LinkedIn
If an agency deck claims “three years managing ChatGPT ads at scale,” stop at that slide. OpenAI only started testing ads in the US on Feb. 9, 2026 and opened its self-serve beta on May 5. That is not a long track record. It is a timeline problem dressed as a credential.
ChatGPT ads are new, the reporting is thin, and semantic matching does not behave like Google keyword targeting. The honest pitch is a small, approved-budget test with tracking set up first, run by someone who will watch what happens rather than hand you a benchmark table. Here are the seven lines I would challenge before signing anything.
Myth 1: “We have a long ChatGPT ads track record”
People believe this because every agency deck needs a timeline. It feels safer to hire the team that says it got there first, especially when the channel itself is unfamiliar. But the dates leave no room for years of campaign results: US testing began on Feb. 9, 2026, and self-serve access opened on May 5. Nobody has years of ChatGPT ads data, seasonal patterns, or a stable playbook built from them.
A team may have paid-search experience. Good. Ask how much of its pitch comes from Google and how much comes from campaigns it has actually run in ChatGPT. Those are different answers, and a polished deck makes them easy to blur.
Ask when its first ChatGPT campaign launched and what it learned. If the answer stays at the level of logos and broad claims, you have your answer.
Myth 2: “It works just like Google keyword bidding”
The setup looks reassuringly familiar: a budget, a bid, and words describing what you sell. I used to tell clients to start with match types when they approached a new search campaign. That advice does not transfer neatly here.
ChatGPT ads have no keywords in that sense. You write one- or two-sentence context hints, and OpenAI matches them to conversations semantically rather than requiring literal word overlap. Its own guidance says hints guide matching but are not exact-match targeting rules. The conversation match determines where an ad might appear; your hint and bid then affect delivery. You cannot import a Google keyword list and assume the same intent will appear on the other side.
That difference matters most when a campaign surprises you. On Google, a search term can often show you the wording that brought someone in. Here, you should expect broader adjacency and less visibility into the conversation behind a click. A hint that sounds precise to a marketer may still reach people discussing a related problem rather than shopping for your particular offer.
Write hints for the intent you want, then judge the delivery you actually get. Do not call a familiar-looking setup a familiar auction.
Myth 3: “You need a big budget to learn anything”
This one has an understandable origin. The early pilot required a $200k minimum, so smaller advertisers heard about ChatGPT ads mostly through other people’s tests. Self-serve changed the size of the decision. It now starts at $25 a day, with a $3 to $5 suggested max bid in a second-price auction.
Say you spend $20k a month on Google. One ChatGPT campaign at $25 a day costs roughly $750 a month, under 4% of that spend. That is enough to learn whether your campaign delivers impressions and clicks, what CPC you pay, whether visitors engage, and whether your conversion tracking works. It is not enough to declare that you have found your long-term CPA, let alone a winning vertical strategy.
What a capped test can tell you
A small test should answer a short list of questions, in order:
- Does the tracking record the action you care about? Set up the pixel before spending. Check that a conversion means the same thing in your ChatGPT test and your other search reporting.
- Does the hint attract useful traffic? Look beyond the click to what visitors do after it. A cheap click that never reaches the form is not a bargain.
- Is there enough signal to continue? If conversions are thin, say so. Do not turn one week of noise into a confident CPA forecast.
The limits are real. In one independent test, $2,319 bought 739 clicks at a 1.06% CTR and $3.14 CPC with zero MQLs; Google produced two MQLs from the same offer. That does not prove ChatGPT ads cannot work. It shows why clicks alone cannot settle the question.
Approve the cap, fix the tracking first, and decide what would justify another round before the first dollar goes out. A small test is useful when you let it be small.
Myth 4: “Buying the ad buys a place in the answer”
Paid search trained people to associate paying with prominent placement. In a chat interface, it is easy to take that one step further and imagine the advertiser buying its way into what the assistant says. That is not what this placement does.
The ad appears below the conversation in a separate, labeled slot. OpenAI says ads do not change answers, and advertisers cannot see chats, only aggregate data. One system produces the answer; another places an ad underneath it. A bid affects the latter, not the former.
That distinction should shape the budget discussion. If you want an ad below the answer, test the ad. If you want your business cited inside the answer, work on the content and authority that might earn a mention. Paying for one does not do the other. An analysis also found that running ads did nothing for organic mention.
Keep paid placement and earned citation as separate jobs. Calling both “AI visibility” makes a tidy slide and a messy measurement plan.
Myth 5: “ChatGPT ads need their own separate agency from Google and Bing”
New channel, new specialist: I understand the instinct. Follow it far enough and you get a TikTok agency, a Shopping agency, and three invoices for one customer. The question is not whether someone knows where the ChatGPT buttons are. It is whether anyone owns the whole search budget and the outcome it is meant to buy.
ChatGPT Ads Manager reports familiar basics such as impressions, clicks, spend, CTR, CPC, CPM, and conversions through a pixel plus Conversions API. There is no MCC equivalent; you add an agency as a user rather than handing over a sub-account. A separate shop can manage that access, but separate management also makes it easier for each team to claim credit for the same customer journey or optimize its own channel without seeing the other one.
I am not arguing that every platform uses the same controls. It does not. I am arguing that one owner should see Google, Bing, and ChatGPT against the same conversion definitions and budget priorities. Otherwise, a ChatGPT report can look promising while the business still has no clear read on whether the additional spend helped.
Ask who resolves an attribution disagreement, who can move the budget, and who checks the account between calls. If the answer names three people in three shops, expect a meeting before you get a decision. Keep one accountable owner over the test.
Myth 6: “A $299 course gets you the same result as hands-on management”
People believe this because a course can teach a working campaign structure. On Google, that can take a newcomer a fair distance. ChatGPT’s problem is not just learning the interface; it is diagnosing delivery with limited visibility.
You cannot read the conversation that triggered your ad. A course instructor cannot read it for you, either. They can explain the controls, but they cannot inspect your live traffic, check your conversion setup, revise a hint that attracts the wrong visitors, and decide whether your offer deserves another week of budget. That work happens after the videos end.
The difference matters when early numbers look ugly. In one 1,500-user GA4 comparison, Google converted at 3.71% with 41 seconds of average engagement, while ChatGPT converted at 0.21% with 17 seconds. Those figures are a reason to investigate, not a universal forecast. A weak result might reflect tracking, the offer, the hint, or the traffic it draws. Lesson four cannot tell you which one applies to your account.
I broke down the choices in ChatGPT Ads Course vs Consultant vs Done-for-You: Who Actually Runs the Test?. A course teaches the interface. It does not run the test. Buy one if you want to learn. If you want someone accountable for a read on CPA, ask who sets up the pixel, watches delivery, and makes the next call.
Myth 7: “We beat the ChatGPT ads benchmarks in your vertical”
This is the hardest myth to kill. Founders want a number before they approve a test. Agencies know it, and a neat industry table calms the room. Google Ads has enough history that vertical comparisons can at least start a conversation. It is tempting to carry that habit into a new ad product.
But OpenAI says there are no performance benchmarks yet across advertisers, industries, or campaign types. Delivery still fluctuates as matching and ad load change. Your hint, offer, and tracking setup can move the result in ways a tidy “ChatGPT CPC by industry” slide cannot explain. The table may contain numbers; that does not make it a dependable target for your campaign.
Your first controlled test is your working benchmark. Record what you spent, what you paid per click, what visitors did, and what the test could not establish. Then decide whether the evidence earns more budget. Do not let a vertical average make that decision for you.
Three questions to ask before you sign
When every deck sounds confident, I would ask these before discussing a benchmark:
- When did you launch your first ChatGPT campaign, and what did an early capped test teach you? Listen for a date, an actual change the team made, and a limit it admits. “We have years of proprietary data” is not an answer.
- How will you track results alongside Google and Bing? Listen for a pixel, Conversions API, consistent conversion definitions, and a way to discuss assisted clicks without counting the same outcome twice.
- Who touches the account between our calls? You want a named human responsible for the decision and execution that does not wait for a weekly review. A course login and a Slack channel where you do the work are not management. I lay out that distinction in ChatGPT Ads on a Small Budget: Course, Consultant, or Hands-On Manager?.
Those questions also keep two purchases from getting mixed up. An ad below an answer is not an earned citation inside it; Can You Buy AI Visibility? Seven Questions Business Owners Keep Asking goes deeper on that measurement trap.
What I would run instead of a benchmark-driven rollout is small, approved, and hands-on: one capped test at $25 a day, consistent conversion tracking across Google, Bing, and ChatGPT, and regular checks on whether the traffic is useful. That is the approach groas offers. Search Ads covers Google, Bing, and ChatGPT under one $1,499 flat fee, with ongoing work on bids, budgets, and hints plus a named human who owns the result. No percentage-of-spend fee. No separate ChatGPT invoice. No course homework disguised as management.
If the test buys clicks that leave after 17 seconds and never fill a form, pause and put the money where it is working. If it produces useful assisted traffic, increase the budget carefully. Neither decision requires an invented industry benchmark. That is precisely why the benchmark table is so hard to kill: it offers certainty before anyone has done the work.