What a $1,500 ChatGPT Ads Test Can Prove: The Conversion Math Agencies Skip

Six conversions. At a $5 click and a 2% conversion rate, that is what a $1,500 ChatGPT ads test buys on paper. It is enough to tell you someone converted. It is not enough to prove a reliable CPA.

I used to tell clients a small test would at least give us direction. I was wrong. On a new placement with thin intent signals and inconsistent tracking, a small test often tells you how noisy small numbers are. Here is the arithmetic I run before anyone funds one, and the denominator I demand when an agency says it has real ChatGPT ads results.

The claim: a few conversions are not proven results

I keep seeing the same pitch. An agency runs a small ChatGPT ads test, gets a few leads, and sells proven ChatGPT ads management. Ask how many clicks and conversions produced the headline CPA, and the story gets vague. That vagueness is the tell.

ChatGPT ads launched Feb. 9, 2026, at $60 CPM with a $200,000 minimum. Self-serve opened in May 2026 with no lifetime minimum and a $25-per-day minimum. Small-budget tests became possible. The question is what their results can actually support.

Start with budget to clicks. Early-beta data puts ecommerce and consumer clicks around $3 to $5, and software and finance clicks at $8 to $18. One aggregator reports a $2.12 median CPC across its clients. I use $2, $5, and $12 as planning cases, not promises. At $1,500, those prices buy 750, 300, or 125 clicks. Same budget, sixfold difference in clicks. If an agency will not tell you its CPC, you cannot tell whether its result resembles the test you could run.

How many conversions can those clicks buy?

Apply conversion rates to the click totals. Projections across 17 verticals put ChatGPT ads conversion rates at 0.2% to 5.7%, with ecommerce at 2.5% and B2B SaaS at 1.2%. For context, WordStream reports an 8.18% median conversion rate for Google and Microsoft search. Those are different placements and datasets, not interchangeable forecasts. The 2%, 5%, and 10% rates below show how much the assumed rate changes the answer; the last is a stretch case, not a baseline.

The table divides budget by CPC, then multiplies clicks by the assumed conversion rate. Conversion counts are rounded to whole numbers. At $5 per click and 2%, the result is six conversions. At 5%, it is 15. A $1,500 test does not have one inevitable outcome; it has a small sample whose meaning depends on CPC, conversion rate, and what counts as a conversion.

Why eight conversions give you a CPA estimate, not a verdict

Say the $5-click test gets eight conversions from 300 clicks. The dashboard prints a $187.50 CPA. That is the observed CPA for this test, not proof that the next $1,500 will buy eight more.

Practitioners discussing paid-search tests want at least 100 conversions per variant, while Google Ads Experiments calls for around 500 before declaring a winner. Those are comparison standards, not requirements for learning anything. They illustrate how far eight conversions sit from a confident winner call. In one confidence-interval example, three conversions from 400 clicks put the underlying conversion rate between 0.3% and 1.7% at 80% confidence. A few additional conversions can move a small test’s apparent CPA sharply.

I learned to distrust early winners after watching a 1 a.m. negative-keyword session turn a winning ad into next week’s loser. Single-digit conversions give you a starting estimate, not a stable price for acquisition.

Three businesses, the same $1,500 budget

Lead gen: cheap clicks still have to become jobs

Say you run home services and ChatGPT clicks cost $3. Your $1,500 buys 500 clicks. At a 1% click-to-enquiry rate, that is five enquiries at $300 each. At 2%, it is ten at $150 each. Both figures follow from the same spend; neither tells you how many enquiries became jobs.

A Perth test makes the distinction concrete: $2,250 bought about 57,000 impressions, just over 400 landing-page views, and four phone-tap enquiries. That is roughly $562 per phone tap, before asking whether anyone booked. A B2B account shows the trap from another angle: ChatGPT clicks cost $0.76 versus $10.46 on Google, but converted at 0.8% versus 6.64%. A cheaper click is not automatically a cheaper customer.

For lead gen, ask for cost per qualified job alongside cost per enquiry. Otherwise, the cheapest-looking metric gets the headline.

Ecommerce: an order count can move fast

Early pricing puts consumer clicks around $3 to $5, so $1,500 buys roughly 300 to 500 clicks at those prices. One public test spent $477 for 19,154 impressions and 370 clicks, with a 1.93% CTR and zero conversions. At the projected 2.5% ecommerce conversion rate, 370 clicks would imply roughly nine orders. Zero is a strong reason to inspect intent, offer, and tracking. It does not identify which one failed.

Even a test that converts has a small denominator. At $5 per click and 2%, the $1,500 scenario yields six orders. A handful of orders cannot carry a confident ROAS claim, particularly if you have not checked what happened after the click.

An ecommerce test needs orders that hold up beyond one odd day. If the first run produces six, use it to investigate the path to purchase, not to announce a durable return.

B2B: the sale may arrive after the test

B2B strains a $1,500 test at both ends. At the cited software and finance click range of $8 to $18, the budget buys about 83 to 187 clicks. Apply the projected 1.2% B2B SaaS conversion rate, and you expect roughly one or two leads.

Then you have to follow those leads. One illustrative funnel multiplies 5% click-to-lead by 30% lead-to-SQL and 20% close, for 0.3% click-to-customer overall. It is an example of why a lead rate is not a customer rate, not a forecast for your funnel. Reach has another limit: ads serve Free and Go tiers, not Plus, Pro, Business, or Edu accounts. That matters if the people you hope to reach sit outside the available audience.

For B2B, a $1,500 test can price raw leads. It usually cannot tell you what the resulting customers cost before those leads have had time to become customers.

Tracking can shrink an already small sample

Before trusting a conversion count, check whether the click survives attribution. In the reported setup, ChatGPT ads append only oppref by default, UTM passthrough is inconsistent, and a rollout stripped tags from some visits. The same account describes a 20% to 40% gap between platform clicks and GA4 sessions and a 30-day click and one-day view attribution window in Ads Manager. Those measures answer different questions. Do not place their totals side by side without saying so.

I used to trust platform conversions on a new placement. I was wrong. If the platform reports eight but your independent tracking confirms fewer, investigate the difference before treating eight as an acquisition count. For pipeline-focused B2B tracking, the distinction matters because a recorded form fill and a lead sales can follow up are not the same outcome.

Conversion definition matters just as much. A 5% rate on brochure downloads is not a 5% rate on qualified calls. And CTR cannot settle the question: one benchmark puts chatbot ad CTR near 0.9%, against about 6.4% for Google search, while a live set averaged about 1.3%. The same account treats below 0.5% CTR as a sign to inspect context hints. Click response may help diagnose delivery. It does not establish customer value.

Time is the final constraint. Google automation is described as stabilizing near 30 conversions per month, with a planning rule of target cost per lead multiplied by 30. At $70 per lead, $900 per month buys about 13 conversions, short of that rule. This is a bidding-stability benchmark from another platform, not a statistical proof threshold for ChatGPT ads. It is still a useful check against calling a tiny count settled.

Size the next test from the question you need answered

Forget round-number budgets for a moment. The planning formula is simple: required conversions ÷ assumed conversion rate × CPC = estimated spend. Each input needs to be stated, because a change in any one of them changes the budget.

Using 30 conversions as a planning floor, rather than a guarantee of proof, produces these cases from the CPC and conversion-rate assumptions above:

The spend figures are arithmetic, not quoted ChatGPT ads prices. They show why a $1,500 B2B run cannot reasonably promise a settled CPA when the working assumption is a $12 click and a 1.2% lead rate. Spend follows the decision; it does not follow whatever budget happened to be left over.

For a smaller learning run, I would frame the question more narrowly. These are planning examples to check against your own CPC and CPA benchmarks, not minimums that apply to every account:

• Local lead gen: At $3 per click and 3% conversion, 30 enquiries require 1,000 clicks, or about $3,000. Qualification still comes later.
• Ecommerce: At $4 per click and 2.5% conversion, 30 orders require 1,200 clicks, or about $4,800. Check feed and tracking structure as you would in an ecommerce revenue recovery before blaming the placement.
• B2B: At $12 per click and 1% click-to-lead, even 15 leads require 1,500 clicks, or about $18,000. Fifteen leads are a learning count, not a pipeline verdict.

Who should skip a $1,500 results test? Anyone whose clicks cost above $8, anyone selling over $2,000 without a way to follow leads to revenue, and anyone who cannot reliably track a lead to a call. You can still spend $1,500 to check delivery and tracking. Put that purpose in writing. It buys questions to investigate, not an answer to scale on.

Ask the agency for the denominator

When an agency says it can show real results, ask for spend, clicks, and conversions, with the conversion action attached. “$1,500 drove 12 leads at $125 each” sounds tidy. “$1,500 bought 300 clicks at $5, and 4% completed a form” gives you something you can examine. Then ask how many were qualified and how the platform total compares with GA4. A lead count without those denominators is just a count.

I want the following on one page, not scattered across screenshots:

• Spend and time in market: exact ad spend, dates, and daily budget. The same $1,500 spread over ten days at $150 per day is a different observation window from 60 days at the $25-per-day self-serve minimum.
• Clicks and what they cost: impressions, CTR, clicks, and CPC. If CTR is below the cited 0.5% context-hint warning level, investigate that before assuming more budget will fix the result.
• Conversions and how they were counted: the action, attribution window, platform total, and independently tracked total. State any click-to-session gap rather than hiding it in a different report.

Then ask what ran as the comparison. If Google and Bing did not run in parallel, you have a ChatGPT cost and outcome, not a side-by-side test. In the paired account cited earlier, ChatGPT clicks looked far cheaper, yet the 0.8% conversion rate trailed Google Search’s 6.64%. That is why I want ChatGPT assessed inside the same Search Ads operation as Google and Bing, with changes logged and the comparison visible. A standalone package that cannot see the control can miss the weak close while celebrating the cheap click.

The decision the numbers support

A $1,500 test can tell you what clicks cost, whether people click, and whether your tracking survives the trip. It can produce an early CPA estimate. It cannot, on single-digit conversions, prove that CPA will hold when you spend more.

Write down the decision before you fund the test: which conversion counts, how many would change your budget, what CPC and conversion rate you assume, and what you will do if the count lands at six instead of 30. If the math says $1,500 buys single digits, run it as a tracking and learning check or save for a test sized to answer the bigger question. Demand the denominator before you pay for anyone’s “proven results.”