A garage-door shop with $750 for ChatGPT ads does not need to prove an entire channel works. It needs to find out whether people asking about the job it wants most turn into leads worth paying for. A small budget can answer one sharp question. The businesses wasting money are the ones trying to answer five questions with that same budget.

Myth 1: “You need a big budget to learn anything”

People believe this because Google taught them that learning needs 30 conversions in 30 days. At $50 a click, that math looks hopeless. But ChatGPT ads start at $25 per campaign per day, with OpenAI suggesting a $3 to $5 opening bid. Say you spend $1,200 in a month at $4 a click. That buys about 300 clicks. At a 4% click-to-lead rate, that is 12 leads.

Twelve leads will not tell you your best offer, geography, landing page, audience and budget. They can help answer whether clicks for one offer turn into quoted jobs at a cost you can live with. That distinction is the whole test. If you change the offer and the geography halfway through, the final cost per lead tells you little about either choice. A small budget buys one answer, not five.

I used to tell clients to wait until they could fund a proper test. I was wrong. Pick one offer in one town, run it on $25 a day for one offer and one geo, and judge it on cost per qualified lead. Keep the landing page constant while you try two or three descriptions of the conversation. Before launch, decide what counts as qualified: a request for a quote on that job, not just a form submission from anyone with a pulse. That is a job for month one. A second month can ask the next question.

Myth 2: “ChatGPT ads need their own specialist”

This one survives because every new placement creates a new job title on LinkedIn. I have watched it with Shopping, with Performance Max, with every acronym since 2016. ChatGPT targeting does not work like keywords: you write context hints that describe the conversations where your ad fits, not match types. In one $2,319 B2B test, keyword-style hints pulled a 1.2% CTR while conversational hints pulled 0.9% on similar impressions.

The wording changes; the underlying job does not. If you already know which search terms turn into quotes in Google Ads, you have a starting point for describing the conversations that matter. You still have to decide which problems are worth paying to appear beside and whether the resulting leads are any good. That is search work, not new-channel magic. Put the test with the team that owns search intent and conversion data, rather than hiring a ChatGPT specialist to work in a separate lane.

Agency guides for Copilot make a related point: those ads run through Microsoft Advertising from existing Search campaigns, and you should fund the test from existing search budget with a target CPL and a decision date agreed before launch. ChatGPT is a different placement, but the planning discipline carries over. One owner, one shared conversion list, one question. A separate retainer for a separate specialist looks a lot like 2015 overhead.

Myth 3: “Clicks and CTR tell you if it is working”

People believe this because CTR is the first number the dashboard shows, and a 1.9% CTR feels like proof. It is proof that people clicked. In one B2B test Workshop Digital ran, ChatGPT clicks cost $3.14 versus $6.76 on Google Search, 54% cheaper, with CTR at 1.1% versus 5.2%. ChatGPT drove zero MQLs. Google drove two.

Barn2 saw the same gap between interest and outcome on a smaller budget: $477 over six weeks bought 19,154 impressions and 370 clicks at 1.93% CTR, with zero conversions. Paid sessions averaged 37 seconds and 1.09 pages. Organic clicks from ChatGPT answers averaged 443 seconds and produced five sales. Cheap clicks without qualification prove nothing.

This is where a small test can save you from a large mistake. If the clicks arrive but nobody asks for a quote, another cheap click does not repair the problem. It gives you another visit to count. Follow the click to the lead or sale, or do not run the test. Choose one conversion event that matters, such as a booked quote or a checkout, and judge the spend on cost per qualified lead rather than CTR. Otherwise, you risk scaling traffic that does not pay you.

Myth 4: “You can paste your Google ad copy in unchanged”

People believe this because the headline limits look familiar, so the old responsive search ad gets copied over. But the query is not the same. A Google search for emergency plumber Austin is three words. The ChatGPT version might be a paragraph about the leak under the sink, the house age, the budget and whether parts are in stock.

There is another distinction worth keeping straight: ads do not influence ChatGPT’s answers; they run on separate systems and sit clearly labeled below the response. You are buying the sponsored slot next to that conversation, not a mention inside it. I explain the split in how ChatGPT separates the answer from the ad.

That context changes the copy’s job. In the $2,319 test, brand-first headlines and industry-specific images lifted CTR. The reader has already described a problem and is scanning for someone who can fix it. Generic Best Plumber Deals copy has less to say to that reader than Licensed Austin Shop, Same-Day Repipe Quote beside a photo of an actual van. The point is not to cram the whole conversation into a headline. It is to sound like the ad belongs beside it.

Make the landing page finish the same thought. If the chat was about the cost of a two-bath repipe, send the visitor to the repipe price section and a quote form, not your homepage. Match the specificity the conversation gave you before you pay for the click.

Myth 5: “You need a whole new tracking stack first”

This myth comes from consultants who sell tracking projects. They are not wrong that measurement is messy. They just put the project before the test. You need a working count before day one, not a rebuild. Tracking runs through the oaiq pixel, server-side Conversions API and a no-JS fallback, reporting into one Pixel ID and deduplicated by a shared event_id. Put UTMs on every ad, install the pixel and send the one event that matters from the server too.

Then check the count against your own analytics. One early small-business test burned through $500 in about an hour at roughly $3.50 a click, with only about 13% of those clicks verifiable in site analytics, zero leads and a dashboard running a day behind. If you cannot reconcile the clicks and outcomes, Ads Manager is a poor place to make a scale decision. Write down the gap rather than choosing whichever dashboard gives the nicer answer.

The event setup matters beyond reporting. Only about 40% of ChatGPT-influenced conversions happen in the click session; 60% arrive later through direct or search. If the pixel and CAPI are broken, oCPC learns from bad data, and you may blame the channel for a measurement failure. I want one quote form, one call number and one thank-you page I can check. I want to know what each records before I let a campaign interpret those records for me. Do not buy a new stack. Prove the small one fires before you spend a dollar.

Myth 6: “If the first two weeks look good, scale immediately”

I watch owners make this call after ten good days, and I get it. The chart points up, the CPC looks cheap, and the rep on the call says more budget means more of the same. But with 12 conversions on a small test, two extra jobs can move your CPA noticeably. Add reporting lag: Ads Manager can run about a day behind, and outcomes may arrive after the click session. Double the budget on week-two CTR and you may simply scale noise.

Shop owner staring at a tiny green chart spike while a delayed sales receipt trails behind it

I treat a budget increase like it needs a signature. Before I add a dollar, I want:

  1. The same cost per qualified lead in my own analytics, not just Ads Manager.
  2. Call or form evidence that those leads were quotable jobs.
  3. Two full weeks where that held while spend stayed flat.

None of those checks asks the chart to predict the future. They ask whether the thing I am about to buy more of is the thing the business actually wanted. That is also how groas runs Search Ads: Google, Bing and ChatGPT under one strategy for one flat fee. You set the goals and budget, and changes come with the reasoning behind them. No percentage-of-spend incentive to push an increase because the dashboard looked pretty. Scale on qualified jobs you can verify, not a green line in someone else’s interface.

Myth 7: “Early movers win automatically”

This is the one I have the hardest time killing, because I want it to be true. Getting in early feels like an advantage you can bank. But ChatGPT ads do not give you a permanent moat. They give you cheap tuition while other businesses wait. Ads appear only for Free and Go users; Plus, Pro, Business, Enterprise and Edu accounts do not see them. So an ad will not reach the paying ChatGPT users in those tiers, however good its copy is. The test pool started February 9, 2026, in the US, and self-serve opened in May 2026. An auction with little competition is an opportunity to learn; a low CPC is not proof of buyer intent.

The distinction matters most when you decide what the first test is for. A $750 to $1,200 test can tell you whether your conversion event fires on real jobs, what a qualified lead costs for one offer in one place, and which of a few context hints draws buyers rather than browsers. It cannot prove scale, lifetime value or whether ChatGPT beats Google. Twelve conversions do not tell you what happens at ten times the spend; a 30-day window does not tell you whether those jobs become repeat work. Treating an early result as a verdict on the whole channel wastes the useful, narrower answer you paid to get.

If I owned a shop, I would start with this: for my highest-margin job in my best zip codes, what does a quoted ChatGPT lead cost when the ad and page repeat the problem the visitor described? I would write down the answer before changing anything else. I would rather be the shop that spent $750 learning whether a prompt about replacing a two-car garage door opener in Plano produces quotes than the shop that spends $7,500 next year asking five questions at $9 a click. Early movers do not win automatically. They get a cheaper chance to learn, and that discount expires.