Most agencies won’t tell you this: a new channel does not have to mean a new retainer. If Google is already running under groas and you have $500 to $1,000 for the next test, the contenders are ChatGPT Ads and Microsoft Search. Your Search Ads service already covers Google, Bing, and ChatGPT under one fee. The decision is where to put the media dollars and what those dollars can teach you.
I’ll call it early: for most small budgets, Microsoft Search wins. You can inspect the intent, measure the click, and stop a bad test quickly. ChatGPT Ads is the interesting loser, worth picking when you have verified access, a placement that fits your buyer’s question, and tracking that survives a real redirect. Novelty is not acquisition economics.
First, separate the management fee from the media budget
Say you spend $20k/month on Google. Under groas, management costs $1,499 a month for Search Ads, whether you run Google alone or add Bing and ChatGPT. Media is separate and goes to the platforms. The management fee does not rise when you add a test, and groas takes no cut of spend. That is the point of flat-fee management: the next $750 is a media decision, not a retainer negotiation.
Before comparing the channels, check that you can actually use both. Log into Microsoft Ads and confirm the account can serve in your market. Then check the ChatGPT ads manager: can your account serve at all, in your country, to the users you want? I have seen small tests die because nobody checked access first. If either answer is no, you do not have a head-to-head. You have one option.
Round 1: Which channel lets you inspect buying intent?
Microsoft Search sells you typed intent you can inspect. Someone types “emergency plumber austin” on Bing, Yahoo, AOL, or DuckDuckGo; your ad shows; and you can see the query, match, and click in a search-terms report you can mine for negatives. Its $0.05 minimum bid and $5 monthly minimum make entry possible on a $500 to $1,000 test, with pay-per-click pricing.
I used to tell clients to import their Google structure as scaffolding, then tighten to Microsoft sites and run in Expert mode with exact match and UET tagging from day one. I was half wrong about the import: imported waste imports fast. But the principle holds. Start with demand you can read, inspect what you bought, and cut the queries that do not belong. Visible search terms give a small test something to optimize, even before the lead count tells you much.
On ChatGPT, you rent inferred intent you cannot inspect in the same way. Ads are sponsored cards below the answer, never inside it, matched to the live conversation through plain-language context hints rather than keywords. There is no Quality Score, search-terms report, or view of the prompt that triggered your ad. For now, there is one ad unit at a time.
Access narrows the audience further. Ads serve logged-in users 18 and older on Free and Go tiers; Plus, Business, Enterprise, and Education are ad-free. In one test of about 97,000 impressions cited by Ranketta, the closer someone looked like the ideal B2B customer, the more likely they were already on a paid plan that never shows ads. Before buying conversational intent, ask who can actually see the ad in your account, not how many people use ChatGPT.

Round 2: Can you trace a lead back to the click?
Microsoft gives you plumbing you already know how to check. UET tags the click, GA4 records the session, and the search term that caused it sits in a report you can inspect. When cost per acquisition moves, you can work back through the query, ad, landing page, and bid. That does not make every lead a good lead. It gives you places to look when the numbers disappoint.
ChatGPT measurement needs more careful handling. You run an oaiq pixel and server Conversions API, deduplicated by a shared event_id, add UTMs for GA4, and choose a click window of 7, 14, or 30 days with a short view-through window. Lose the oppref identifier in a redirect chain and the click is gone. Send one bad event in a batch and the API rejects the batch; send an event older than 7 days and it is rejected. I have lost cleaner setups to dumber redirect mistakes. A dashboard total is not enough if your own analytics cannot verify what happened after the click.
The early numbers are another reason to check before you trust. In one same-advertiser test of about 1,500 users, Google averaged 41 seconds of engagement and a 3.71% conversion rate, against 17 seconds and 0.21% for ChatGPT. Reporting was only by country, even when targeting ZIP or DMA. A coaching firm reported a small-budget version of the problem: $500 spent in just over an hour at about $3.50 CPC, 140 reported clicks, only about 13% verifiable in Wix and GA4, zero consultations, and a dashboard lagging a day.
Those examples are not your forecast. They are a warning about what to verify. Use the same definition of a qualified lead and cost per acquisition on both channels. Do not scale ChatGPT until its clicks show up in your analytics at the same rate Microsoft clicks do.
Round 3: What can $750 actually teach you?
$750 does not buy certainty. Size the test from your own baseline and ask what it could reveal. Say your current Google cost per qualified lead is $120. At roughly $3 a click, a $750 Microsoft test buys about 250 clicks. If your page converts at 3%, that is seven or eight leads. It is not a final verdict on long-term CPA. It is enough to start checking whether Microsoft intent behaves like your Google intent and whether clicks make it through to your CRM.
For ChatGPT, $25 a day with $3 to $5 starting bids gives you a way to pace a 30-day test. If clicks cost roughly $3 to $5, the same $750 buys about 150 to 250 clicks. But without search terms to prune, and with weaker engagement in the cited early test, a handful of leads tells you less about what to fix next. You still need to know whether the placement reached the right person, the page carried their question forward, and the measurement survived the trip.
Do not split $750 into two $375 channel tests and call it a head-to-head. You will have two thin samples and a report saying both channels need more spend. Run the channels in sequence, Microsoft first. You can hold back half the allocation while you check the first stretch of the Microsoft test; that is different from dividing the money between two channels before either produces a clean read.
And do not assume Microsoft wins because its clicks must be cheaper. Across 73 advertisers and 4.5 million clicks, Microsoft averaged $1.54 against Google’s $2.93, with 64 of 73 cheaper and a median saving of 59%. That discount flipped when Google clicks were already under a dollar: Microsoft ran 40% more expensive. Price your test from your own cost per qualified lead, not a platform average or an attractive CPC in somebody else’s account.
Round 4: What work must happen before and after launch?
A Microsoft test is execution you can audit the next morning. Build separate campaigns from the Google import. Start with exact match on the five to ten queries that already close, carry over shared negatives, and match landing pages to intent instead of sending every ad group to one generic page. Check UET and GA4. Mine search terms daily for the first two weeks, and log each bid or budget change with a reason.
That daily work sits inside your groas fee, which is why adding Bing does not add headcount. Put a simple request to your manager before launch: show me the queries, the landing page for each, and the CPA that kills the test. If they cannot answer before spend starts, do not start.
A ChatGPT test is a different job, and most decks gloss over it. There are no keywords to structure. The work shifts to plain-language context hints, a tight set of conversational openings worth appearing under, and a landing page for someone arriving mid-question rather than after a typed search. The page has to carry the answer forward: the ad sits below the response and has to earn the next click from there.
Check the unglamorous parts first. Do redirects preserve the click identifier? Do the pixel and server events fire with the same ID? Do UTMs arrive intact in GA4? Reports lag about a day and split only by country, so daily management means watching verified sessions in your own analytics, not admiring a platform dashboard. A ChatGPT test is not ready just because the creative is ready.
The guardrail for either channel is boring and non-negotiable: nothing scales without your sign-off. I set three numbers before launch:
- The test budget: say $750.
- The kill line: say $150 with zero verifiable clicks in analytics, or $300 with zero qualified leads.
- The threshold for more spend: click verification and the lead definition must match the baseline before any increase.
groas runs with that approval discipline. You set goals and budget, actions are logged with reasoning, and increases wait for approval. If your agency proposes ChatGPT before it can put those three numbers in writing, it is not proposing a test. It is proposing a subscription to hope.
Both channels sit under the same $1,499 Search Ads management fee. Bids, keywords where applicable, budgets, and pages are worked around the clock. The media spend is what differs, and media should be judged by what it proves per dollar. Here is the head-to-head in one view.

| What you decide on | Microsoft Search | ChatGPT Ads |
|---|---|---|
| Intent you can see | Typed queries, search-terms report, negatives you control | Conversation hints, no keywords or prompt data |
| Click you can verify | UET and GA4, with a query to investigate | Pixel and server API, identifiers to protect, country-only reports |
| Learning from $750 | About 250 clicks at an illustrative $3 CPC, with queries to prune | About 150 to 250 clicks at an illustrative $3 to $5 CPC, with no terms to prune |
| Work to run it | Campaigns, matched pages, daily mining, logged changes | Hints, an answer-fit page, redirect and event checks |
The verdict: give the first test to Microsoft Search
For the groas customer with $500 to $1,000 and Google already running, Microsoft Search takes the test budget first. You can see what you bought, check it in your own analytics, and name the CPA that ends the experiment. That is not a vote against conversational ads as a category. It is arithmetic. A small test needs to produce a readable answer, and typed search gives you fewer ways to lose that answer between click and booked job.
Pick ChatGPT Ads anyway when three conditions hold at once: your account is verified to serve to the users you want, the placement fits a question your buyer actually asks mid-conversation, and your pixel, server events, and UTMs survive a live redirect test before spend starts. If any one fails, keep the money in the baseline that already converts. I learned that lesson the expensive way, mining negatives at 1am for accounts that never needed a second channel, just better execution on the first. Under one flat fee with approval before any increase, restraint is free and waste is a choice.
Before approving either test, get one short email from your manager covering:
- The exact queries or conversation openings you will buy.
- The landing page for each one.
- The tracking check that proves clicks reach GA4 and your CRM.
- The kill line in dollars and qualified leads.
- The written rule that spend does not rise without your sign-off.
If Microsoft can answer that email today and ChatGPT cannot, you have your order. Spend the next $750 where the answer is readable: usually Microsoft Search first, ChatGPT second, and neither until the measurement holds.

