ChatGPT Ads vs Earned AI Citations: Where B2B SaaS Should Spend in 2026

The evidence below comes from OpenAI's own ad pages, read on 23 September 2026, and from our own AI visibility data at LoudFace.

ChatGPT ads vs earned citations at a glance

Every ad fact in that table comes from OpenAI's help center. OpenAI does not publish a CPM rate or an average CPC. Where a number is someone else's reading, I say so.

What ChatGPT ads are and what they cost

ChatGPT ads are OpenAI's own ad product, bought through OpenAI Ads Manager, which OpenAI still labels a beta. Testing started in the US on February 9, 2026. After a pilot from February through April, OpenAI opened self-serve sign-up at ads.openai.com. Self-serve works only if the legal entity that pays is based in a listed country. The availability list on 23 September includes the United States, United Kingdom, Canada, Australia, India and every EU country. Some new accounts can only advertise in their home country until identity verification clears and they reach a required amount of spend there.

How you pay. You pick one of three campaign objectives, per OpenAI's campaign guide:

• Views (CPM): you pay per 1,000 impressions served.
• Clicks (CPC): you pay per valid click.
• Conversions: the system optimizes toward a conversion event, but you are still billed per valid click (oCPC) or per 1,000 impressions (oCPM). You never pay per conversion.

You set a maximum bid per ad group. For click campaigns, OpenAI recommends starting with a maximum bid of $3 to $5 per click. Ads compete in a relevance-weighted, second-price auction, so relevance to the conversation matters as well as your bid.

What it costs to start. The minimum daily budget is 25 USD, 15 EUR or 15 GBP, depending on your billing currency. A daily budget is an average across a Sunday to Saturday week, and a single day can spend up to twice that figure.

What OpenAI does not publish. Its own ads FAQ says the platform "does not yet have performance benchmarks across advertisers, industries, or campaign types," and warns that delivery and spend can fluctuate during the beta.

The launch pricing you may have read about. Neither of these figures appears on OpenAI's help pages today. In January, The Verge, citing The Information, reported OpenAI was asking around $60 per 1,000 views, and Adweek reported that OpenAI confirmed it was asking select advertisers to commit at least $200,000 during the beta. Treat both as reported history from the invite-only phase. The self-serve product today has the daily minimum above and a bid you choose.

Targeting. This is where B2B teams should slow down. ChatGPT matches ads to the context and intent of the current conversation, your landing page and copy, and "context hints" you write at ad-group level. OpenAI is blunt that context hints "are not exact-match keywords and do not guarantee delivery in specific conversations." On top of that you get countries and sub-national locations, five platforms (Android app, Android web, desktop web, iOS app, iOS web), and custom audiences. Custom-audience inclusion needs at least 25,000 matched users. The help pages I read list no job-title, company-size or industry targeting.

So if you were hoping to upload a list of target accounts and show ads to their buying committee, check the list size first. Most target-account lists I see fall well short of 25,000 matched users, and the pages I read offer no firmographic targeting to fall back on.

What is new this month. On September 16, 2026, OpenAI announced Sponsored Agents (a labelled conversation with a business's own agent after someone clicks its ad, in testing with select US advertisers), AI-suggested ad copy and images in Ads Manager, and integrations with HubSpot and Shopify. The HubSpot one matters for B2B SaaS: you can connect a ChatGPT Ads account, create ads, track performance and follow up on leads inside HubSpot.

Where an ad sits versus where a citation sits

A ChatGPT ad appears below the end of a response. OpenAI's help article says ads are "clearly labeled as sponsored and visually separated" from the answer. It also says ads "run on separate systems from our chat model," and that advertisers cannot shape, rank or alter ChatGPT's responses. By default ChatGPT cannot even see the ad the user is looking at.

An earned citation lives inside the answer. When someone asks ChatGPT which tool to use for compliance automation, or which payroll provider handles stablecoins, the answer names vendors and links the pages it drew from. Being one of those pages, and being the brand named, is the organic position.

So an ad sits under whatever the answer recommends, and only a citation gets your brand into the recommendation itself. That difference decides the budget question.

Our own data shows how tightly naming follows sources. In a sample of 120 recent AI answers (40 each on ChatGPT, Perplexity and Google AI Overviews), LoudFace was named 17 times, and every one of those answers had retrieved a loudface.co page. ChatGPT retrieved one of our pages in 24 of its 40 answers and named us in 7. In that sample, LoudFace was never named in an answer that had not retrieved one of our pages, and an ad bid has no effect on which pages an answer retrieves.

There is a second gap: who sees each one. OpenAI says ads may appear for users on the Free and Go plans, and it also shows ads in logged-out ChatGPT. Plus, Pro, Business, Enterprise and Edu accounts do not get ads. If your buyer shortlists vendors inside their company's ChatGPT Business or Enterprise workspace, or pays for Plus, they never see your ad, but they can still see your brand named in an answer. A Free user can also switch to an ads-free mode with lower message limits.

Ads also run only inside ChatGPT, while the same page that earns a ChatGPT citation can be cited by Perplexity and Google AI Overviews, which trust different corpora. In our category, the domain ChatGPT cited most was loudface.co, followed by listicle sites, arxiv.org and reddit.com, while Google AI Overviews leaned on YouTube, LinkedIn and Reddit.

When ChatGPT ads make sense for B2B SaaS

I am not against the channel. It is new, it is cheap to test, and conversational intent is real intent. Ads make sense when most of these are true:

• You have a self-serve conversion. A free trial, a template, a calculator or a signup. The Clicks and Conversions objectives are built for this, and reporting covers impressions, clicks, spend, CTR, average CPC, average CPM and conversions.
• Your buyer skews to free accounts. Founders at early startups, freelancers, students heading into a role, SMB owners. Enterprise buyers on company workspaces are out of reach.
• You already run HubSpot. The new integration lets you tie ad clicks to contacts and deals where your pipeline already lives, which is the only way to judge this channel on revenue instead of clicks.
• You are already in the answer, or close to it. Seer Interactive reports that when paid ads ranked alongside organic results for a brand it works with, organic CTR went up 156%. Seer does not say that figure came from ChatGPT. I would expect the same logic to carry over: an ad under an answer that already names you reinforces it.
• You need reach this month. Once a campaign clears review, impressions can appear within minutes. Nothing organic moves that fast on a contested prompt.

What a sensible test looks like: one campaign, the Clicks objective, the $3 to $5 starting bid OpenAI suggests, a daily budget you would not miss, UTM parameters on every landing URL, and a fixed end date. Judge it on qualified pipeline. CTR tells you little on its own, so compare the test against your own pipeline data.

When earned citations win

For most B2B SaaS teams, most of the time. Four reasons.

The recommendation is the answer. A buyer asking ChatGPT to compare vendors reads the answer first. If the answer recommends a competitor and your ad sits underneath, you paid to be the second opinion. Seer's advice before approving any LLM ad budget is to run the exact prompts you plan to bid on and read what comes back about your brand, and I agree with it. If the answer ignores you or frames you badly, fix that first.

Citations reach the accounts ads cannot. Business and Enterprise workspace users never see ads. They do see answers.

Ads stop when the budget stops. Turn off a campaign and the impressions stop the same day. In our experience, a page that earns a steady slot in the sources keeps getting cited, though no engine guarantees that slot will survive its next re-evaluation. The results we publish show the curve. Our own AI visibility went from 0.18% in April 2026 to 12.42% in August 2026, measured in Peec AI, and ChatGPT went from never naming LoudFace to 15.4% of answers. Prospects now find us through ChatGPT recommendations, and we booked 2.0x more sales calls than in the months before August. We were Toku's growth partner for 18 months, and the program reached 97.8% visibility on the category's top stablecoin payroll prompt in August 2026. A stealth payments company went from 0.53% to a 10.46% peak in seven weeks.

Speed is better than people assume. A well-structured page from a brand with some existing authority can be picked up by Google AI Overviews and Perplexity within 24 hours of publishing. The slow part is holding a dominant share on a contested prompt cluster, which takes months. That is the same patience you would give a new paid channel to exit learning mode, except that, in our experience, the result does not vanish when the budget does.

The weakness of earned citations is control. You cannot buy a guaranteed slot tomorrow, and the work needs to reach both your own pages and the third-party lists and threads each engine reads. If that sounds like a lot, our playbook on getting cited in ChatGPT covers the pages side, and the AEO guide covers the fundamentals.

How to run both without wasting spend

If you do both, sequence them. The mistake I would bet on seeing most in 2026 is a team buying ads for prompts where the answer already recommends a rival.

1. Read the answers before you write the ads. List the 10 to 15 prompts your buyers actually ask: "best [category] for [segment]", "[you] vs [competitor]", "[category] pricing". Run each one in ChatGPT, several times, and write down who gets named and which pages are cited. An AI visibility audit does this across engines and against competitors.

2. Split the prompts into three buckets. Prompts where you are named: candidates for a small ad test, since the ad reinforces the answer. Prompts where you are absent but a page of yours could fit: organic work first. Prompts where the answer is negative or wrong about you: fix the sources, do not advertise.

3. Fund the organic work with the bigger share. Your pages need to answer the prompt in the first screen, with a table or a clear verdict an engine can lift. Your brand also needs to appear on the third-party pages each engine already cites. That is the work a GEO agency does, and in our view it is where the more durable result is.

4. Cap the ad test and give it an end date. Clicks objective, a starting bid in OpenAI's $3 to $5 range, a daily budget at or near the minimum, and four to six weeks. Tag every landing URL. If you run HubSpot, connect the account so ad clicks land on contact records.

5. Measure both on pipeline. For ads: qualified pipeline per dollar, from UTMs and CRM data. For citations: AI visibility and named mentions on your target prompts per engine (we track ours in Peec AI), then the branded searches, demo requests and signups that follow. Impressions and clicks on their own never tell you whether a deal came in.

6. Re-check the answer every month. An ad leaves the answer alone. Your organic work changes it. As your pages start getting cited, move ad budget toward the prompts where the answer now names you.

What I would do with a B2B SaaS budget this quarter

If you sell to mid-market or enterprise teams, put the money into earned citations and skip ads until you are named on your core prompts. Your buyers sit on workspace plans that never see an ad, and the ad pages I read offer no firmographic targeting to reach them anyway.

If you sell self-serve to founders, SMBs or prosumers, run both, with the organic work holding the larger share and a capped, tagged ad test on prompts where the answer already mentions you.

In both cases, look at the answer first. You can check where your brand stands across ChatGPT, Claude, Gemini and Perplexity against your top competitors before you spend a dollar on either side.