ChatGPT Ads, run by operators
Since February 2026, brands can buy a labeled placement beneath a ChatGPT response, arriving at the moment someone describes exactly what they need. QueryBid does one thing: plan, launch, and scale those campaigns. Eligibility, structure, creative, and the measurement almost everyone gets wrong.
Self-serve since May 2026. No minimum, roughly $25 a day to start. The catch nobody leads with: whether you may buy at all depends on your category, and that list is shorter than most agencies admit.
A labeled card below the answer, visually separated from it. This is the unit we buy and build. All four examples sit in categories OpenAI currently permits.
What QueryBid does
QueryBid is an advertising agency that manages ChatGPT Ads for brands and manufacturers. Category eligibility, account structure, campaign strategy, creative, product feeds, and the conversion tracking that decides whether any of it can be measured at all.
We deliberately do not spread across six channels. ChatGPT Ads is months old, the rules change monthly, and depth on one surface beats a shallow pass at all of them. If a different surface is the better buy for you, we will say so and point you at it rather than take the retainer.
The name is literal. Every ChatGPT answer starts with somebody's query, and the placement beneath it is won at auction. We bid on your behalf.
Based in Austin, Texas. Serving clients across the United States. Every platform fact on this page last verified against OpenAI's published ad policies and advertiser documentation.
The unit
It is not a search ad and it is not a social ad, and treating it like either is the most common way early budgets get wasted. Here is the specification, plainly, before anyone asks you for money.
| Attribute | Status | What it means | Why it matters to you |
|---|---|---|---|
| Format | Sponsored card | A labeled card carrying a brand, a headline, a line of copy, and a link. A product carousel format pulling multiple items from retailer feeds is rolling out alongside it.Not a text link, not a banner, not a native paragraph woven into the response. | You are writing an offer, not a keyword ad. The headline does almost all the work, and there is no ad rank trick that rescues weak copy. |
| Placement | Below the answer | The card sits underneath the response, visually separated and clearly marked as sponsored.OpenAI has been explicit that ads do not change what the model says. The answer is written first, independently. | You cannot buy your way into the recommendation itself. You buy the slot after it, which means your ad has to survive being read second. |
| Audience | Free and Go | Ads show to Free and Go tier users. Plus, Pro, Business, Enterprise, and education accounts do not see them.Consumer reach, not enterprise reach. | If your buyer is a senior professional on a paid work account, they are outside this inventory. That is a reason to test small, not a reason to skip. |
| Targeting | Contextual | Topic, intent, and conversation context rather than a keyword list. Audience Tools allow first-party customer data upload and suppression lists.Relevance weighted auction, so creative quality affects ranking alongside your bid. | There is no keyword research phase and no exact match. The work is mapping the questions your buyers actually ask, then writing to the moment. |
| Geography | Expanding | Advertiser accounts open in the US, Canada, Australia, and New Zealand. Ads serve in those markets plus the United Kingdom, Mexico, Brazil, Japan, and South Korea.Five markets were added on August 11, 2026 alone. | Where you can buy from and where you can reach are two different lists, and both are moving. Confirm both before committing to a plan. |
| Measurement | Bring your own | A JavaScript pixel and a server-side Conversions API, plus AppsFlyer and Adjust integrations for app advertisers.Neither surface is optional in a serious build. See the measurement section below. | This is where most early accounts quietly fail. Attribution is not automatic, and a misconfigured install produces confident numbers that are wrong. |
Scroll the table sideways for the full detail.
One honest note on the rest of the market. Google sells placements in AI Overviews and AI Mode, and Microsoft sells them beneath Copilot answers, so both are real options. Claude, Perplexity, and the Gemini app carry no advertising at all, and anyone offering to sell you ads there is selling something that does not exist. We concentrate on ChatGPT because it is the newest, the least understood, and the one where a few months of reps still counts for something.
What ChatGPT ads cost
OpenAI has not published performance benchmarks by industry, so nobody can honestly tell you what a good result looks like in your category yet. Anyone who quotes you a target return on ad spend for this channel is guessing. What we can tell you is what it costs to get in and how the auction works.
Minimum commitment
None
The $200,000 pilot minimum was removed when self-serve opened in May 2026.
Daily budget floor
~$25
Per campaign, per day. Testing is genuinely cheap right now.
Reported cost per click
$3 to $5
Entry range reported by early advertisers. Varies widely by category.
Reported CPM
$25 to $60
Higher than social, against a far more specific moment of intent.
Figures reflect publicly reported ranges from early advertisers as of August 2026, not official OpenAI rate cards. Auction pricing is relevance weighted, meaning creative quality affects ranking alongside your bid, so a sharper ad can win placement at a lower cost. Prices and platform rules change often. We re-verify this page monthly. Official platform documentation: openai.com/advertisers.
What we run
ChatGPT Ads does not behave like Google Ads or Meta. There is nothing to bid on by keyword, the creative is a card rather than a text ad, and the measurement is a build rather than a checkbox. Seven pieces of work, all of them on one platform.
Before anything else, we confirm your category can actually buy, and where approval is required we prepare the application and the substantiation rather than promising a launch date we cannot control. Then account structure, billing, and campaign architecture inside the OpenAI Ads Manager.
Conversions API and pixel, deduplicated, with event taxonomy mapped to your real outcomes and signed off before launch. We validate against the platform in test mode, then with a live click, before we spend anything. This is the work that makes every later decision trustworthy.
There are no keywords here. Targeting is topic, intent, and the moment in a conversation. We map the questions your buyers actually ask, identify the point where they would be glad to hear from you, and mark the conversations where you should not appear at all.
Your ad sits directly under a genuinely useful answer, so it lives or dies on the brand line, the headline, and the offer. We write, test, and rewrite. The auction is relevance weighted, which means sharper creative can win the placement at a lower cost.
The carousel format pulls from retailer feeds, so for retail and DTC the feed is the campaign. We manage it so the right product appears against the right question, priced correctly and in stock. A great placement pointed at a dead SKU is wasted money.
For lead generation, the click is not the outcome. We feed closed-won results back as conversions on a nightly cadence, inside the seven-day window the platform accepts, so optimization runs against revenue rather than form fills.
A consistent report you can actually act on, with view-through kept out of the headline. Plus a standing watch on the policy changelog, because category rules have changed four times since March and the announcements come after the fact, if at all.
Why QueryBid
Specialists, not generalists
We have spent real budget inside ChatGPT Ads since the platform opened, and we have the losing tests to prove it. That head start is the entire reason to hire us now rather than in a year.
Operators, not advisors
We own the numbers and run the campaigns ourselves. You are not buying a strategy deck to go implement on your own after the call ends.
Profit over vanity
We manage to cost per acquisition and payback period, not reach. If a placement does not make money, we cut it, even when the dashboard looks impressive.
A direct line
Short emails, clear calls, no account manager relay. When a platform changes its rules overnight, you hear it from the person with their hands on the account.
Who it fits
Most write-ups rank categories by how well they perform and skip the part where OpenAI simply will not accept your money. During the current test period the eligible list is short and specific. Here is the version with the gate included, which is the version that decides your quarter.
| Category | Can you buy? | Fit if eligible | What we would do |
|---|---|---|---|
| Local services | Eligible | Strong | Cost and how-to questions come first, the hire comes second. Region, DMA, and postal targeting map cleanly onto a service area. Run it now. |
| Household and consumer goods | Eligible | Strong | The core permitted vertical and the one the format was designed around. Feed quality does most of the work. |
| Travel and experiences | Eligible | Strong | Long, descriptive planning questions with a real budget attached. One of the best matches on the platform. |
| Digital products and education | Eligible | Strong | Comparison cycles run almost entirely through research questions. Creative carries more weight here than bid. |
| Considered e-commerce | Eligible | Strong | Product research converts well with a clean catalog behind it. The carousel format makes this better every month. |
| Business software | Eligible | Workable | Real demand, harder attribution, and much of the buyer base sits on paid tiers that never see ads. Worth a contained test rather than a headline budget. |
| Financial services | Restricted | Case by caseApproved advertisers only, manual review, US market. Credit repair, debt settlement, and alternative investments stay prohibited. | We prepare the application and the substantiation before we promise you a launch date. Nobody gets to skip the queue. |
| Healthcare and medicine | Restricted | Case by caseApproved advertisers only, manual review, US market. Named eligible areas include insurance, dental, vision, supplements, testing, and hospitals. Ads still never serve into personal health conversations. | Same process, plus a claims review on the creative and the landing page. Pharma and clinical treatment claims are not on the table. |
| Legal services | Not eligible | No paid routeAds for legal advice, representation, or legal services are not permitted, and the policy names immigration, personal injury, legal claims, and document preparation directly. General legal education offering no services is the only opening. | We will say so on the first call and decline the engagement. If an agency is pitching your firm a ChatGPT ad buy today, ask to see the written approval. |
| Impulse and low-cost goods | Eligible | Weak | Nobody consults an assistant before a $12 purchase, and current CPMs do not clear the margin. Allowed, but we would talk you out of it. |
Scroll the table sideways for the full detail.
The distinction that trips everyone up
Placement eligibility and advertiser eligibility are two different things. In April, OpenAI stopped categorically blocking ads from appearing near medical, legal, and financial advice conversations. That governs where other advertisers' ads can show. It did not make law firms, clinics, or lenders eligible to be the advertiser. A lot of pitches this year have quietly conflated the two, and a personal injury firm can end up paying a retainer against inventory it cannot legally buy.
Not sure which row you are in? Send us the category and we will tell you within a business day.
Check my categoryCategory rules are the fastest-moving part of this channel. Eligibility has changed four times since March 2026, always through quiet edits to the policy changelog rather than announcements. We re-check before every launch and again before every renewal. Source of record: OpenAI ad policies. Our full write-up: which categories can advertise on ChatGPT →
What changed this quarter
Nothing below was announced with a press release. All of it came from policy changelogs, documentation updates, and account-level changes we watched land. This is what paying attention to one channel looks like.
Aug 11, 2026
Five new markets. ChatGPT Ads live in the United Kingdom, Mexico, Brazil, Japan, and South Korea, with more expected before year end.
Jul 15, 2026
Health and finance loosen. A policy update spelled out which financial and health subcategories may qualify, added enhanced verification and credential proof, and kept both restricted to the US market.
Jun 2026
Legal confirmed closed. The ad content policy explicitly named legal advice, representation, immigration, personal injury, claims, and document preparation as not permitted, after a month in which several agencies had begun pitching law firms.
Jun 17, 2026
Ad Tools Terms published. Audience Tools for first-party customer data upload and suppression lists, and Creative Tools for generating and localizing creative from your own catalog and site. Policy-announced ahead of the feature.
Q2 2026
App attribution arrives. AppsFlyer and Adjust integrations brought install, re-engagement, and in-app event attribution to the channel for the first time since the pilot opened.
Q2 2026
Product carousel. A format showing multiple products in one placement, pulled from retailer feeds. This is the change that moved e-commerce from interesting to worth a real budget.
Apr 2026
Placement rules refined. Medical, legal, and financial advice contexts stopped being categorically blocked from carrying ads. Widely misread as opening those categories to advertisers. It did not.
May 5, 2026
Self-serve opens. The $200,000 pilot minimum removed entirely, 86 days after launch. Daily budgets from roughly $25.
Measurement
A campaign that cannot be measured cannot be improved, and this platform makes it easy to install tracking that looks correct and reports nonsense. We run the same eleven-point gate on every account before a dollar goes live, because the failures below are silent and cost weeks to find.
The architecture we build every time
Conversions API as the system of record, JavaScript pixel alongside it, both sending the same deduplication ID. Pixel-only loses events to ad blockers and script failures. API-only forfeits automatic click-identifier capture and the browser reference that hybrid matching depends on, which quietly degrades your match rate. Anyone who installs one and calls it done has left money on the table and will not know it.
| Symptom | Usual cause | Consequence if missed |
|---|---|---|
| Conversions recorded, none attributed | The click identifier stripped by a redirect, a link shortener, a consent gateway, or a geo-router before the landing page loads. | The campaign looks like a total failure. Budget gets cut from something that was working. |
| Conversion counts roughly doubled | Deduplication IDs that do not match across the pixel and the server, or a standalone image tag firing alongside the pixel. | Reported cost per acquisition is half of reality. You scale spend into a loss. |
| Match rate far below expectation | A normalization error producing valid-looking hashes that never match anything, or the browser reference never forwarded to the server. | Worse than sending nothing, because everything appears to be working. |
| EU traffic converts but never reports | Consent set after initialization instead of before it. Blocked events are never replayed once consent is granted. | Those conversions are permanently gone. There is no backfill. |
| An entire batch rejected | One malformed event in the batch, or a timestamp outside the seven-day window the API accepts. | Silent data loss, and monthly CRM exports never land at all. |
| Events fail silently on a hardened site | A Content Security Policy missing a required directive, most often the one governing script elements specifically. | Zero data from your most security-conscious client's site, with no error anyone sees. |
Scroll the table sideways for the full detail.
What we will not do in a report
View-through conversions stay out of the headline number. They sit in a fixed one-day window after an ad impression, they are excluded from the platform's own Conversions metric, and cost per acquisition, bidding, and billing all remain click-based. In a QueryBid report, view-through is directional color in a secondary section, never the number the engagement is judged on. We put that in writing at kickoff, before the first report lands, so nobody is surprised by an honest number later.
Offline and CRM outcomes can be fed back as conversions, which matters for every lead-generation client. The constraint that dictates the build: event timestamps must fall inside the last seven days. That makes a monthly CRM export worthless and a nightly sync mandatory, and it is the kind of detail that is far cheaper to design around than to discover in month two. Read the full measurement walkthrough →
How it starts
STEP 01
We map where your buyers already ask questions, which platforms sell that attention, and what a placement there is realistically worth to you. If the math does not work, we say so.
STEP 02
We build the accounts, set up tracking, write the first round of creative, and ship live campaigns, usually inside two weeks of a signed start.
STEP 03
We cut what does not convert, pour budget into what does, and move onto each new platform the week it opens its doors to advertisers.
Who runs it
Will Tjernlund
Founder
QueryBid exists because a genuinely new advertising channel opened in February 2026 and almost nobody moved. The same thing happened with Amazon around 2013, and the brands that got in early spent the next decade at an advantage that later entrants never quite closed.
The work is not complicated, but it is unfamiliar. There are no keywords. Creative quality changes what you pay. Half the assistants people use will not sell you a placement at any price. Somebody has to sit inside these accounts every day, run the tests, and report honestly on what is and is not working.
That is the whole company. No layers, no relay, no deck that ends in a proposal to build another deck.
Notes
We publish what we find. Not thought leadership, just the operational detail that is cheap to design around and expensive to discover in month two. Every platform claim is dated and re-checked monthly.
The eligible list is much shorter than the coverage suggests, three categories sit behind manual approval, and one is closed outright. Here is the whole thing, with the distinction that keeps costing advertisers money.
ChatGPT Ads / Policy · 9 min read
The pixel alone is not enough, the Conversions API alone is not enough, and the failures are silent. A walkthrough of the architecture, the seven-day rule that dictates your build, and the six ways accounts quietly break.
ChatGPT Ads / Measurement · 11 min read
Some setup choices lock permanently at confirmation, negative keywords are capped far lower than you expect, and a landing page that blocks OpenAI's crawler will not serve. What to get right on the first pass.
ChatGPT Ads / Operations · 7 min read
Questions
The honest state of the channel, including the parts that do not help us sell anything.
Yes. OpenAI began showing ads inside ChatGPT on February 9, 2026, and opened a self-serve Ads Manager to US advertisers in May 2026.
Ads appear as clearly labeled sponsored placements alongside responses for people on the Free and Go tiers. Access started in the United States and has since expanded to additional markets. Anyone with a business, a tax ID, a billing method, and a landing page can open an account.
Maybe not. During the current test period OpenAI limits ads to a short list of consumer verticals: household and consumer goods, local services, travel and experiences, and digital products and education.
Financial services and healthcare sit behind approved-advertiser status, granted one at a time through manual review and limited to the US market. Legal services are not permitted at all. Everything outside those lists is disallowed for now. Check the eligibility table above before you plan a budget around this channel, because the gate matters more than the forecast. Full category breakdown →
There is no minimum spend commitment, campaign daily budgets start around $25, and reported entry costs run roughly $3 to $5 per click or $25 to $60 per thousand impressions.
Placements are awarded through a relevance weighted auction, which means creative quality affects ranking alongside your bid, so a sharper and more relevant ad can win at a lower cost. OpenAI has not published official performance benchmarks by advertiser or industry, so early results vary widely. Treat any agency quoting you a guaranteed return in this channel with suspicion.
It is early, which is the argument for a controlled test rather than either a large commitment or a wait.
The honest case for going now: there is no minimum, daily budgets start around $25, the auction is thin compared to search, and reported entry costs per click sit in the $3 to $5 range. The honest case against: OpenAI has published no industry benchmarks, so nobody can tell you what good looks like in your category, and ads reach Free and Go tier users rather than people on paid work accounts. Our recommendation for most eligible advertisers is a budget you would be comfortable writing off entirely, spent properly with tracking built first, treated as buying information rather than buying revenue. If your category is not eligible, the question does not arise yet.
No. OpenAI's ad content policy states that ads for legal advice, representation, or legal services are not permitted, naming immigration, personal injury, legal claims, and document preparation directly.
The confusion comes from an April change that let ads appear near legal-advice conversations. That governs placement, meaning which conversations can carry an ad. It says nothing about permission, meaning who is allowed to run one. Both things can be true at once: a competitor's travel ad can appear beside someone describing a car accident, and your firm still cannot buy that slot. The one narrow opening is general legal education that offers no services, and OpenAI reviews the landing page as well as the creative, so routing an educational ad to an intake form does not survive review. Until that changes, GEO and your existing search program are the honest answer. Why legal is closed, in detail →
Among categories that can actually buy: local services, travel and experiences, education and digital products, and considered e-commerce.
All four share the same trait, which is that buyers ask long, descriptive questions before deciding. Business software works but attribution is harder, since those buyers research across many channels before converting. Impulse and low-cost consumer goods are the weakest fit, because current CPMs do not clear the margin on a small basket.
Targeting is contextual. Placements are matched to the topic and intent of the conversation someone is already having, not to keywords you bid on.
Signals include the current conversation, prior chat context, and previous ad interactions. Advertisers can target and exclude by country, region, designated market area, and postal code. In practice this shifts the work away from keyword lists and toward understanding the actual questions your buyers ask, then writing creative that belongs in that moment.
Logged-in adult users on the Free and Go tiers in eligible countries.
Subscribers on Plus, Pro, Team, Business, Enterprise, and education plans do not see advertising at all. OpenAI has also said it does not serve ads to accounts it predicts belong to people under 18. This shapes the audience meaningfully: you are reaching the free tier, which is the large majority of users but skews away from the heaviest paying professional users.
No. OpenAI states that advertising runs on systems separate from the model and does not alter the response it generates.
Placements are labeled as sponsored and sit alongside the answer rather than being written into it. You are buying adjacency to a moment of intent, not influence over what the assistant says. That distinction matters, and it is worth understanding before you set expectations internally.
Advertiser accounts are open in the United States, Canada, Australia, and New Zealand. Ads serve in those markets plus the United Kingdom, Mexico, Brazil, Japan, and South Korea.
These are two separate lists and they are frequently confused. Where you can open an account and where your ads can reach people are governed differently, and both have expanded repeatedly during 2026, most recently with five markets added on August 11. Confirm both against current documentation before you build a plan around a particular country.
With a server-side Conversions API as the system of record and a JavaScript pixel alongside it, both sending the same deduplication ID.
Neither alone is sufficient. A pixel-only build loses events to ad blockers and script failures. An API-only build gives up automatic capture of the click identifier and the browser reference that hybrid matching depends on, which degrades your match rate without any visible error. We map your real conversions to the platform's event taxonomy, agree the deduplication convention on both sides, validate in test mode, then confirm with a live click from an actual ad before any budget goes live. For lead generation we also sync closed-won CRM outcomes back as conversions, nightly, because the platform only accepts event timestamps from the last seven days. The full tracking walkthrough →
No. We run ChatGPT Ads only, and we would rather be the best option on one surface than an average one on four.
For completeness, since it comes up on every call: Google sells placements inside AI Overviews and AI Mode through Google Ads, and Microsoft sells them beneath Copilot answers through Microsoft Advertising, so both are legitimate buys. Claude, Perplexity, and the Gemini app carry no advertising at all. Perplexity retired its ad program in early 2026 to protect trust in its citations, Anthropic has never sold ads in Claude, and Google places ads in Search rather than in the Gemini assistant. If someone offers to sell you placements on those three, they are selling something that does not exist. Where another surface is genuinely the better fit for you, we will tell you and point you to it.
These platforms do not behave like Google Ads or Meta, and most search and social teams have not yet spent meaningful budget in them.
There are no keywords to bid on, the ad unit is a labeled card inside a generated answer, creative quality directly affects auction ranking, and native benchmarks barely exist. We work only in this channel and coordinate with the search and social teams you already have rather than trying to replace them.
Most accounts launch within two weeks of a signed start.
Week one covers the audit, account structure, and tracking setup. Week two covers creative and launch. Expect meaningful optimization signal in the first 30 to 45 days of spend, which is when we start cutting and scaling with real data instead of assumptions.
A flat monthly management fee, billed month to month with 30 days notice, separate from the media budget you pay platforms directly.
No long-term contracts and no percentage-of-spend arrangement, because that structure rewards an agency for spending more of your money rather than making more of it. Reporting is monthly, with a direct line to the people running your accounts in between.
Austin, Texas, working with clients across the United States.
These platforms are managed entirely through web-based ad interfaces, so engagements run remotely and geography does not limit who we can work with.
Start here
Send us your category and we will tell you within a business day whether OpenAI will accept you as an advertiser today, whether you need to go through approval first, or whether ChatGPT Ads is closed to you for now.