ChatGPT Ads for Agencies: Subcontract, Refer, or Train In-House?

Five clients ask your lead-gen agency for ChatGPT ads in the same month. You can subcontract the work, refer the clients out, or train someone in-house. For an agency with five clients ready to test, I would subcontract. This is not primarily a hiring decision. It is a decision about who owns the client, the ad account, and the lesson if the test fails.

ChatGPT ads are live, appearing below answers with a sponsored label. Delivery depends on conversation context and landing page relevance rather than exact-match keywords, and the minimum starts at $25 a day per campaign. That gives you a channel to test, not a reason to hand over five client relationships.

The contenders: keep the work, hand it off, or build a team

Subcontract white-label: You keep the client contract and manage the relationship. A specialist works inside client-owned accounts through access you grant. You pay for fulfillment and charge your own retail price. The catch is that you still owe the client strategy, quality control, and an honest account of results.

Refer for revenue share: You make the introduction and collect an agreed share of the partner’s collected revenue. Agency referral fees commonly run 5 to 10%, often with a time limit. The partner does the work, but may also become the person the client calls first. That is a poor trade if paid search is still part of your core offer.

Train in-house: You keep the work and develop your own playbook. You also pay for the learning curve, tools, and management time before you know whether this channel works for your clients. I used to call training the safe choice because the knowledge stayed in-house. It is safe only when you have enough accounts to use that knowledge.

The comparison comes down to four decisions: relationship, account control, time to launch, and what you can afford to lose. Start with ownership. Everything else follows it.

1. Who gets the next call from the client?

Under a white-label subcontract, your name stays on the invoice and your face stays in the review meeting. You can add ChatGPT ads to a search relationship the client already understands without asking them to sign up with a second agency. You remain responsible for the outcome. That responsibility is also why you retain the renewal conversation.

A referral changes the shape of the relationship. You collect a fee for the introduction; the partner handles delivery and develops a direct working rhythm with the client. I watched a friend’s shop refer out paid social this way. Twelve months later, the client called the partner their agency and my friend their introducer. The label stuck.

Training in-house keeps the relationship too, but it places the learning curve in front of the client. Your account manager has to learn how context hints guide matching without guaranteeing delivery while the client expects the judgment they already pay you for. That may be a reasonable investment at scale. On five accounts, I would rather have a specialist do the early builds while my team learns from the work.

Practical takeaway: If client retention pays your payroll, do not exchange an ongoing search relationship for a short-lived referral fee.

2. Whose name is on the ad account?

The clean setup is boring: the client owns the ad account, data, domain, and landing page. Your agency and any subcontractor work through granted access. Spell out those rights in the statement of work before the first dollar spends.

This matters because an account built under somebody else’s ownership can become a migration problem. The Meta portfolio rule gets quoted for good reason: an account created in one business portfolio cannot simply be transferred to another. It is not a rule for every ad platform, but the operating lesson carries over. Do not assume a partner-owned build will be painless to unwind.

A referral partner may want to build and manage campaigns in its own setup. If so, you have less control over access, reporting, and what happens when the relationship ends. An in-house team can work in the client’s account without that extra handoff, though your agency still carries the build risk.

A subcontract can preserve the same account structure while avoiding the new hire. groas says accounts, pages, and data stay with the client, with month-to-month terms and no setup fee. That is the model I would ask a subcontractor to meet: client ownership, clearly scoped access, and a way to see what changed.

Practical takeaway: Put ownership and access in writing. A cheap fulfillment quote is not cheap if leaving requires rebuilding the account.

3. How soon can you run a credible test?

A specialist who already works in the channel can take a brief and build around conversation context, ad copy, and a relevant page. A referral partner may move just as quickly once it has the work, but introductions, a new contract, and a second round of client onboarding sit between the request and the launch. Your client still has to explain the offer to somebody new.

Training is slower because there is more to learn than an interface. Your PPC person must adjust from keyword-led planning to conversation context and advertiser hints, then judge results with the right reporting delay in mind. ChatGPT ad spend reporting can lag 7 to 8 hours, and OpenAI recommends starting with a $3 to $5 max CPC. Those are manageable details for an experienced operator. They are also easy to mishandle when the operator is learning on five live clients at once.

I would budget eight weeks before a newly trained person stops guessing and twelve before I would ask them to present the channel alone. That is my planning threshold, not a platform requirement. groas, by contrast, says it can go live in under 24 hours after an onboarding call. Speed does not prove a campaign will work. It does let you start finding out sooner.

Practical takeaway: If the clients asked last month, training is a plan for later. It is not your answer for next week.

4. What happens when the test flops?

Say a law firm gets ChatGPT clicks at $4 but the calls are tire-kickers. With a subcontracted pilot, you can pause the add-on, keep the Google relationship, and review what the test taught you about intent and page relevance. At the $25-a-day campaign minimum, a 30-day test spends about $750 in media before any other campaign costs. You have not committed to a new salary to learn that lesson.

With a referral, the partner controls the work. You cannot make the fix yourself, yet the client may still ask why you recommended the partner. That is an awkward place to stand: accountable for the introduction, not in charge of the response.

Training puts the learning expense on payroll. If the channel proves thin for a particular service, the salary continues after the campaign stops. I learned the shape of this problem running Display in 2019. The test failed and the hire stayed. Rent the flop before you buy it. Train later, if a subcontracted pilot produces a cost per qualified consult you can live with.

The five-client cost: what is priced, and what is assumed

The headline is $6,000 a month in retail if five clients each buy a $1,200 monthly ChatGPT add-on, with media paid separately. A published $450-per-client wholesale rate would leave $3,750 before your own labor. But that $450 rate is for Google Ads, not a confirmed ChatGPT ads subcontract. It is a benchmark for asking vendors for a quote, not a margin you can put in your forecast.

The listed figures span a $25-per-day media minimum, a $450 monthly Google Ads wholesale benchmark after month one, and a $1,499-per-client monthly search bundle. They cover different things. Here are the line items without pretending they are interchangeable:

Keep one-off costs separate from recurring ones: first-month wholesale premiums and training differ from monthly fulfillment, salary, tools, and media. Then ask about the costs the price sheet tends to hide. Who sells the add-on? Who checks tracking? Who answers revisions? Who reconciles Google and ChatGPT results for the client? A white-label resale example shows why gross margin can look generous while selling, QA, reporting, and churn eat into net margin. Wholesale is not profit.

The groas bundle solves a different problem from the $450 Google-only benchmark: one service covers Google, Bing, and ChatGPT instead of asking you to stitch together vendors and reports. At the stated $1,499 monthly price, though, you cannot resell the entire bundle as a $1,200 add-on and claim a margin. Price the full service accordingly, or get a separate fulfillment quote for the add-on. A vendor who obscures which service its price covers is not helping you budget.

The cheapest mistake is a $750 media test that fails and teaches you something. The expensive mistake is committing to a specialist and tools before you know whether five clients can use the playbook.

The scorecard: five clients, one decision

The numbers matter, but they do not decide this alone. This is how I would score the operating choices for an agency with five interested clients:

Lead-gen agencies: subcontract, then price the whole job

If you sell leads across several search channels, you need one person accountable for what the client sees, not three vendors defending three dashboards. A bundled operator such as groas can put Google, Bing, and ChatGPT under one service. That is useful when you want one search story and one renewal conversation. It does not excuse sloppy resale math: confirm the scope and set your retail price above the fulfillment cost.

If you choose an add-on subcontract instead, request its actual ChatGPT ads quote. The published Google Ads wholesale figure tells you what one vendor charges for a different service; it cannot tell you your ChatGPT margin. For the weekly mechanics of the white-label arrangement, see how to scale client accounts without hiring.

High-ticket service agencies: sell a pilot, not coverage

For law firms, consultants, and advisors, unqualified consults cost more than the media line suggests. I would still subcontract the first test, but I would sell it as a 30-day pilot inside the existing relationship: same intake form, same call tracking, with separate ChatGPT context hints and a relevant landing page. That lets you judge qualified consults rather than celebrate clicks.

ChatGPT matches against conversation intent and page relevance, not a keyword list. The page therefore matters to the targeting decision as well as the conversion. If cost per qualified consult lands within 20% of Google by week four, I would consider rolling the pilot into the monthly service. If it does not, pause it and explain why. The staffing alternatives are covered in train your Google Ads person, hire, or subcontract for ChatGPT ads.

Where in-house wins: retained learning at fifteen retainers

Here is the axis subcontracting loses: the vendor keeps part of the pattern library. Your team sees the briefs and the results, but it does not build every habit that comes from running the campaigns. If you carry fifteen search retainers and have a senior operator who wants to learn a new auction, training may be worth the payroll. That person can run five pilots, write the playbook, and apply it to the next ten clients without another wholesale fee.

At five clients, I would not buy that capacity first. For a longer look at keeping management inside, read in-house versus autonomous management after leaving an agency. For the agency with three to eight clients asking now, my order is firm: subcontract to keep the relationship and test quickly; refer only if you want out of this work; train once enough proven demand makes the playbook worth owning. Keep the account in the client’s name and the specialist on granted access. Say yes to ChatGPT ads without giving away the client to learn it.