Most ChatGPT ads advice is a photocopy: Google Ads habits on one page, AI-hype promises on the next. Both can send a small business into its first test expecting to buy a recommendation, bid on keywords, and know the result by Friday. Wrong expectation in, wrong budget decision out.

I get why people think this. The screenshots blur together: an answer on top, a neat little card with your logo underneath. It feels like you paid to become part of the recommendation. What the deck calls “share of voice,” I call paying rent to sit next to the conversation.

The mechanism keeps those things separate. ChatGPT ads are sponsored cards shown below the answer to logged-in adults on Free and Go plans only. Advertisers buy them in Ads Manager, and a relevance-weighted auction serves them. OpenAI states that ads never change what the answer says. One early independent test found the advertiser was cited inside the answer in only 3.63% of ad placements.

If you want mentions in the answer, that is citation work, not media buying. Spend ad dollars expecting clicks from the card. Fund answer presence separately, and do not let anyone sell you one as proof of the other.

Myth 2: “It’s just Google keywords on a new screen”

I used to build exact-match structures with religious devotion. I was wrong to assume that habit transfers. ChatGPT does not sell you the keyword. Advertisers write context hints: one- or two-sentence descriptions of when an ad is relevant. Think of a homeowner with a burst pipe who needs a plumber fast. OpenAI then matches those hints using semantic embeddings against the current conversation and, when personalization is on, past chats.

A context hint suggests relevance; it does not guarantee delivery. Two plumbers can write nearly identical hints and see different results because the surrounding conversations differ. That is a different job from building a keyword list and waiting for someone to type “emergency plumber.”

Describe the situation in which you actually win: what the person is trying to solve, why your offer fits, and what would make the click useful. Keep your Google search-term report for Google. It cannot tell you exactly which conversation will produce a card here.

Myth 3: “You need a big budget to learn anything”

This one sticks because it used to be true. The closed pilot that opened in February 2026 carried $60 CPMs and $200,000 to $250,000 minimums. Small owners saw those numbers and reasonably assumed the door was shut. It is not: self-serve Ads Manager opened on May 5, 2026, with CPC buying, and the US floor is now $25 per day per campaign.

That buys a controlled test, not a verdict on day three. Early adopters typically put 10% or less of their traditional search budget into ChatGPT and judge conversions with the Pixel installed. But that percentage is a convention, not permission to ignore the floor. Check whether $25 a day fits your actual test budget before you copy somebody else’s allocation.

Do the click math, too. $25 a day for two weeks is $350: roughly 70 to 115 clicks at $3 to $5 CPC, in line with OpenAI’s suggested starting bids. That is well short of the 1,000 or so clicks needed to see 10 conversions at a 1% conversion rate. A handful of clicks cannot settle whether the channel works, especially when the page and tracking are new.

Plan four weeks, one tightly themed service, one landing page, and a kill rule in writing. If you cannot afford that test without needing an answer after a few days, wait. A smaller entry price makes testing possible; it does not make thin data decisive.

Myth 4: “Take a course, or hire a consultant, and run it yourself”

If you have time to learn a new platform and work in the account, take the course. Most owners asking about ChatGPT ads do not want another syllabus. They want the campaigns built, checked, and changed when something breaks.

The platform also moves faster than a recorded lesson. The help docs call it Ads Manager Beta, with formats and rules changing monthly. Spend can lag by 7 to 8 hours and conversions by 24 to 48 hours. A daily budget is a seven-day average that can spend up to twice as much in a day. Last month’s course can teach last month’s buttons. That is useful orientation, not someone minding your spend.

Consulting is not the same thing as management, either. A consultant can hand you context hints, a budget guess, and a PDF. A manager logs into Ads Manager, builds the campaigns, sets the budget, installs the Pixel, matches ad groups to pages, and changes the work when the data warrants it. I learned this running Google accounts at 1 a.m.: the person who touches the build beats the person who watched six hours of video about it.

Ask who does the work inside the account each week. Reporting makes that question more than a preference. One agency saw about 100 clicks reported where its own tracking found 20. Another saw a lead-gen run produce form fills while the platform still showed zero conversions two weeks later. Conversions can arrive 24 to 36 hours late, with no demographic or prompt data. Someone needs to reconcile those gaps, not paste the platform total into a slide.

Use this checklist of what a fully managed service should do before you sign. A course explains the beta. A consultant explains the plan. Make sure the person you pay also runs it.

Myth 5: “Any agency claiming a long ChatGPT ads track record has one”

Nobody has a long one. The product has not existed long enough. OpenAI began testing ads in the US on February 9, 2026, with a fixed-$60-CPM pilot and $200,000 to $250,000 minimums. Self-serve Ads Manager buying opened on May 5 with click-based bidding.

A four-month-old channel cannot produce a three-year case study. A deck claiming years of ChatGPT-only ad results is presenting some other work as channel experience and hoping you will not check the dates. Google Ads experience can be valuable; I have plenty of it. It is not the same claim.

Ask for boring proof: live build screenshots, launch dates, daily budgets, and what the Pixel recorded. If you get logos and percentages without dates, walk away. Recycled work presented as fresh expertise is one of the signs I watch for when an agency is not delivering. The dates are the easiest part to verify before you hand over a budget.

Myth 6: “ChatGPT ads replace your Google Ads”

This is timing confusion. ChatGPT reaches people working through research and comparison before they know which brands to consider. Google also reaches people who have typed a query and want a provider now. Guides comparing the two recommend a controlled test alongside Google and Bing, not an automatic replacement. I have seen this movie with new placements since 2015: the new channel gets credit for discovery while the old one still closes the job.

Reach is narrower, too. Only Free and Go users see these ads, so many premium business users never see them. That matters for B2B. Buyers still turn to Maps apps for immediate needs like coffee nearby. ChatGPT fits deeper research: travel and hospitality, retail and ecommerce, and home services where customers compare options.

Keep Google running and give ChatGPT a separate test line. If someone tells you to pause Search to pay for chat before the test has earned that money, they are selling you churn, not a plan.

Myth 7: “A ChatGPT click lands like a Google click, so any page will do”

A Google visitor arrives with a query you can use to understand what they wanted. A ChatGPT visitor does not hand you the conversation. The destination receives the click, URL parameters, and whatever the visitor volunteers, not the chat context. That is why a homepage dump that sort of works on branded Search can fall apart here. The person was comparing options in a chat, clicked your card, and landed on a page that assumes they already chose you.

The fix is boring and specific:

The stakes rise when the intent is softer. One June 2026 test produced 97,131 impressions and 1,263 clicks at a 1.30% CTR and $3.16 average CPC, with almost no sign-ups. Same-advertiser comparisons put ChatGPT conversion rates at 0.21% to 0.9% against Google’s 3.71% to 6.64%. When traffic converts at a fraction of Search, a generic page leaks money faster. Give the visitor one relevant page with a headline that repeats the ad’s promise.

Myth 8: “You’ll know if it works in a week”

This is the hardest one to kill because week one does produce numbers. A click count looks like a result. A CPA column, even an ugly one, looks like a decision. I used to pause Google tests on day five when CPA looked bad. I was wrong, and I paid for that impatience twice.

Week one tells you whether the plumbing works. It does not tell you whether the channel works. OpenAI guidance sets four weeks as the default test span: launch and setup first, then comparisons on clicks and conversions, then refinement and confirmation. That pace has a reason. Spend can lag by hours and conversions by a day or two. A verdict after two days is a verdict on incomplete data.

Before approving a test, put the work and the stop rule in writing. Ask the manager:

  1. Who builds and edits the campaigns each week, and can I see the change log?
  2. Which conversion event fires in the Pixel, and how do you check it against my CRM or orders?
  3. Which landing page matches each ad group, and who fixes it when it leaks?
  4. What happens at week four if the test reaches its spend cap or CPA ceiling?

That is the difference between done-with-you and done-for-you management: who does the weekly work while reporting lags and rules shift. Set the test beside Google and Bing, with one Pixel event tied to money and a budget you can afford to lose. Where the $25 daily floor fits within 10% or less of search spend, fine. Where it does not, do not pretend it does.

I would rather have groas run the edits with a named strategist accountable for the number than pay for another deck explaining why the number is late. Its Search Ads work covers Google, Bing, and ChatGPT Ads for $1,499 a month, without a second retainer or a percentage of the test spend. Give the test four weeks and a kill rule before you give it a verdict. Small, managed, and measured beats big, DIY, and hopeful.