OpenAI Ads Agency for B2B SaaS

An OpenAI ads agency plans, launches and runs your paid campaigns inside OpenAI Ads Manager. The platform that places sponsored cards beneath ChatGPT answers. For a B2B SaaS company there is a second job that matters just as much, and almost nobody names it: OpenAI does not show ads to anyone on a Plus, Pro or Business plan, so the buyers sitting inside a company-provisioned ChatGPT workspace never see the card at all. Reaching those people means being cited inside the answer itself. We run both halves off one prompt map.

That is an uncomfortable thing for an ads page to say, so let us be direct about why we say it. ChatGPT Ads is a real channel with real intent, and for the right SaaS category it is worth buying. But it is sold today on claims that do not survive ten minutes with OpenAI’s own documentation. Guaranteed “partner access” that anyone can get, B2B conversion benchmarks that OpenAI says do not exist, and audience reach that excludes every paid seat your buyers work in. You should know all three before you commit budget, not after.

What an OpenAI ads agency actually does for a SaaS company

The work splits into three jobs, and the first one surprises most growth teams because it happens before a single campaign is built.

Access comes first, and it cannot be delegated. OpenAI Ads Manager has no agency manager account. OpenAI’s setup documentation is explicit: each advertiser requires its own account, each account needs a distinct email address, and account verification runs through Persona in a rolling review queue. So you create and verify the advertiser account in your own name, then invite us into it as a member or viewer. A role you can revoke on any Tuesday afternoon. No agency can create that account for you, and any agency that claims it can is describing something that does not exist.

Structure comes second. Campaigns, ad groups and bids get built around the questions your buyers type into ChatGPT instead of around a keyword list, because there is no keyword list to build. The objective locks at campaign creation and cannot be changed afterward, so choosing between Reach, Clicks and Conversions is a decision you make once, deliberately, with the tracking consequences understood in advance.

Measurement comes third, and it decides whether the first two were worth doing. The OpenAI Pixel and the Conversions API have to be firing before spend starts. A conversational surface hides its own performance until you instrument it. Run two weeks without measurement and you have bought traffic, not knowledge.

Underneath all three sits the prompt map: the list of things your buyers ask before they choose a vendor. That single artifact drives your context hints, your card creative, your landing pages, and the organic ChatGPT citation work that reaches the buyers the ads cannot. One piece of research, two surfaces.

Who actually sees your sponsored card

This is the fact that should shape your entire decision, and we have not found a single competing agency page that states it plainly. OpenAI puts it in one sentence: “We do not show ads to users on Plus, Pro, or any Business plan, or to any accounts identified as belonging to users under 18.”

Read that against how B2B software gets bought. Your economic buyer is a VP of Engineering or a Head of RevOps whose employer bought them a ChatGPT Business seat. That seat carries zero ad inventory. You cannot bid your way onto their screen at any price, because there is no auction running there.

What you can reach is the same person on their personal account at 9pm, the individual contributor evaluating tools before escalating, the founder of a twenty-person company who never bought a team plan, and the long tail of self-serve buyers who were never going to talk to sales anyway. For a product-led SaaS with a self-serve motion, that is a valuable audience. For an enterprise sales motion with a six-month cycle and a procurement committee, it is a much thinner one.

Two further limits stack on top. OpenAI does not show ads to accounts identified as under 18, using stated age and age-prediction signals. And self-serve buying is live in exactly nine countries. The United States, United Kingdom, Canada, Australia, New Zealand, Japan, Korea, Brazil and Mexico. Thirty-one European markets are listed as coming soon, reachable today only through OpenAI’s ads team or a partner. If your pipeline is majority EU, that is a scoping conversation before it is a campaign.

None of this makes the channel worthless. It makes it a channel with a defined edge, and knowing where the edge sits is the difference between a test that teaches you something and a quarter of wasted budget.

Is B2B SaaS even an eligible category?

Here is the second thing nobody publishes. ChatGPT Ads does not run a list of banned categories with everything else permitted by default. It runs the opposite. A permitted-category allowlist, with everything outside it disallowed. OpenAI’s ad policies state that the supported categories are household and consumer goods, local services, travel and entertainment, and digital products and education. All other categories are disallowed at launch.

Notice what is missing from the permitted list: B2B software. It is never named. “Digital products” plausibly covers a SaaS product, and two signals suggest OpenAI expects SaaS advertisers. The standard conversion events include trial_started and subscription_created, which is a SaaS funnel and nothing else. But an allowlist plus manual advertiser review means eligibility is decided by a human reading your account rather than by you reading a policy page.

So we will not tell you B2B SaaS is an approved category, because OpenAI has not said so. We will tell you it is very likely in scope, that we check your specific category and creative against the policy during scoping, and that a rejected account costs you weeks in a queue nobody can expedite. That check is free and it happens before anyone builds a campaign.

If you sell into regulated verticals, read the middle column carefully. A fintech SaaS or a healthtech platform may land in case-by-case review, US-only, with licensure proof requested. That is survivable, but it is not a two-day launch.

Context hints, not keywords

ChatGPT Ads has no keyword list, no match types, and no negative keywords. Targeting runs on context hints: plain-language descriptions, set at the ad group level, of the conversations where your product belongs. OpenAI is careful about what they are not: hints are not exact-match keywords, not audience targeting rules, and not instructions to show your ad only in specific conversations. They guide matching. They do not guarantee delivery.

This breaks every habit a paid search team has. The instinct is to write short commercial phrases, because that is what a keyword is. The platform wants the opposite. OpenAI’s own worked example replaces “running shoes” with “cushioned everyday running shoes for beginners training for their first 5K.” The hint carries the situation.

Translated to B2B SaaS, the same move looks like this:

The left column will either fail to deliver or deliver against conversations that have nothing to do with buying. The right column describes a person mid-decision. Every one of those hints comes out of the prompt map, which is why the research step is not optional busywork. It is the targeting.

Alongside hints you get controls that behave conventionally. Country targeting at campaign level, plus state, DMA and ZIP in the United States. Platform targeting across iOS app, Android app and web. Custom Audiences built from hashed email or phone lists, with a hard floor of 25,000 matched users before an audience is usable and 100,000 recommended, applied as campaign-level include or exclude and ad-group-level bid multipliers from 0.1× to 10×. There is no documented lookalike product and no documented pixel-based site retargeting. Audiences are list-based only.

The sponsored card, spec by spec

A ChatGPT ad is one sponsored card rendered below the end of the assistant’s response, clearly labeled and visually separated from it. Six elements make it, and every one has a published constraint that almost no agency page prints.

OpenAI does not publish the field limits on its public help pages, Ads Manager enforces them at upload, but the guidance it does publish points the same direction. Titles should stay concise and spell out the value. Copy should complement the title with new information rather than repeat it. Images should be simple, uncluttered and aligned with the message. In practice the card is far tighter than a display unit, so copy written for a landing page hero truncates the moment it lands. We write to the card.

One more instruction matters more than any character count: build for coverage instead of one message. OpenAI is explicit that a high volume of distinct title and copy variations lets the system find more conversations where you belong. Four near-identical ads is not four ads.

One rule sits outside the creative entirely, and it kills more launches than anything else: your landing page must be reachable by OAI-AdsBot. Link to the most relevant destination rather than defaulting to your homepage, and add your own UTM parameters directly to each destination URL.

Objectives, bidding, and where the auction sets your cost

Three objectives exist, each with its own bidding model, and the objective locks at campaign creation. You cannot switch a Clicks campaign to Conversions later. You build a new one.

oCPC has a wrinkle worth stating plainly, because it is the most misunderstood control on the platform. Your Bid Cap is the maximum you are willing to bid for a conversion, and the system uses predicted conversion likelihood to set a per-click auction bid, but billing remains based on valid clicks. It is not a guaranteed cost per acquisition.

Three more oCPC constraints are worth knowing before you build. Conversion tracking has to be live first. Only supported standard conversion events work. Custom events are not currently supported for oCPC. And each campaign optimizes toward exactly one event, chosen at creation and unchangeable afterward. Existing CPM or CPC campaigns cannot be converted to oCPC either, so a Conversions campaign is always something you build fresh.

Because relevance is weighted against your bid, the cheapest lever on this platform is not money. It is a hint that describes the conversation more precisely than your competitor’s. That is unusual, it will not last forever, and it currently rewards research over budget.

For opening bids, OpenAI recommends starting CPC campaigns at a maximum bid of $3 to $5 per click. Bids are set at the ad group level, and Ads Manager surfaces bid-strength guidance as you go.

Tracking that has to exist before you spend

Measurement on this platform is unusually well documented and unusually easy to get wrong. Four things have to be true before your first impression.

The pixel is live. The OpenAI Pixel loads the oaiq browser SDK and fires your standard events. For SaaS, the relevant ones already exist as standard events: trial_started, subscription_created, lead_created, registration_completed and appointment_scheduled. Custom events are supported for anything else.

The Conversions API is wired alongside it. Server-side events give you the conversions a browser will drop. Send both with the same event_id and OpenAI deduplicates. It keeps the first event for a matching key and ignores the rest.

The click reference survives your redirects. OpenAI appends a parameter called oppref to your landing page URL. The pixel captures it and stores it in a first-party __oppref cookie so later conversions can be tied back to the click. If your marketing site bounces visitors through a redirect chain that strips query parameters, and a great many SaaS sites do, you lose attribution silently. Nothing errors. The number is just wrong.

OAI-AdsBot can reach your landing pages. OpenAI states you must allow it, and recommends allowing OAI-SearchBot too. Three layers have to cooperate: robots.txt, your WAF or CDN bot mitigation, and any CAPTCHA or JavaScript challenge. Ads whose landing pages cannot be crawled are not eligible to run, and support cannot manually bypass validation. This is the single most common reason a technically correct campaign serves nothing.

One reporting detail matters for how you read early results. View-through attribution is fixed at one day, is reported separately as VTA, and is excluded from your headline conversions total. It does not affect CPA, bidding, billing or optimization. If both a click and an impression qualify for the same conversion, the click wins. Delivery data lags up to seven hours; conversions lag 24 to 48. Do not judge a Tuesday on Tuesday.

What ChatGPT ads cost

Your investment splits three ways: media you pay OpenAI, the setup work, and a monthly fee. Only the first has published numbers.

Media starts at a minimum daily budget that varies by your ad account’s billing currency. OpenAI publishes the applicable amount in the Minimum Campaign Spend table inside Ads Manager rather than on a public page, so we confirm your market’s floor during scoping instead of quoting a number that may not be yours.

How the budget paces matters more than the floor. A daily budget is an average over a seven-day period rather than a hard daily cap. Spend on any single day will not exceed twice your daily budget, and spend across the seven-day period will not exceed seven times it. Underspend earlier in the week can be used later in the same week. So a campaign that looks like it overspent on Tuesday is usually pacing correctly.

You can also set a campaign-total budget instead, which is a hard cap over the campaign duration. Choose carefully: some budget-type changes are not reversible, and switching a campaign-total budget to a daily budget cannot be undone.

Whatever the floor is in your market, it is a floor and not a plan. At the $3 to $5 opening CPC OpenAI recommends, a minimum-budget day buys a handful of clicks: enough to confirm delivery, nowhere near enough to learn which hint clusters convert. We model a realistic test budget against your own funnel on the call rather than publishing a number that would be wrong for most companies.

On our own fee: it is a fixed monthly retainer, never a percentage of spend. A percentage fee pays us to talk you into bigger budgets, which is precisely backwards on a channel where the honest answer is sometimes “spend less here and more on citations.” The number moves with market count, tracking maturity, and how much creative you already own, so we scope it on a call. Our published plans and pricing show how we scope the earned side.

You own everything. The ad account, the pixel, the audiences, the conversion data, and the creative. We work inside your account under a role you grant and can revoke. If you leave, nothing walks out with us, because none of it was ever ours.

ChatGPT Ads vs Google Ads vs LinkedIn Ads vs earned citations

The useful question is not which channel is best. It is which job each one does, because for B2B SaaS they do different jobs.

Read the third row again. It is the only row where ChatGPT Ads scores a flat no and earned citations score a flat yes, and for B2B SaaS it is the row that decides your mix.

The half that reaches everyone

The sponsored card buys the slot under the answer. It does not put your brand inside the answer, and OpenAI is emphatic that it never will. Ads do not influence the answers ChatGPT gives, they run on separate systems from the chat model, and advertisers have no ability to shape, rank or alter responses. Advertisers never receive user conversations either; they get aggregated, non-identifying performance data.

That separation is not a limitation to work around. It is the reason the other half of the work exists. Getting named inside the answer is an authority problem and not a media-buying problem, and it is the problem this agency was built for.

The earned half runs on the work we already do: digital PR, brand mentions, and content built to be the source an answer engine reaches for. If you want the mechanics of that side, we have written them up in full: AI search for SaaS and our GEO framework for B2B SaaS.

Running the two together is not a cross-sell. Prompt research for context hints is the same research that tells you which questions to earn citations on. Do it once, spend it twice.

Your first 90 days

The clock on everything after account approval starts when OpenAI clears your account rather than when you sign. That queue is rolling, OpenAI does not expedite it, and no agency can move it, so we build everything else while it runs.

1. Before day one: approval and mapping. You create the advertiser account and complete Persona verification in your own name. We check your category against ad policy first, build the prompt and intent map, draft context hints, and prepare landing pages and creative. Nothing here waits on approval.
2. Days 1 to 30: launch and instrument. Pixel and Conversions API verified against live events. Landing pages confirmed open to OAI-AdsBot through robots.txt, WAF and any challenge layer. Campaigns built by objective, since the objective locks. Ads submitted for review, edited and resubmitted on rejection. Weekly delivery notes from first serve.
3. Days 31 to 90: read, then scale. We judge hint clusters on conversions rather than clicks, which is why this phase runs sixty days and not thirty. Losing hints get cut, winning clusters get widened, and a separate oCPC campaign gets built rather than switched. Earned citation targets get set against the same prompts.

At a realistic starting budget, a single month produces too few conversions per hint cluster to act on. Anyone promising you a readable signal in thirty days is promising you noise.

Where the numbers have to land

Before you commit budget, the arithmetic has to clear a bar you set. This is not a forecast. It is a worked example, and only one number in it comes from OpenAI.

Only the bid range is OpenAI’s. Every other rate has to come from your own data, and if you cannot supply them the honest conclusion is that your tracking is not ready for this channel yet. Two inputs decide whether ChatGPT Ads works for you: what a qualified opportunity is worth, and whether your instrumentation can see one when it arrives. We run this with your real numbers on the call, before anyone commits budget.

Running it in-house or with us

Plenty of SaaS teams can run this channel themselves, and some should. The difference is who carries the tracking work and how fast you get a signal you can trust.

One thing we will not claim: a long track record on this specific platform. Self-serve ChatGPT Ads opened in May 2026. Nobody has years on it, including us. What we can show you is a decade of earned authority work on the half of the answer that reaches every plan, and that is the half we were building before ads existed.

Frequently asked questions

What does an OpenAI ads agency do?

It runs your paid campaigns inside OpenAI Ads Manager, the platform that places sponsored cards below ChatGPT answers. The work covers prompt research, context hints, card creative, campaign and bid structure, conversion tracking, and reporting. For B2B SaaS we also run the earned citation inside the answer, because the paid card does not reach anyone on a paid ChatGPT plan.

Who can see ads inside ChatGPT?

Everyone except the paid tiers. OpenAI states that it does not show ads to users on Plus, Pro, or any Business plan, or to accounts identified as belonging to users under 18. For B2B SaaS this is the single most important fact about the channel, because your buyers usually sit in company-provisioned workspaces where no ad inventory exists at all.

Can B2B SaaS companies advertise on ChatGPT?

Almost certainly, but OpenAI has not said so explicitly. The permitted-category list names “digital products and education,” which plausibly covers SaaS, and the standard conversion events include trial_started and subscription_created. Since it is an allowlist with manual advertiser review, eligibility is confirmed by a human reading your account. We check your category and creative against the policy during scoping.

How much do ChatGPT ads cost?

The floor is a minimum daily budget that varies by your ad account’s billing currency; OpenAI shows the applicable amount in the Minimum Campaign Spend table inside Ads Manager. For CPC campaigns OpenAI recommends starting with a maximum bid of $3 to $5 per click. A daily budget is an average over seven days. Single-day spend can reach twice it, and seven-day spend will not exceed seven times it. Our management fee is a fixed retainer scoped on a call, never a percentage of spend.

How is ChatGPT ads targeting different from Google Ads?

There is no keyword list, no match types, and no negative keywords. You write context hints in plain language describing the conversations your card should sit beside, and the system pairs them with live intent. OpenAI is explicit that hints are not exact-match keywords and do not guarantee delivery. Geo, platform and custom-audience targeting work roughly as you would expect.

Can you set up the ad account for me?

No, and no agency can. OpenAI requires one account owner at your business to create the advertiser account, each advertiser needs its own account with its own email address, and there is no agency manager account. You create and verify it through Persona, then invite our team as a member or viewer under a role you can revoke.

Do you need a certified OpenAI ads partner?

No. Some agencies market “official partner access” as a scarcity play, but self-serve onboarding has been open to any verified advertiser in an eligible country since May 2026. OpenAI’s named ad partners are holding companies and adtech vendors: Dentsu, Omnicom, Publicis, WPP, Adobe, Criteo and similar. There is no public per-agency certification to hold.

Which countries can run ChatGPT ads?

Nine have self-serve access: the United States, United Kingdom, Canada, Australia, New Zealand, Japan, Korea, Brazil and Mexico. Thirty-one European countries are listed as coming soon; advertisers there can start today through OpenAI’s ads team or a partner. We check your target markets before scoping.

What is oCPC bidding and when should you use it?

oCPC is the bidding model for the Conversions objective. Your Bid Cap is the maximum you will bid for a conversion, and the system converts that into a per-click auction bid using predicted conversion likelihood, but billing remains based on valid clicks, so it is not a guaranteed cost per acquisition. Conversion tracking must already be live, only supported standard events work, and each campaign optimizes toward one event chosen at creation that cannot be changed afterward.

Can a competitor bid on conversations where your brand comes up?

Yes, and you cannot fence it off. Context hints are not exact match, so there is no brand-term equivalent to a Google Ads keyword you can defend. Presence is the only defense, and the more durable half of it is being the brand cited inside the answer rather than the one advertising underneath it.

Do ChatGPT ads change the answers people get?

No. OpenAI states that ads do not influence ChatGPT’s answers, that ads run on separate systems from the chat model, and that advertisers cannot shape, rank or alter responses. Advertisers also never receive user conversations, only aggregated, non-identifying performance data.

How do you track conversions from ChatGPT ads?

Through the OpenAI Pixel and the Conversions API, sent with a shared event_id so duplicates collapse. OpenAI appends an oppref click reference to your landing page URL, which the pixel stores in a first-party __oppref cookie so later conversions tie back to the click. If your redirects strip query parameters, attribution breaks silently.

Why would ads be approved but never serve?

Most often because OAI-AdsBot cannot reach the landing page. OpenAI requires it to be allowed, and three layers must cooperate: robots.txt, your WAF or CDN bot mitigation, and any CAPTCHA or JavaScript challenge. Ads with uncrawlable landing pages are not eligible to run and support cannot bypass validation. A missing advertiser name or logo in account settings will also stop delivery.

What are the ChatGPT ad creative specs?

Every ad unit needs six things: an advertiser name, a favicon-style logo, a title, copy, an image asset, and a landing page. OpenAI enforces field limits at upload in Ads Manager rather than publishing them on its help pages, and its published guidance is directional. Keep titles concise and specific, make the copy complement the title instead of repeating it, and keep images simple and aligned with the message. It also recommends running many distinct title and copy variations rather than one polished ad.

How quickly can a campaign go live?

As soon as your account clears verification and your ads clear review. OpenAI reviews advertiser accounts in a rolling queue and does not expedite them, so that step sets your date, not us. We run prompt research, hint drafting, creative and pixel work in parallel while the queue moves.

Are there benchmarks for B2B SaaS performance on ChatGPT ads?

No. OpenAI states plainly that ChatGPT Ads does not yet have performance benchmarks across advertisers, industries or campaign types. Every “B2B SaaS converts three times better on ChatGPT” figure in circulation is agency marketing with no primary source behind it. We will model your economics from your own funnel data instead.

What is the difference between ChatGPT ads and GEO?

They buy different halves of the same result. Ads buy the sponsored card below the answer, deliver within days of approval, and stop the day you stop paying. Generative engine optimization works on the answer itself so your brand gets named and cited inside it, takes weeks to months, reaches every ChatGPT plan including the Business plans ads cannot touch, and persists until the model reweights.

Can you run ChatGPT ads through a DSP instead of buying direct?

Some adtech partners offer that path, and OpenAI’s advertising terms formally define an advertising intermediary and reseller program. For most B2B SaaS budgets, buying direct in your own advertiser account is simpler, cheaper, and keeps the pixel and conversion data in your hands rather than an intermediary’s.

Should you run ChatGPT ads or spend the budget on citations?

It depends on where your buyers sit. A product-led SaaS with self-serve signups reaches real demand on the consumer tiers and should test the card. An enterprise sales motion selling into companies that provision ChatGPT Business seats will reach very few of its actual buyers through ads, and should put the budget into being cited instead. Most companies need some of both, weighted differently than they expect. That weighting is the first thing we work out on the call.