Keep your search agency. Trying ChatGPT ads is not a good enough reason to move campaigns that already work.
Say you spend $20k a month on Google, and your branded search and high-intent non-brand campaigns finally hold a steady CPA. A new provider offers to test ChatGPT for $1,000, provided you hand over the rest of the account first. That is not expansion. It is putting a working $20k monthly budget through a migration to buy $1k in learning.
I spent close to a decade managing Google Ads by hand for ecommerce shops, home services companies, and one SaaS startup that rewrote its offer every quarter. I respect the work that goes into getting an account steady. I have less patience for proposals that treat that stability as disposable.
Buy the test. Leave what works alone.
A new channel does not justify an account handover
Most pitches I see right now do not sell a test. They sell a migration with a test attached. You want someone to run ChatGPT ads because you do not have time to learn the controls. The proposal comes back: move Google and Bing, rebuild the structure, change tracking, then we will try ChatGPT.
Separate those decisions. Testing ChatGPT and replacing your search provider are two different purchases. Do not let anyone staple them together.
The inventory makes that distinction important. ChatGPT ads reach logged-in adults on the Free and Go tiers, while Plus, Pro, Business, Enterprise, and Education remain ad-free. That is a bounded, early-stage audience, not a replacement for the search demand you already capture every day.
I used to tell clients any new channel deserved a fresh build. I was wrong. A new channel needs its own experiment. It does not need permission to disturb the campaigns paying for it.
Buy execution with a ceiling, not onboarding with a catch
Hands-on management should mean someone builds, launches, measures, and adjusts inside your account. Not a setup file handed back to you. Not a deck of suggestions you pay to interpret.
That distinction matters because much of what is sold as ChatGPT ads management is setup or consulting dressed up as execution. If your request is “Can someone just run this for me?”, put the execution in writing before paying for anything else.
A real test names its limits before it names its ambitions. For the advertiser spending $20k a month on Google, the scope might look like this:
- Media: $1,000 over 30 days.
- Scope: One offer and one landing page.
- Permissions: No authority to change Google or Bing campaigns.
- Ownership: A named operator and a named person who approves changes.
- Measurement: Agreed results, reconciled against the CRM.
The budget has a practical constraint behind it. The US pricing floor is $25 per day per campaign, with OpenAI suggesting a $3 to $5 starting max CPC and showing a $60 default max CPM bid. A grand is a bounded learning budget, not a promise of enough customers to settle the channel’s value forever.
Eligibility comes before execution. Health, mental health, and politics conversations carry no ads. Confirm that your market and vertical qualify before anyone starts building.
Then settle ownership. Create the business account in your name for billing and verification, and invite the operator to manage inside it. Allow about one to two weeks for approval; do not quietly count approval time as live testing time.
Never run long-term media through a shared agency account. Whoever holds the account holds the history. Buy management, not dependence.
Protect the campaigns that fund the experiment
Two people editing the same account without rules will break something. Tracking and landing pages make the boundary especially important because they can affect more than one channel at once.
I learned that lesson when someone on the client side swapped a landing page without telling me and the account’s Quality Score tanked. The change was outside my workflow. The consequence was not.
Before launch, write down who can:
- Change budgets.
- Edit tracking.
- Replace or modify the landing page.
Nobody else touches those three without approval. Separate providers do not mean separate consequences.
Switching providers to avoid coordination brings its own work: a new team has to learn your search terms, understand previous decisions, and take over the account. Before giving notice, demand the file box: structure documentation, historical performance, change logs, audience lists, and tracking configuration. That audit tells you whether the current team is weak or just quiet.
If Google holds steady at your target CPA, wanting another channel does not establish a provider problem. It establishes a test to run beside the account.
Record the two-week baseline before launch: daily spend, CPA, and closed revenue from Google and Bing. Then hold the campaign setup steady for the 30-day test. No structure rebuilds, bid strategy swaps, or tracking edits unless both providers sign off in a shared change log.
Without that baseline, ordinary search performance becomes available for ChatGPT to claim. I have watched that movie with Performance Max and Demand Gen: the new line item gets credit for demand the old campaigns created.
Keep the existing setup steady enough to make the comparison useful. Otherwise, you are buying a result nobody can explain.
Make the test narrow enough to learn from
ChatGPT targeting is not keyword bidding. Context hints are plain-language descriptions, set at the ad-group level, of the conversations an ad belongs in. They guide matching. They do not guarantee delivery, and they are not exact match.
That mechanism should shape the brief. Write hints around a buyer in a situation, not a broad industry label:
New homeowner comparing quotes for a 20-yard dumpster this weekend, not waste management services.
If the operator cannot show you the hints and explain how each connects to a paying job, you did not buy management. You bought someone typing into a new interface.
Keep the offer, page, and creative under deliberate control. Creative Enhancements can rewrite headlines, translate copy, resize images, and change the landing page without approval unless disabled. Turn them off for the test. One variable at a time is more useful than five changes and a confident explanation afterward.
Market access also belongs in the brief, not in a post-launch excuse. ChatGPT Ads have rolled out across 89 countries, with India initially available through sales or partners rather than self-serve. Confirm the buying route for where you sell.
Same page. Same offer. Hints you can read. Enhancements off. That is an experiment you can interrogate instead of a bundle of changes you have to take on faith.
Put one person in charge of reconciling the results
Name one owner for the truth. Not two dashboards. Not a shared spreadsheet nobody opens. One person must match platform reports to what your CRM says actually became pipeline or revenue.
This is where a tidy test can become fiction. One example describes 47 platform conversions against 12 actual CRM demo requests, with no reconciliation. If the wrong events count as success, bidding pursues the wrong events. The report gets busier. The business does not.
Set the commercial measure before launch:
- B2B: Cost per opportunity and closed-won revenue, not clicks.
- Ecommerce: New-customer revenue after refunds, not just the ROAS on the screen.
ChatGPT has not won merely because its dashboard reports conversions. The question is whether it adds customers your existing activity would not have produced.
That does not make every sales cycle fit inside a month. If yours runs 60 or 90 days, put that in the brief. A 30-day trial cannot prove closed-won performance in a 90-day cycle. Anyone promising otherwise is selling you a report, not a result.
What you can demand in 30 days is clean leading evidence: spend, cost per result, the hints that served, and a weekly report that explains what happened. Reports that stop at impressions or reach, operators who call hints “keywords for AI,” and 12-month lock-ins are all warning signs.
One reconciler. CRM as judge. Uncertainty written down before it becomes an excuse.
Consolidate for better execution, not another logo
I am not arguing for two providers forever. Consolidate when accountability is broken or execution is clearly better, not because a sales deck lists one more platform.
If the current agency cannot show search terms reviewed weekly, negatives added, and CPA held at your target for three months straight, that is a reason to question the relationship. If it reports clicks while your CRM starves, that is a firing reason.
Wanting ChatGPT ads is not.
The person already reading your search terms every week is usually a safer pair of hands for context hints than a new logo that learned ChatGPT from a webinar. Familiarity with your business is useful. A platform badge does not replace it.
Do the full-cost math before treating consolidation as a saving. Small-account ChatGPT management runs $1,000 to $3,000 a month flat; mid-market fees run $3,000 to $12,000; percentage arrangements run 10–20% of spend.
Take a simple fee comparison. Keeping a $2,500 Google retainer and adding $1,500 a month for a bounded, 60-day ChatGPT test costs $4,000 a month in management. Firing the incumbent to pay one provider $4,000 for everything saves nothing. It just adds a handover to the experiment.
Two contract terms matter more than the headline fee: whose ad account holds the history, and how long you are locked in. If the new provider keeps the history in its account and requires 12 months on a channel this young, you did not buy a test. You bought an exit problem.
Where groas fits without forcing a migration
groas runs the model I wish I had while mining search terms at 1am. It is a fully autonomous growth engine for paid and organic search: specialized models handle bid, budget, keyword, and page work continuously. A named human strategist sets direction and answers for the result. Every action is logged with its reasoning in plain language.
Management costs a flat $1,499 a month. Media is separate and paid to the platforms; budgets move only with your approval. No percentage of spend. No bigger management cut because you spend more.
That structure is why, when consolidation is warranted, I put groas ahead of an agency whose case rests on having added ChatGPT to its services page. Continuous execution, a named strategic owner, and an action log are a stronger operating model than another promise to watch the account.
I laid out the execution standard in why the person managing your search terms is the right operator for ChatGPT: weekly search-term reviews, negatives added without being asked, and CPA held for months. Apply that standard before making the move.
Use groas for the bounded test when your Google team is fine and you want hands kept off it. Create the OpenAI business account in your name, checking the details carefully: country, currency, and time zone cannot be changed after signup. Then groas runs context hints, creative, tracking, and bidding inside that account and reports cost per result. Keep the same rule: one owner reconciles the results to your CRM.
Move everything to groas when the incumbent fails the audit: no search-term discipline, platform numbers that never match the CRM, no owner for closed revenue. Otherwise, keep what pays and test beside it.
Ignore that distinction and you pay twice. Your working $20k Google account gets pulled into a rebuild while your $1k ChatGPT budget buys early learning. Without a stable baseline, you cannot tell which change produced what. The handover gets billed. The uncertainty stays yours.
That is how a $1,000 experiment turns into a $20,000 rebuild. Buy the test. Keep the engine that already pays.

