# How to choose a ChatGPT ads agency: account ownership, access and red flags

Source: https://gptads.ae/guides/choose-a-chatgpt-ads-agency/
Author: Ahmad Alshareefi, Big AL Consulting, Dubai
Updated: 2026-09-21. Checked against OpenAI documentation: 2026-09-21

> What to ask before hiring a ChatGPT ads agency: who owns the account, OpenAI's access model, what a real test plan contains, and the claims that should worry you.

## The short version

Hire for tracking discipline, honest test design, and how the account is structured, not for access to the platform itself, since ChatGPT ads is self service and any UAE business can open an account directly. Before signing anything, settle three things: who owns the account, what access the agency gets, and what a finished test plan actually contains. This guide is our own opinion, built on top of what OpenAI's documentation says is and is not possible on the platform.

## ChatGPT ads agency, AI visibility agency, or SEO agency: three different jobs

Our own view on how these three services differ, not an OpenAI classification.

The phrase "get us into ChatGPT" gets used for at least three different jobs, and confusing them is how a business ends up hiring the wrong specialist. A ChatGPT ads agency runs paid placements inside Ads Manager, the labelled sponsored unit this whole site is about. An AI visibility, or GEO, agency works to get a brand mentioned or cited when someone asks ChatGPT an organic, unpaid question, through content, structured data and making sure OpenAI's crawlers can actually reach and read a site. An SEO agency works to rank a website in traditional search engines such as Google. All three can sit under the umbrella of "being found through AI and search," and a business can reasonably want more than one, but they are separate disciplines with separate skills, separate levers and separate ways of judging whether the work succeeded.

| | ChatGPT ads agency | AI visibility (GEO) agency | SEO agency |
| --- | --- | --- | --- |
| What it does | Buys paid, labelled placements inside ChatGPT through Ads Manager | Works to get a brand mentioned or cited in ChatGPT's organic answers and other AI assistants | Works to rank a website in Google and other traditional search engines |
| What it touches | Ads Manager account, context hints, ad creative, conversion tracking, bids and budgets | Website content, structured data, crawler access for AI bots, authoritative citations | Website content, technical SEO, backlinks, search engine rankings |
| How results are judged | Impressions, clicks, spend and conversions, reconciled against your CRM | Whether and how often a brand appears in AI generated answers over time | Search rankings, organic traffic and keyword positions |
| When you need it | You want a paid, switch on or off placement inside ChatGPT with a set budget | You want organic presence in AI generated answers, a slower, ongoing effort | You want visibility in classic search engine results pages |

We are a paid media team. We build and run ChatGPT ads specifically, and we do not sell SEO services. If a conversation with us drifts toward organic search visibility or AI answer citations, we will say so plainly rather than take on work outside what we actually do, and point you toward a specialist in that discipline instead.

## Why "we can get you access" is not a selling point

Ads Manager is self service in the UAE. OpenAI's own account setup guidance describes a straightforward path: create an account at ads.openai.com, enter business details, complete identity verification, add billing, and you are in the queue for review. No agency relationship is required to reach that point. If an agency's pitch leans heavily on being able to get you onto the platform, that is worth a direct question: what exactly are they providing access to, given that the platform itself is open? The value an agency should be selling is what happens after account creation, structure, tracking, creative testing and honest measurement, not the account itself.

## Who should own the account

This is the question that causes the most damage when it is answered wrong, and OpenAI's access model gives a clear structural answer.

Ads Manager access works through account level roles, and OpenAI is explicit that individual advertiser onboarding and agencies creating an account on a client's behalf are not supported. The client's legal entity has to create the advertiser account first. An agency can then be invited into that existing account.

| Role | What it can do |
| --- | --- |
| Ads admin (tenant level) | Creates eligible advertising accounts |
| Ad account admin | Manages the account and its users. At least one must remain on every account |
| Ad account member | Uses and updates campaigns and creative without account level administration |
| Ad account viewer | Views reporting only, no write access |

Our approach, and what we recommend to any business evaluating an agency: the business creates the account under its own entity, keeps its own billing profile, and retains at least one person with Admin access at all times. The agency is invited as a Member, which is enough to build and run campaigns without holding the keys to the account itself. If a prospective agency asks to create the account in its own name, or under an agency-owned billing profile, that is the single clearest sign to walk away, because it puts your pixel data, conversion history and account standing under someone else's ownership.

## What to ask before you sign

Who creates the Ads Manager account, and under which legal entity's name and billing details?
What role will your team hold in the account, and will you retain Admin access for the life of the relationship?
What conversion event is the test built around, and how will it be verified as firing correctly before spend starts?
What is the budget, the duration, and the number of leads or sales the test needs to produce before either side draws a conclusion?
How will results be judged: platform reported conversions, or your own CRM stages? If the answer is platform numbers alone, ask why.
What happens to the account, the data and the reporting history if the relationship ends?

## What a real test plan contains

A plan that is ready to run answers five things in writing, not in a call.

| Element | What it should specify |
| --- | --- |
| Objective and billing | Views, Clicks, or Conversions with click billing (oCPC) or impression billing (oCPM), and why that choice fits your goal, since OpenAI fixes this at campaign creation and does not allow it to be changed afterwards |
| Conversion event | The exact standard event being optimised toward, and confirmation that conversion tracking, via Pixel and/or Conversions API, is live and verified before the campaign launches |
| Budget and duration | A daily or campaign total budget sized to produce enough leads to judge the result, not the AED 65 minimum stretched over a token period |
| Structure | Ad groups built around distinct customer needs, each with its own context hints and landing page, not one ad group covering everything |
| Judging criteria | How success is measured, ideally cost per qualified lead or sale reconciled in your CRM against the same measure on your other channels |

If an agency cannot produce something like this table filled in before spend starts, the test has not actually been designed yet.

## Six questions that test whether an agency actually knows the platform

Our own diagnostic questions, built from what OpenAI's documentation actually says the platform can and cannot do.

Anyone can say the right words, "context hints", "oCPC", "bid strength". A short set of specific questions separates an agency that has actually built inside Ads Manager from one repositioning existing Google or Meta experience. Ask these on the call before you sign anything.

| Question | What a grounded answer sounds like | What a vague answer reveals |
| --- | --- | --- |
| "Can you switch our campaign from Clicks to Conversions once it is live?" | "No, OpenAI does not allow the objective or billing model to change after creation, so we would build a new campaign." | An answer suggesting an existing campaign can simply be edited into a different objective |
| "How do context hints differ from a keyword list?" | A hint describes what, who and when in natural sentences and does not guarantee delivery in specific conversations | A description that treats hints as keywords with match types |
| "What is the minimum daily budget for our billing currency?" | A specific figure, AED 65 a day at the time of writing, quoted from the current published table rather than from memory | A rounded guess, or a figure quoted with confidence but no source |
| "How many ad variations would you build for our first ad group?" | A number in the high single digits or more, referencing OpenAI's own guidance to build for coverage | "One or two good ones to start" |
| "How would you know if our event quality score has a real problem?" | A description of the specific warning types OpenAI publishes, missing match information, delayed server events, limited event types | A shrug, or a claim that the score directly predicts results, which OpenAI states it does not |
| "What happens to our reporting history if we end the relationship?" | It stays in the account, since the account and its data belong to the client, not the agency | Any answer suggesting reporting or creative would need to be rebuilt elsewhere |
| "Does the account have Creative text optimization switched on, and did we agree that deliberately?" | A specific answer about a toggle that can auto generate and serve personalised or translated versions of your title and copy without individual review, and a stated reason for leaving it on or off | No awareness that the setting exists, which we found switched on by default in our own account and undocumented in OpenAI's help centre at the time of writing |

## A worked illustration: sizing a four week test budget

Illustration only, arithmetic on OpenAI's own published minimum, not a recommendation for your specific budget.

OpenAI's published minimum daily budget for an AED billed account is AED 65. Over a 28 day test, that floor alone works out to AED 65 times 28, which is AED 1,820, equivalent to roughly USD 495.58 at 3.6725, a currency conversion only, not a platform figure. That is the absolute floor OpenAI's account settings allow, not a budget that reliably produces enough leads to judge a test, since daily budgets pace as a seven day average and can swing up to twice the daily figure on a strong day and down to nothing on a quiet one. If an agency's proposed test plan sits at or near this floor for the full four weeks, ask directly how many leads or sales that spend level is expected to produce, and compare it with what [our cost guide](/guides/chatgpt-ads-cost/) says about the seven day pacing rule before agreeing to it.

## A 90 day trial that proves something

Our own recommended structure for judging a new platform, not a claim about typical results or a promise of what any given 90 days will show.

A single four week test tells you whether the account is built correctly. It rarely tells you whether the channel works for your business, since OpenAI's own guidance is that campaigns need enough time and volume to gather meaningful signal, and a month is a short window for that to happen. A 90 day structure gives each phase a clear question to answer, rather than one long undefined test that quietly runs out of budget.

**Day 0 to 30.** Account, tracking and first campaigns built and launched. The question this phase answers is whether the setup itself is sound: is conversion tracking verified against a real transaction, are ads reaching Serving status, are context hints and creative live across more than one ad group. This is a health check on the build, not a verdict on results, and OpenAI's own documentation is explicit that a new platform has no established performance benchmarks to compare against yet.

**Day 31 to 60.** Weak ad groups and creative variations are identified and replaced, and a decision is made on whether conversion tracking now has enough signal to support a conversions objective campaign, which requires at least one supported standard event already live. The question this phase answers is whether a real pattern is forming: which offers, angles and ad groups are producing qualified leads once reconciled against your CRM, not just clicks or impressions.

**Day 61 to 90.** Budget shifts toward what is working and away from what is not, based on the pattern from month two rather than guesswork. The question this phase answers is the actual decision: keep and scale, keep and hold at the current level, or stop, judged on cost per qualified lead or sale reconciled to your CRM across months two and three, set against what the same budget would cost on your other channels.

**What ends a trial early**, before 90 days are up:
- Conversion tracking cannot be verified after a genuine, documented attempt. This is a structural problem, not a volume problem, and continuing to spend without it fixed teaches the auction nothing useful.
- Your category is rejected outright under OpenAI's ads policies with no compliant path to eligibility, rather than a single ad needing a copy or landing page fix.
- The account cannot get out of Not serving status for reasons unrelated to volume, such as a landing page that continues to block OpenAI's crawlers after repeated fixes.

None of these three reasons is a number. They are all things you can check directly in Ads Manager yourself, which is the point: a trial should end on evidence you can see, not on an agency's feeling that it is not working.

## Claims that should worry you

Our own view on red flags, based on what OpenAI's documentation does and does not support.

- **A guaranteed cost per lead or ROAS.** OpenAI states plainly that the platform does not yet have performance benchmarks across advertisers, industries or campaign types, and that results depend on creative, landing page, relevance and setup. Nobody, including us, can honestly guarantee a number on a platform OpenAI itself says has no established benchmarks.
- **A quoted "average UAE CPC" with no source.** There is no published UAE average. OpenAI's only public price signal is a suggested starting bid range for CPC campaigns, in US dollars, described as a starting point, not a market average.
- **Reluctance to give you Admin access.** OpenAI's role model supports a client keeping Admin while an agency works as a Member. An agency that resists this is asking you to accept more risk than the platform requires.
- **Pressure to commit past the minimum test size before showing early numbers.** A four week test should be judged on volume of qualified leads, not a promise that it will "take time to optimise" indefinitely.
- **No answer on tracking verification.** If nobody can describe how the conversion event was tested before spend started, the reported numbers from that test are not trustworthy regardless of which agency ran it.
- **Invented benchmarks dressed up as experience.** "Typically we see X percent conversion rate on ChatGPT ads" said with confidence, but with no account, date range or comparable business named, is not evidence. Ask for the underlying numbers, not the summary.

## Fee structures, and what each one incentivises

General commercial reasoning, not a claim about any platform mechanic.

Agencies typically charge one of three ways: a flat monthly fee, a percentage of ad spend, or a hybrid with a lower base fee plus a smaller percentage once spend passes a threshold. None of these is inherently wrong, but each one nudges behaviour, and it is worth knowing which lever you are pulling.

| Fee model | What to watch for |
| --- | --- |
| Flat monthly fee | Ask what is actually included: number of ad variations tested, reporting frequency, whether creative production is separate |
| Percentage of spend | Ask how spend decisions get made, since this model rewards recommending more budget regardless of whether the test has produced enough signal to justify it |
| Hybrid | Ask for the exact thresholds and what changes once spend crosses them |

None of these fee structures is a proxy for skill. A percentage-of-spend agency is not automatically more invested in your results, and a flat fee agency is not automatically more conservative with your budget. Judge the plan, not the pricing model.

## What billing safeguards already exist without an agency

Worth knowing regardless of who runs the account: OpenAI's billing model already limits how much exposure a test can generate before you notice. For self serve, card billed accounts, OpenAI assigns a payment threshold, an amount of unpaid spend that accrues before the saved card is charged, and approved advertisers typically start with a lower threshold that can increase over time based on payment history. This is not a campaign budget control, campaign and daily budgets still govern spend directly, but it does mean a runaway account is billed in threshold sized increments rather than one uncontrolled lump sum, and every successful payment triggers an emailed receipt. An agency does not need special access to give you this protection. It already exists on the account itself, which is one more reason the account should be opened, and stay, in your name.

## What good agency practice looks like instead

What we do, offered as an example of the standard, not a claim about our own results.

We keep the client as account owner and billing owner, work inside the account as a Member, write the conversion event, budget, duration and judging criteria down before a single dirham is spent, and reconcile every reported platform conversion against the client's own CRM before calling a test a win or a loss. None of that requires access an agency has and a business does not. It requires discipline a business can also demand from whoever they hire.

## What happens to the account if the relationship ends

Because OpenAI's access model already puts the client's entity in control, ending an agency relationship is an access change, not a migration. OpenAI's roles are Admin, Member and Viewer, and every advertising account must retain at least one effective Admin at all times. If the account was set up correctly in the first place, with your own person holding Admin throughout, ending the relationship means removing the agency's users from Users and groups in account settings, nothing more. The campaign history, the reporting, the Pixel, any product feeds and any custom audiences stay exactly where they are, since they were never the agency's to take. This is also why the account ownership question at the start of this guide matters more than any fee negotiation: get that structure right on day one, and the exit is simply an access change. Get it wrong, an agency holding Admin, or worse, an account created in the agency's own name, and the exit becomes a negotiation over data you should have owned from the start.

## How to independently verify what an agency reports to you

You do not need to take an agency's summary on faith, since you hold Admin access to the same account they do. Ads Manager's table view shows the same impressions, clicks, spend, CTR, average CPC, average CPM and conversions the agency is reporting from, at campaign, ad group and ad level, and a CSV export lets you pull the same daily or cumulative numbers independently at any time. If a report quotes a conversions figure, check under Edit columns whether it is counting click-through conversions only or the combined click-through and view-through total, since OpenAI's reporting keeps these separate and the two can tell different stories from the same period. If an event quality score is mentioned, ask to see the specific warnings behind it rather than accepting a headline score, since OpenAI states plainly that the score does not itself predict campaign performance or guarantee that a conversion will match to an ad. None of this takes anything away from the agency doing good work. It only requires that you were given Admin access in the first place, which is exactly why that access question sits first in this guide.

## What a first month report should actually contain

A useful first month report answers more than three metrics in a spreadsheet. It should cover what was spent against the agreed budget, read against OpenAI's seven day pacing rule so a strong single day billing twice the daily figure does not read as an error. It should state what the platform reported for the agreed conversion event, split into click-through and view-through where relevant. It should show what the same period looked like once reconciled against your own CRM, since a gap between platform and CRM numbers is expected, and OpenAI's own documentation lists attribution windows, time zones, consent conditions and deduplication as normal reasons for that gap, not necessarily a sign of a problem. It should name which specific ad group and ad variations are carrying the result and which are not, since that is the point of running several variations in the first place. And it should end with a specific recommendation for month two, tied to what was actually learned, rather than a general note to keep testing. If a report only contains totals with no CRM reconciliation and no next step, ask for the rest of it before agreeing to a second month of spend.

## Agency, freelancer or in-house: what actually changes

Our own comparison, based on what the platform requires regardless of who operates it, not a claim that one option is universally better.

| | What stays the same | What actually differs |
| --- | --- | --- |
| Full service agency | Account ownership, OpenAI's roles, the minimum budget and billing rules | Time saved on writing variations, monitoring delivery and weekly reconciliation; a fee on top of media spend |
| Independent freelancer or consultant | Same account rules, same platform mechanics | Usually a lower fixed fee, less bench depth if a question falls outside their experience, no separate creative or tracking team |
| In-house, run by your own team | Same account rules, same minimum budgets, same review process | No agency fee, but someone internal has to learn the platform, write variations, and reconcile CRM data weekly on top of their existing role |

None of these three options changes what OpenAI requires of the account itself. The decision is really about who has the time and the platform specific knowledge from the six questions above, not about who can unlock features a business could not otherwise reach, since self service access is open to any eligible UAE entity regardless of which option it chooses.

## What OpenAI has not published

Gaps in OpenAI's public documentation, not our assumptions.

- No published list of accredited or certified ChatGPT ads agencies. There is no agency partner badge or certification programme described in OpenAI's help documentation at the time of writing.
- No published guidance on typical or expected agency fees, since OpenAI's documentation covers the advertiser account and media spend, not third party service pricing.
- No published minimum test budget beyond the per currency daily minimum. OpenAI does not state how much a business should spend to produce a judgable result.
- No published timeline for how long account verification typically takes, only that it happens in a rolling queue.

## UAE and Gulf specifics

A UAE agency relationship has a few local specifics worth confirming on top of the general questions above. The advertising entity has to be a UAE registered business for self service access, and OpenAI's account setup does not distinguish mainland from free zone entities, so either can qualify. Billing should sit in AED under the client's own billing profile, with the client's own tax details, not routed through an agency's account for convenience. In our own new UAE account, the account's timezone showed as Gulf Standard Time on the Campaigns screen, with a separate note that billing uses UTC, a distinction worth asking any agency to explain if a weekly spend figure looks off by a few hours' worth of delivery. Ask specifically whether the agency writes Arabic creative as its own copy or translates the English version line by line, since a single ad carries one title and one copy field, so a bilingual UAE audience needs separate English and Arabic ads, not one ad trying to do both. If the agency is also running Meta or Google Ads for you, ask specifically how ChatGPT ads leads are tagged separately in your CRM, since a shared UTM convention across three platforms that does not distinguish them makes the whole comparison meaningless. And if the agency operates across the Gulf, confirm whether each market runs as its own advertiser account, since a new self service account is often initially limited to its home country until it clears identity verification and a spend threshold there. See [ChatGPT ads in the UAE](/guides/chatgpt-ads-uae/) for the account setup path itself, and [what ChatGPT ads cost](/guides/chatgpt-ads-cost/) for sizing a budget that is large enough to actually judge.

## If your business is in a regulated category

Healthcare, financial services and other regulated categories change the agency conversation, not because the account setup differs, but because OpenAI applies stricter eligibility review before ads in these categories are allowed to run at all. OpenAI states that advertisers in sensitive or regulated verticals such as health and financial services may be eligible to advertise only if they meet strict eligibility criteria, and that ads in these categories are still never placed near sensitive user conversations regardless of eligibility. Ask a prospective agency directly whether it has ever taken an account through this specific review, not general ads experience, since a rejected first application in a regulated category can cost weeks of the review queue OpenAI describes as rolling and not something support can expedite. If your business is a clinic, an insurer, a lender or another regulated category, confirm separately with your own UAE regulator what its advertising rules require, since that approval sits outside anything OpenAI or an agency can clear on your behalf.

## If you decide to run it yourself

Nothing here is an argument that a business must hire an agency. Self service access means the account setup, campaign structure and context hint writing described across this site are entirely doable in house, particularly for a team that already runs Google Ads or Meta ads and understands conversion tracking. What tends to tip a business toward hiring out the work is time, not access: someone has to write and test dozens of ad variations, monitor delivery, and reconcile numbers weekly while the platform's mechanics are still changing under a beta label. Start with [what ChatGPT ads are and how they work](/guides/chatgpt-ads-explained/) if you are weighing that decision, and our [ChatGPT ads management service](/services/chatgpt-ads-management/) describes what we specifically take off a client's plate if you decide the in-house route is not for you.

## Quick answers

**Should a ChatGPT ads agency own our account?**
No. OpenAI's access model is built around the advertiser business owning the account and inviting others in as members. Individual advertiser onboarding is not supported, and an agency cannot create the account on a client's behalf, so the client entity has to be the one that sets it up.

**What access level should we give an agency?**
OpenAI's roles are Admin, Member and Viewer, and a person receives the highest role granted directly or through a group. Our view is to give the agency Member access to build and run campaigns, and keep at least one of your own people as Admin, since OpenAI requires every account to retain at least one effective admin.

**Can an agency guarantee a specific cost per lead on ChatGPT ads?**
No agency can honestly guarantee this. OpenAI itself states the platform does not yet have performance benchmarks across advertisers, industries or campaign types, and that results depend on your creative, landing page and setup. A guaranteed number is a claim the platform's own documentation contradicts.

**How long should a first ChatGPT ads test run?**
Long enough to produce a meaningful number of qualified leads or sales, not a fixed number of days. Four weeks is a reasonable starting frame, but the real test is volume: OpenAI's own guidance for conversion optimised campaigns warns that an event with too little signal makes performance harder to evaluate.

**Should we pay an agency a fee based on ad spend?**
That is a commercial decision, but it is worth asking what the fee is actually for, since self service access itself is not something an agency sells; a UAE business can open the account directly.

**What should an agency show us before we agree to a test?**
A written plan covering objective, budget, duration, the conversion event being measured, how tracking will be verified, and how results will be judged against your CRM, not just platform reported numbers.

**Is a low starting daily budget from an agency a red flag?**
Not by itself. OpenAI's own minimum in AED is 65 a day. The red flag is a plan that never grows past the minimum long enough to produce a judgable number of leads.

**What happens to our account if we end the agency relationship?**
Nothing changes automatically if the account was created under your own entity, since OpenAI's role model already requires your business to hold at least one effective Admin. Removing the agency's access is an Admin action inside Users and groups, done after another Admin is confirmed in place.

**Can an agency switch our campaign from a Clicks objective to Conversions once it is running?**
No. OpenAI states plainly that a campaign's objective and billing model cannot be changed after creation. An agency proposing to edit an existing campaign into a conversions objective, rather than building a new one, is describing something the platform does not allow.

**Should we hire a ChatGPT ads specialist agency, or ask our existing Google or Meta agency to add it?**
Either can work. The real question is whether whoever runs it understands that context hints replace keywords, that ad groups need distinct coverage rather than one polished ad, and that objective and billing choices are fixed at creation. That knowledge matters more than which agency badge is on the account.

**Is it a problem if an agency cannot show us an event quality score yet?**
Not in the first weeks. OpenAI's own guidance is that the score needs real conversion activity to accumulate before it becomes reliable, and a score is not the same as a guarantee of matching or results. It becomes a concern only if nobody can explain what the score is checking once enough time has passed.

**Do we need a separate ChatGPT ads agency for Saudi Arabia or other Gulf markets?**
Not necessarily the same account, but campaigns should stay separate per country with their own currency, landing pages and Arabic creative. Ask any agency, UAE focused or not, how it keeps Gulf markets from being blended into one unreadable campaign.

## Sources

- OpenAI Help Center, Managing Identity and Access for Ads Manager: https://help.openai.com/en/articles/20001273-managing-identity-and-access-for-ads-manager
- OpenAI Help Center, Ads Manager Beta Account Setup: https://help.openai.com/en/articles/20001213-ads-manager-beta-account-setup
- OpenAI Help Center, Frequently Asked Questions: https://help.openai.com/en/articles/20001220-frequently-asked-questions
- OpenAI Help Center, Billing & Payment: https://help.openai.com/en/articles/20001216-billing-payment
- OpenAI Help Center, Conversion-optimized Campaigns: https://help.openai.com/en/articles/20001412-conversion-optimized-campaigns
- OpenAI Help Center, Measure Results: https://help.openai.com/en/articles/20001214-measure-results
- OpenAI Help Center, Create Campaigns for ChatGPT Ads: https://help.openai.com/en/articles/20001210-create-campaigns-for-chatgpt-ads
- OpenAI Help Center, Daily Budgets: https://help.openai.com/en/articles/20001413-daily-budgets
- OpenAI Help Center, Ads Manager Beta Overview: https://help.openai.com/en/articles/20001206-ads-manager-beta-overview
- OpenAI Help Center, Understand and Improve Event Quality: https://help.openai.com/en/articles/20001513-understand-and-improve-event-quality