There is no such thing as a ChatGPT ads specialist with years of ChatGPT ads results. Hire one on that promise and you are paying for a track record the calendar will not allow.

I spent nine years running Google Ads by hand for ecommerce shops and home services clients, mining search terms at 1am and watching CPA fall apart when tracking broke. I know what a track record looks like: years of accounts, hundreds of tests, benchmarks you can argue with. Nobody has that on ChatGPT ads. Self-serve buying only opened in May 2026. The work belongs with whoever already runs your Google and Bing accounts, on the same conversion data, not with a new boutique selling you a specialist badge.

The specialist pitch has a calendar problem

Your inbox already knows the script. A boutique shop promises a dedicated ChatGPT ads team, proven playbooks and category case studies. What the deck calls specialist coverage, I call a new logo on old Google slides.

Do the track-record math before you believe it. OpenAI started US testing on Feb 9, 2026, at $60 CPMs with $250k minimums, then dropped to around $25 CPM and $50k minimums by April before opening its self-serve Ads Manager beta on May 5, 2026, with $25-a-day minimums. That is months of buying history, not years. There is also no formal OpenAI agency certification, and buying is self-serve with no invite list. Ask who issued the badge on the homepage. The answer is no one.

That does not mean nobody can run an ad. It means a shop cannot borrow its Google Ads history, rename it ChatGPT expertise and charge you for a mature playbook. Adjacent experience matters. A claimed record of proven ChatGPT category results is a different proposition, and it should survive a look at the dates.

Ask when the results were earned before you ask how impressive they look.

Early results are tests, not category benchmarks

OpenAI says ChatGPT Ads has no cross-advertiser performance benchmarks yet, calls the platform a beta and warns that delivery and spend will fluctuate. Read that twice. The platform owner is telling you not to treat somebody else’s benchmark slide as a reliable forecast. Even ranking pages selling specialists concede that self-serve only opened in May 2026, so rankings reflect adjacent PPC experience rather than years of performance data on this placement.

The early numbers show why one tidy success metric cannot settle the question. In one same-advertiser test, Workshop Digital spent $2,319 for a 1.06% CTR, $3.14 CPC and zero MQLs, while Google Search delivered a 5.2% CTR and two MQLs on the same offer. Another B2B account saw $0.76 CPC on ChatGPT against $10.46 on Google, but only a 0.8% conversion rate against 6.64%; its ChatGPT lead still cost $92 against Google’s $158. One test produced no qualified leads; the other produced a cheaper lead despite a much lower conversion rate. Neither gives you a category benchmark or permission to declare victory from CPC alone.

I judge a new placement on added enquiries at an acceptable cost per qualified lead. A click can look cheap and go nowhere. A lower conversion rate can still produce a workable lead cost. You only find out which story you bought when the same qualification standard applies to both channels.

If a deck promises a proven ChatGPT CPL for your category, ask which months it covers and what counted as a qualified lead.

Search skills transfer. Search controls do not.

The mechanics are new. The discipline is not. ChatGPT ads do not run on keywords. They match conversation context through free-text context hints and interpretation of your ad and landing page, and you cannot see the triggering conversation. Custom audiences need 25,000 matched users, and Plus, Pro and Business users do not see ads. You are buying intent you cannot fully inspect, with thinner controls than the search accounts you know.

I have bought traffic with that uncomfortable mix of intent and ambiguity before. It was called broad match in 2018. The lesson was not to pretend every click came from the query I wanted. It was to keep the offer tight, inspect what the platform would show me and refuse to confuse traffic volume with demand.

That is why query-to-offer matching, negatives by theme in search and landing pages that answer the user’s question still matter. A search ad that promises one thing and lands on a generic homepage pays for curiosity and gets no pipeline. A chat user has been reading an answer before the ad appears. If the landing page drops that thread, a clever ad will not rescue it. Continue the answer, name the price or process and ask for one action.

The third transfer decides whether you learn anything: conversion tracking needs to connect a chat click to revenue before the first dollar spends. The setup is specific. You need the OpenAI Pixel plus Conversions API with shared event IDs, the oppref click ID stored in a first-party cookie, and the default 30-day click and one-day view attribution window. Miss the UTM discipline and ChatGPT clicks can dissolve into GA4 as AI Assistant or Paid Other traffic. Then a boutique can celebrate a 1.39% CTR and a $2.12 CPC portfolio median while you wonder whether pipeline moved at all.

I used to tell clients tracking could be fixed after launch. I was wrong. On a channel with no established benchmarks, dirty data is not a minor launch delay. It can waste the test.

Put the test beside Google and Bing

Practitioners spending on ChatGPT treat it as an incremental test benchmarked against Google Search at the same intent moment, judged on shared conversion data and added enquiries. That comparison only works if someone can see both sides. Split the channels across vendors and each gets to present its own version of the same customer journey: Google claims the close, ChatGPT claims the assist, and you still have to decide where the next dollar goes.

This is why fully managed execution beats a hybrid patchwork when inventory is new and volatile. One owner needs the conversion feed and the authority to move budget across Google, Bing and ChatGPT. Otherwise, every pacing decision becomes a scope negotiation. A new channel does not need another retainer to watch an old one do the useful work.

Hands-on management also beats a consulting deck here. A consultant hands you a point of view about prompts and positioning, then leaves you to run a beta auction that shifts week to week. An operator watches delivery, cuts the context hints pulling curiosity clicks, rewrites the landing page that loses chat traffic and moves $500 from a ChatGPT test into Search when lead quality says to move it.

The honest shops admit the limit. One UK team running budgets says it is too early for mature case studies and will not pretend otherwise. That is a better answer than a laminated specialist badge. Look for evidence that someone can manage intent, budgets and bad data, then make that person accountable for this test. Do not pay extra because they have renamed themselves around a young placement.

groas is built for that operating model. Its Search Ads coverage includes Google, Bing and ChatGPT under one flat fee, with bids, keywords and budgets managed as one system rather than three retainers. One strategist owns the outcome, actions are logged with reasoning, and shared conversion data lets ChatGPT be judged against qualified pipeline and closed revenue instead of CTR. The point is not that a prompt trick makes the new inventory predictable. It is that the person responsible for the result can see what happened and act on it without waiting for three vendors to agree.

Keep ownership of the result with the team that already owns your Search revenue signal.

Skip the test if your Search data is broken

This advice has a starting condition. If you have no Google or Bing spend and no conversion tracking worth trusting, do not open a ChatGPT test to find product-market fit. You will pay for clicks without a useful read on what they became. A chat ad cannot repair an offer or a measurement setup that already fails on Search.

Fix the landing page, wire CRM revenue back into the ad account and get 30 days of clean Search data first. Then you have an offer to carry over and a comparison worth making. The new channel can wait. Do not buy a second source of uncertainty before you have measured the first.

Four questions that expose a badge-only operator

Forget the case study PDF. On a channel with months, not years, of self-serve history, operational answers tell you more than a specialist label. Ask these four questions on the sales call. Short answers are good. Vague answers are the tell.

  1. Who touches the account daily, by name, and what did they change yesterday on another account? Consulting shops talk about cadence. Operators talk about yesterday.
  2. What tracking is live before the first dollar? Look for the OpenAI Pixel and Conversions API with shared event IDs, oppref in a first-party cookie and UTMs that keep clicks from blending into AI Assistant traffic.
  3. What budget cap needs my approval, and who moves money between Google, Bing and ChatGPT? If nobody owns the shared conversion read, nobody can defend the next allocation.
  4. What will we know by week four in cost per qualified lead? Set the acceptable CPL against Search and name what gets killed if the test misses. Skip this and a low CPC can keep an unproductive test alive; hide behind the learning phase as an excuse and it may never face a decision.

Hire the boutique and here is what the next ninety days can look like. They open an isolated test with its own budget and its own read on performance because they cannot touch your Google account without stepping on your existing operator. Week two brings a cheerful report: low CPCs, solid CTR, healthy delivery. Week six brings the awkward question about why CRM shows no added pipeline. By month three, you have paid a retainer to teach someone else search fundamentals on beta inventory: who can see the ad, why an audience is too small and what happens when conversion data is split. The channel did not fail you. The operating model did.

Keep ChatGPT where search lives and the same ninety days can teach you something usable. One team runs Google, Bing and ChatGPT against one definition of a qualified lead and a shared view of revenue. Say you spend $20k a month on Search. Carve out $1,500 for ChatGPT as an incremental test with a written kill rule: match Search CPL on qualified enquiries by week four or move the money back. The operator checks context hints daily, rewrites the page that loses chat clicks and shifts budget when the data calls for it. No specialist badge required. Just the old work done fast: intent matched to offer, tracking tied to revenue, budgets moved without a meeting.

If a pitch deck promises proven ChatGPT category benchmarks, close it. The platform owner says cross-advertiser benchmarks do not exist yet, and the buying timeline offers months of history, not years. Put the test with whoever already owns your Search revenue signal and judge them on added pipeline at a CPL you can afford. Pay for the boutique story instead, and you may spend the next ninety days funding someone else’s education while your own Search operator waits for the data.