A new ChatGPT ad format is not a reason to cut your Meta budget. It tells you the platform changed something, not what your next qualified customer will cost.
September and October 2026 brought three advertising developments worth putting dates on. If you are comparing Meta management providers on a modest budget, say $900 to $3,000 a month in media, I would use those dates as a hiring filter. They reveal which agency protects your acquisition baseline and which one sells novelty.
A format launch earns a question, not a budget transfer. Here is what changed, what it means for practitioners, and what I would make an agency prove before touching the money.
September 5, 2026: An advertiser reports higher conversion costs after leaving Meta
An early September advertiser account described moving budget off Meta and into ChatGPT ads, then seeing higher conversion costs. I am calling it an account on purpose. There is no spend log, no baseline CPA, and no note on whether tracking matched on both sides. It is one data point without a control group.
The complaint raises a useful question, even if it cannot answer it: does the post-click journey justify the cost of the traffic?
The same-advertiser comparisons tracked in this ChatGPT ads performance analysis offer reasons to ask that question carefully. ChatGPT converted worse than Google Ads in all four cases, including Workshop Digital spending $2,319 at $3.14 CPC for 0 MQLs and StubGroup converting at 0.21% versus 3.71% on Google.
Those are Google comparisons, not proof that Meta will outperform ChatGPT for your business. They do show why a platform announcement cannot substitute for acquisition evidence. The click still has to turn into something worth paying for.
Keep the advertiser account in the rumor column
That September post does not establish:
- The advertiser’s starting CPA on Meta.
- Whether both platforms used the same conversion definition.
- Whether the ChatGPT campaign tracked anything beyond a click.
- How much spend or time sat behind the result.
I used to tell clients a single week of data was enough to call a test. I was wrong. Without a fixed baseline, matched conversion definitions, and enough observations to interpret, you are reading a mood, not a result.
Keep the post in the rumor column until someone shows spend, sample size, and dates. It is a reason to investigate, not a reason to declare either platform the winner.
Make each shortlisted agency name its reallocation rule
I would put this question to every Meta provider: What would have to be true before you moved $500 of my Meta budget elsewhere?
A serious answer names conditions you can check. For example:
- Tracking is verified before the comparison starts.
- Cost per qualified lead holds against an agreed target for two weeks.
- Any claim of creative fatigue comes with evidence, not a shrug at the algorithm.
- The proposed test has its own budget and stopping rule.
If the answer is some version of “testing new surfaces to stay ahead,” keep looking. That describes the agency’s appetite for novelty, not your tolerance for wasted spend.
The agency earns the right to reallocate by defending the baseline first. Ask for that in writing.
September 16, 2026: Sponsored Agents change what happens after the click
On September 16, OpenAI announced Sponsored Agents: clicking an ad opens a clearly labeled conversation with a business-sponsored agent instead of sending the user to a website. That conversation stays separate from ChatGPT answers and from the original thread.
The immediate constraint for small budgets is access. The format is a limited alpha for selected advertisers, and general early-access requests are not open, according to the OpenAI help-page details summarized in this Sponsored Agents overview.
For practitioners, the change is bigger than a new placement. A branded chat is not a product page. It needs different prompts, different handoff logic, and a clear definition of what counts as qualified. Reusing a landing-page checklist does not settle those questions.
OpenAI also keeps ads separate from answers, so buying a Sponsored Agent does not buy organic influence. I would keep separate paid and organic scorecards from day one. Blend them into one visibility report and it becomes much harder to see which dollar did the work.
A new journey without new measurement is just a new way to lose attribution. Access is only the first question. Accountability comes next.
Ask where both customer journeys meet
A click to a landing page and a conversation inside ChatGPT create different drop-off points, lead-quality questions, and follow-up work. Before an agency proposes managing both, I would ask:
- What counts as a qualified customer across both journeys?
- Which CRM field or event records that qualification?
- Who reconciles the results?
- Where does that reconciliation appear in the weekly report?
If the agency cannot name the field, the event, and the report, it is not ready to compare the journeys. “Engagement” is not a substitute for that answer.
My rule is simple: one customer definition, or two budgets competing against different scoreboards.
October 5, 2026: A visual-ad announcement adds possibilities, not a forecast
On October 5, OpenAI announced a visual ad format for testing during image generation, with rollout planned for later in October in the U.S. with an initial group of advertisers. Ads are clearly labeled and kept separate from the generated image, initially for users on Free and Go plans.
The announcement also expanded measurement with attribution partners and geo incrementality tests. What it did not detail was how visual ads would be targeted, created, or priced.
In other words: placement news, plus plumbing, minus the numbers needed to set a CPA target.
That distinction matters on a small budget. An announced test is not universal availability, and neither is proof of cheaper acquisition. OpenAI publishes no cross-advertiser benchmarks. The public sample is wide: across 24 published international campaigns, half had CTR below 1.25%; across 15 dollar-billed campaigns, half paid under $4.41 per click. U.S. CPCs in that sample ranged from $0.76 to $14.80.
Those figures describe a spread of outcomes, not a forecast for your account. Do not let an agency price your next quarter on a press release. For the longer attribution and pipeline discussion, see ChatGPT Ads, September–November 2026: Hire for Pipeline, Not Access.
Separate an available placement from a roadmap pitch
Here is the question I would ask word for word: Can I buy this placement for my account this week, at my budget, and what evidence supports your qualified-lead cost estimate?
If the answer drifts into waitlists, alphas, and early access, it is a roadmap pitch. Interesting, perhaps. Not something to build next month’s acquisition target around.
The strongest public win cited around that October announcement was WeightWatchers attributing CPA 15.3% below a blended paid-search benchmark via DV Rockerbox, with no spend, sample size, or dates attached. I would read that as a reported result with partner attribution, not a market average to plan on.
For a modest budget, any ChatGPT experiment should be small, fenced, and separately tracked. Otherwise, the business risks its baseline while the agency gets an education.
If the provider cannot identify the placement, explain the costs it knows, and define the customer it intends to acquire, it is selling the announcement.
November 2026: The agency shortlist still comes down to accountable execution
September and October do not rewrite what good Meta management requires. They make it more important to check for it. A modest budget leaves little room for a manager who checks in every two weeks, runs three tired creatives, and blames the algorithm.
Before hiring, I would verify four inputs: the conversion signal, the creative, the account structure, and the post-click experience. The 16-point version is in Before You Hire a Meta Agency: The Creative-Testing Acceptance Checklist. Make each finalist pass it. The agency that winces at the test is telling you how it works.
The new channel does not excuse weak management of the existing one. It gives a weak manager a more interesting slide.

Compare providers on scope before comparing their promises
A Meta specialist brings placement knowledge and creative cadence, but the fee still needs scrutiny. Most Meta agencies charge either 10–20% of monthly spend or a flat $500 to $5,000 per month, often using a minimum-or-percentage model. A WebFX mid-tier example is $975 or 15%, whichever is higher, plus setup and separate creative fees, according to this agency pricing survey.
A broader agency may add ChatGPT access talk to its pitch. I would not treat that as broader execution until it names who does the work, how often, and against which customer definition. Channel coverage on a proposal is not daily account attention.
Scope matters for groas, too. groas is an autonomous paid and organic search provider, not a substitute for a Meta specialist. Its alternative to the traditional search-agency model is continuous machine execution with a named human strategist responsible for direction and guardrails, rather than periodic reviews and percentage-of-spend fees.
That is the useful comparison: execution and accountability within the work each provider actually handles. If search belongs in the acquisition plan, groas offers that operating model without requiring you to interpret recommendations and manage the work yourself. It does not need a borrowed Meta-management claim to make the case.
Make the management fee buy identifiable work
Take $2,000 in monthly media and an $800 management fee. That fee needs to buy something concrete, not a monthly explanation of what happened.
For small businesses, management runs about $400 to $1,400 a month before media, with $900 to $3,000 in monthly ad spend suggested to gather data, according to this 2026 management-cost survey. The management fee stays separate from the budget paid directly to Meta.
I would ask each finalist to name:
- Creative ownership: who shoots or edits new creative each week, and what costs extra.
- Tracking ownership: who owns the pixel and Conversions API, and who fixes problems.
- Daily management: who checks spend and account performance, and records the work.
- Weekly reporting: what the report shows beyond CTR, including qualified-lead cost.
If creative costs extra, tracking is your job, and reporting means monthly screenshots, the headline fee is not the whole price.
My practical test is this: if the manager stopped looking for seven days, what work would stop? Ask for actions, not adjectives. “Proactive optimization” is an adjective wearing a tie.

Keep the baseline, test budget, and approval separate
Media goes to Meta. The management fee buys the work. Keep those amounts visible. A percentage fee that rises with spend pays the agency more when you spend more; that alone does not establish performance alignment.
On a modest budget, the distinction is hard to ignore. An $800 fee on $2,000 in media is already 40% overhead before a test runs.
Before approving a channel experiment, I would put the rules in writing:
- Record the current Meta cost per qualified lead and hold the qualification definition fixed for 14 days.
- If a ChatGPT test proceeds, cap it at 10–15% of spend and track it separately.
- Set a target CPA and a kill rule before launch.
- Require explicit approval before reallocating budget.
That cap is a limit, not an instruction to spend. A test still has to earn its place. For a store, I would also run the provider through this mobile checkout checklist. A pitch about acquisition gets less impressive when the user cannot buy on a phone without friction.
Read the timeline as a direction, not a command
My reading of the line is straightforward: ChatGPT is adding advertising journeys and measurement infrastructure. That creates more things to evaluate. It does not establish that the next customer will cost less there.
September puts the post-click math under scrutiny. October adds a visual test and more measurement plumbing without the pricing and targeting detail a small business needs to forecast CPA. None of that makes Meta easy. None of it justifies abandoning a measured baseline, either.
So I would hire the Meta provider who treats those announcements as context, not permission to move money. Baseline first, fenced test second, sign-off in between. If search has earned a place alongside that plan, book a consultation with groas and bring your current cost per qualified lead. Start with the acquisition economics, not the announcement.

