close to unique among paid channels. LinkedIn gets more expensive as you target more senior titles. ChatGPT ads simply cannot show them anything.
This does not make the channel worthless. It makes it a specific channel for specific businesses:
- Products with a self-serve, low-ACV motion where the evaluator is an individual practitioner rather than a committee, and where that practitioner may well be on the free tier.
- Products bought by small businesses and solo operators, where the buyer is the owner and the owner is price-sensitive about everything including their own AI subscription.
- Prosumer-adjacent tools that straddle the line between personal and business use. If you sell six-figure ACV enterprise software, the reachable audience for your ad is people researching your category who are not on a paid AI plan. Interrogate who that actually is before you fund it.
The Second Gate: Where Your Company Is Registered
There is a separate eligibility layer that catches European companies in particular, and it is not about your customers.
Advertiser accounts are gated by the country where your legal entity is registered, not by where you sell. Serving markets and advertiser-eligible markets are two different lists, and they do not match. A company registered outside the eligible list cannot open an Ads Manager account even if its entire customer base sits inside a serving market.
For continental European SaaS, including Swiss, German, and French companies, this is currently a hard stop rather than a queue. The account country is set at creation and is not editable afterward, so the only route in is an entity that is already registered in an eligible market.
Country availability is the fastest-moving fact in this whole space, so check OpenAI's current advertiser availability rather than any third-party list, including this one. What is stable is the rule: entity registration governs access, and setting up an entity purely to buy ads is almost never worth it at test-budget scale.
Shopping Ads and Product Feeds Are Not Built For You
A lot of B2B software content has attached itself to the phrase "ChatGPT shopping ads." Worth being direct: product feed campaigns are documented as being for retail advertisers with broad or frequently changing catalogs. The feed schema is a commerce catalog format built around SKUs, price, availability, and checkout. During the beta, feed products are eligible for paid placements only and do not surface in organic conversations.
A SaaS product with three pricing tiers and no inventory does not map onto that schema in any useful way. If you sell software, you are running standard ChatGPT ad campaigns with hand-built creative and context hints. Skip the feed documentation entirely. Any advice telling you to prepare a product feed for your SaaS is advice written by someone who has not opened the spec.
Channel Comparison for B2B Software
| Channel | Intent stage | Cost model | Targeting | B2B software fit | Best for |
|---|---|---|---|---|---|
| ChatGPT Ads | Mid, active research | CPM (Reach) or CPC (Clicks), second-price auction, $3 to $5 suggested starting CPC | Context hints plus conversation intent, no firmographics, no exact-match keywords | Limited by Free and Go tier exclusion; weakest fit for senior or enterprise buyers | Self-serve, low-ACV products with individual evaluators |
| Google Search Ads | High, keyword-triggered | CPC, auction | Keyword plus audience layers | Strong, and still the highest-intent capture available | Buyers who already know the category and are comparing vendors |
| Google Performance Max | Mixed | CPC/CPM, automated | Algorithmic across Google properties | Moderate; needs real conversion volume to optimize against | Scaling after you have proven conversion data |
| LinkedIn Ads | Low to mid, awareness | CPM/CPC, expensive | Job title, seniority, company size, industry | Strong, and the only channel that reliably reaches the exact people ChatGPT ads exclude | Account-based targeting and reaching buying committees by role |
| Review platforms (G2, Capterra) | High, comparison | CPC, category placement | Category and competitor intent | Strong, and doubles as retrieval surface for AI answers | Shortlist-stage capture with compounding organic benefit |
The row that matters is the last one. Review platform presence buys you paid placement and simultaneously improves the odds of being cited organically when someone asks ChatGPT to compare tools in your category. For B2B software, that is a better use of the next thousand dollars than a ChatGPT ad campaign in most cases.
If You Test It Anyway, Test It Properly
Assuming you have entity eligibility and a buyer profile that plausibly sits on a free tier, here is a defensible structure.
Pick the Clicks objective, not Reach. Reach optimizes for impressions and bills per thousand. You do not need conversational impressions. You need people on your site.
Start at the recommended $3 to $5 max CPC and let the bid guidance tell you if you are uncompetitive. Do not open at $12 because your Google CPCs are $40. Different auction, different supply, and you are buying data before you are buying pipeline.
Write context hints as buyer situations, not as keyword lists. The system reads them as descriptions of relevant conversations. "Teams evaluating tools to replace spreadsheet-based inventory tracking" gives the matcher something to work with. "inventory software, inventory tool, best inventory" does not, because exact matching is not what is happening.
Write copy for a person mid-research, not mid-purchase. The unit is a headline, a description, and an image sitting under an answer. The job of that unit is to earn a click into a comparison page, not to close. Lead with the specific situation your product solves and one concrete differentiator (no per-seat pricing, native Jira sync, deploys in your VPC). Skip the trial CTA. Someone three questions into researching a category is not starting a trial from a sponsored card, and optimizing your landing page for that will make the numbers look worse than the channel deserves.
Set the conversion event at the top of your funnel. Fire the pixel on demo request or trial start, not closed-won. Your sales cycle is longer than any sane optimization window, and the system needs volume to learn against.
Do the arithmetic before you spend. This is illustrative, and you should replace every number with your own: at a $4 CPC and a 3 percent visit-to-trial rate, a trial costs roughly $133. At a 20 percent trial-to-paid rate, a customer costs roughly $665 before sales cost. Whether that is good depends entirely on your ACV and payback target. Run this calculation first. If it fails at plausible inputs, the test is not worth running regardless of how novel the channel is.
Cap it and time-box it. Set daily budgets in Ads Manager. Give it 60 to 90 days and a predefined kill threshold. New channels get extended on vibes far more often than on numbers.
Measurement, Honestly
The measurement stack is better than most people expect from a beta. Pixel plus Conversions API plus persistent UTMs means you can attribute a click to a trial with the same rigor you apply to Meta or Google.
What you still cannot do:
Separate paid influence from organic influence. Your ad appeared under an answer. That answer may also have mentioned your product, or three competitors, or given advice that reframed the buyer's requirements entirely. You get the click data and none of the conversational context, by design, since advertiser access to conversations is explicitly walled off.
Attribute assisted conversions across a 90-day cycle. A ChatGPT ad that plants your name in week one and a branded Google search that converts in week eleven will hand the credit to Google under any last-click model. This is not specific to ChatGPT ads, but it bites harder here because the channel sits so early in the research process.
See the organic side at all. If ChatGPT recommends you without an ad, the visitor arrives with no paid marker and often no referrer at all.
That third gap is the one worth sitting with, because for most B2B software companies it is larger than the paid opportunity.
The Bigger B2B Opportunity Is Not The Ad
Reachable ad audience for B2B software is structurally capped by the paid tier exclusion. Organic citation has no such cap. When a buyer on a Business plan asks ChatGPT to compare tools in your category, they see no ads at all. They see an answer. Being in that answer is not something you can buy.
What moves it is unglamorous and largely already known: specific comparison and category pages that answer the questions buyers actually ask, presence and rating on the review platforms that retrieval systems lean on, clean structured data, and third-party coverage. None of it is new. All of it compounds.
The part that is genuinely new, and that almost nobody is measuring, is what happens after the answer. AI agents increasingly arrive at your site on a buyer's behalf to check pricing, confirm an integration, or pull spec details. They arrive with broken or absent referrers, so your analytics stack largely does not see them. And when they do arrive, they frequently cannot complete the task, because your pricing sits behind a form, your docs need JavaScript, or your comparison page renders client-side.
That is a different question from whether your ad got a click. It is whether the thing that showed up on your buyer's behalf could actually get what it came for. Serge exists to answer that question: passive monitoring of real agent traffic hitting your site, plus active replay that walks the same journey from the agent's perspective and tells you where it breaks.
Run the ad test if the arithmetic supports it. But if your buyers sit on paid ChatGPT tiers, the ad was never going to reach them, and the answer they do see is the thing worth instrumenting.
Frequently Asked Questions
Can B2B software companies advertise on ChatGPT?
Yes, if your legal entity is registered in an advertiser-eligible country. Software is not among the restricted ad categories. The practical constraint is not permission, it is reach: ads do not serve to Plus, Pro, Business, Enterprise, or Edu accounts, which is where a large share of B2B software buyers sit.
How much do ChatGPT ads cost?
You choose CPM under a Reach objective or CPC under a Clicks objective and set a maximum bid at the ad group level. OpenAI recommends starting CPC campaigns between $3 and $5. There is no spend minimum since the self-serve launch. Delivery runs through a relevance-weighted second-price auction, so your effective cost depends on competition and relevance rather than a rate card.
Do ChatGPT ads appear inside the answer?
No. They appear below a completed response, labeled as sponsored and visually separated. OpenAI runs ads on separate systems from the model, and advertisers cannot influence what ChatGPT says. Any claim that paid placement changes the recommendation is inaccurate.
Can I target by job title or company size?
No. There are no firmographic or professional targeting controls. You provide context hints describing relevant conversation topics, and the system matches on conversation intent, your hints, your copy, and your landing page. If you need to reach a specific title at a specific company size, that is still LinkedIn's job.
Should a SaaS company build a product feed?
No. Product feed campaigns are built for retail advertisers with catalog inventory, and the feed spec is a commerce schema around SKUs, price, and availability. Run standard campaigns with hand-built creative instead.
Is it better to buy ChatGPT ads or work on organic AI visibility?
For most B2B software companies, organic. Ads cannot reach the paid-tier accounts where senior and enterprise buyers concentrate, while organic answers reach everyone. Ads are a reasonable test for self-serve, lower-ACV products with individual evaluators. They are a poor substitute for being in the answer.
The Bottom Line
ChatGPT ads are a real, self-serve, properly instrumented performance channel. That part of the hype is now true.
The B2B software caveat is not a timing problem that resolves as the platform matures. It is a design decision. Paid tiers are ad-free because that is the value proposition of paying, and the people who pay for AI tooling are disproportionately the people who buy software for a living. Assume that stays true.
So: check your entity eligibility, run the cost-per-trial arithmetic at a $4 CPC, and if it clears, test with a hard cap and a kill date. Then spend the larger share of your attention on the answer itself, and on whether anything arriving from that answer, human or agent, can actually complete what it came to do.