We Scored 44 ChatGPT Ads Case Studies. Only 2 Passed Every Evidence Check
Only 2 of 44 published ChatGPT ads case studies pass basic evidence checks. We scored every case we could find on 27 October 2026 against five tests: disclosed spend, disclosed dates, a business outcome, a result checked outside Ads Manager, and a comparison channel. Both cases that passed all five show ChatGPT ads losing to the advertiser’s own baseline. None of the 44 comes from an Arab market.
Key takeaways
- 2 of 44 cases (4.5%) pass all five checks, and both report ChatGPT ads underperforming Google or organic ChatGPT mentions.
- Only 13 of 44 cases (30%) checked a result anywhere outside the ad platform’s own dashboard.
- 11 cases claim ChatGPT ads beat a comparison channel. 9 of those 11 were published by a ChatGPT ads seller or selected by OpenAI and its partners.
- Of 9 cases where ChatGPT ads lost to a baseline, only 2 came from a seller.
- 13 Arab markets can buy ChatGPT ads as of 27 October 2026, and no Arabic-language result has been published.
Why did we run this audit?
Advertisers in Riyadh, Dubai and Cairo are now pitched ChatGPT ads with case studies attached. OpenAI announced on 16 September 2026 that ads were expanding to Saudi Arabia, the UAE, Egypt, Qatar, Kuwait, Bahrain, Jordan and Morocco. The case studies arriving in their inboxes were written for US and UK budgets. We wanted to know which deserve weight.
The trigger was the UK trial. Agencies that joined it from June 2026 have started publishing what they learned, and their write-ups are mostly about setup rather than results.
One UK agency’s review reports no CTR, no CPA and no ROAS at all, only platform guidance. A global network agency’s write-up in Campaign frames the trial as something to treat differently from paid search. Neither gives an advertiser a number to plan around.
So we went the other way. Instead of collecting more claims, we graded the claims already public.
What did we expect to find?
We expected the strongest evidence to be the least flattering. Vendors publish wins. Advertisers testing with their own money publish whatever happened, and they have fewer reasons to stop measuring at the click. So a case showing the full chain from budget to sales should look worse than one that stops at the dashboard.
If a case study discloses spend, dates and a result checked outside Ads Manager, then it will report weaker ChatGPT ads performance than cases that do not.
Our working hypothesis before scoring
The hypothesis is falsifiable. If the fully documented cases had shown ChatGPT ads beating search, we would have published that instead. They did not, but the sample is small, and we say exactly how small below.
How did we score each case?
We scored 44 cases on five yes or no checks, using only what each published source discloses. Where a source was ambiguous, the case scored no. We excluded anonymous forum posts and any result that blended ChatGPT ads with Meta, Google or TikTok, because those cannot be attributed to ChatGPT ads at all.
The 44 cases came from two places. 30 came from the most complete public roundup we found, a compilation of 49 international cases by a Prague PPC agency, verified as of 29 September 2026 and updated 14 October.
The other 14 came from primary sources: five tests on a SaaS ads agency’s case study page, three anonymized cases from a ChatGPT ads agency, two OpenAI partner results published on 5 October 2026, three agency or advertiser reports we opened directly, and one trade press report.
The five checks:
- Spend disclosed: A currency figure for the test budget, not a share of an undisclosed budget.
- Dates disclosed: A start and end date, or a stated length in days or weeks.
- Business outcome: Leads, sign-ups, sales or revenue, including zero. Clicks and CTR alone do not count.
- Checked outside Ads Manager: The result was compared with a CRM, analytics, sales records or a controlled test.
- Comparison channel: The same advertiser reported a baseline: Google, Meta, LinkedIn, organic ChatGPT traffic or a geo-holdout.
We did not score whether the publisher sells ChatGPT ads, because many sources do not say. We recorded it where the page states it, and we report it separately.
What did the scores show?
Most published cases fail on verification, not on disclosure. 34 of 44 cases (77%) report a business outcome, but only 13 (30%) checked that outcome anywhere other than the ad platform. A result that lives only inside Ads Manager is the platform grading its own homework, and that describes 70% of the public record.
| Check | Cases passing | Share of 44 |
|---|---|---|
| Business outcome reported | 34 | 77% |
| Spend disclosed | 23 | 52% |
| Comparison channel | 22 | 50% |
| Dates disclosed | 20 | 45% |
| Checked outside Ads Manager | 13 | 30% |
| All five checks | 2 | 4.5% |
The distribution is lopsided. 17 cases score 2 out of 5, 10 score 3, 9 score 4, and only 2 score 5. Four cases score zero: they report a CTR or CPC with no budget, no dates and no outcome.
The gap is easy to find. In the roundup, one US advertiser’s dashboard showed 36 conversions against 2 real inquiries. A Brazilian agency saw the opposite: zero conversions in Ads Manager while three sales existed.
A case study that never left the ad platform tells you what the platform counted, not what the business earned.
Which cases passed every check?
Two cases pass all five, and both are small tests run with the advertiser’s own money. A desktop software maker billing in GBP and a brand-monitoring SaaS each disclosed spend and dates, counted real conversions, checked them outside Ads Manager and kept a comparison running. Both found ChatGPT ads converting worse than the alternative.
The software maker spent GBP 289.52 between 5 and 28 August 2026, and the ads drew a 7.9% click-through rate, the highest in the roundup, at GBP 0.10 a click. Paid visitors converted at 0.46%, against 5.3% from Google Ads and 4.2% from ChatGPT’s unpaid recommendations. Paid visits lasted 7 seconds on average. Google’s lasted 20.
The SaaS spent just over EUR 600 across two weeks at EUR 50 a day, which bought about 25,000 impressions and just under 200 clicks. It recorded zero sign-ups from the ads and 10 sign-ups from organic ChatGPT recommendations over the same period.
Two cases are not a trend. They are, though, the only two where every link in the chain from budget to outcome is visible to a reader.
Does the publisher’s business predict the result?
Yes, more clearly than any other variable we recorded. Of 11 cases where ChatGPT ads beat a comparison channel, 9 were published by an agency selling ChatGPT ads or chosen by OpenAI and its measurement partners. Of 9 cases where ChatGPT ads lost, 2 came from a seller.
| Result against baseline | Cases | From a seller or OpenAI-selected | Checked outside Ads Manager |
|---|---|---|---|
| ChatGPT ads won | 11 | 9 | 2 |
| ChatGPT ads lost | 9 | 2 | 4 |
| Mixed | 2 | 1 | 2 |
Two details make the pattern sharper. Three of the favorable cases sit on one agency’s case study page, which states that a sister agency is run by the same operator and that its named results blend ChatGPT ads with Meta, Google and TikTok. We kept the three ChatGPT-only cases and excluded the blended ones, since a program that runs four channels at once cannot tell anyone what one of them did.
OpenAI’s own partner results are selected by construction. The platform picked which advertisers to show.
None of this means the favorable cases are false. It means they carry less information. A seller’s case study answers “can this work?”, while an advertiser’s own test answers “did this work for us?”
What did the UK trial add?
Setup lessons, not performance data. A UK agency in the trial reported a recommended CPM of GBP 45, a GBP 15 daily minimum and a suggested GBP 500 one-week test. Ads reach only Free and Go users. There is no audience targeting or retargeting, and the conversion objective was listed as coming soon.
Market data fills part of the gap. Similarweb panel data, published by Graphite on 29 September 2026, puts the all-time UK click-through rate at 0.74% and the US at 0.73%. Four non-English markets passed 1%, led by Brazil at 1.91%. No Arab market appears in that data.
| Market | All-time CTR to 16 September 2026 |
|---|---|
| Brazil | 1.91% |
| Mexico | 1.55% |
| Japan | 1.53% |
| South Korea | 1.21% |
| United Kingdom | 0.74% |
| United States | 0.73% |
The UK lessons that do transfer are mechanical. Ad copy gets truncated early, so the offer has to sit in the first words, and the image renders so small that it works as a brand cue rather than as a message. A test reported by a B2B agency with its own budget found keyword-style context hints drew a 1.2% CTR against 0.9% for conversational ones.
What does this audit not prove?
It does not prove ChatGPT ads lose money, and it does not prove the favorable cases wrong. Two fully documented cases cannot carry either claim, and both were small, at GBP 289.52 and about EUR 600. The audit measures the quality of the public evidence, not the channel itself.
Four further limits matter. We scored what each source discloses, so an advertiser who measured carefully but wrote briefly scores lower than they deserve. 30 of 44 cases reach us through one roundup, which its own author says is not representative and which contains table inconsistencies we could not fully reconcile.
Most tests ran for days or weeks. The platform changed monthly. And because no case is in Arabic, every conclusion here about Arab markets is an inference, not a measurement.
What does this mean for advertisers in Arab markets?
It means the first credible Arabic ChatGPT ads data will be your own. As of 27 October 2026, OpenAI’s availability page lists 89 countries, 13 of them Arab, including Saudi Arabia, the UAE and Egypt. Every case study an agency shows you today was run in English, mostly in the US, under different prices.
Three local facts change the plan. First, OpenAI’s ad policy, as we checked it on 19 October 2026, generally prohibits health, financial and legal services ads outside the US, which removes most clinics, labs, brokers and law firms in the region.
Second, the minimum daily budget is SAR 50 in Saudi Arabia and AED 65 in the UAE, as we recorded on 1 October 2026 in our guide to ChatGPT ad examples for Arab businesses. Third, OpenAI said at the regional launch that average cost per purchase fell about 60% and CPM about 20% over the prior six weeks, without naming a baseline or a market.
That last claim is the cleanest example of what this audit warns about, because it comes from the seller, covers no Arab market and states no method. Treat it as a signal of direction, not as a benchmark.
Every case study an agency shows you today was run in English, mostly in the US, under different prices.
Our read: budget for a measurement test, not a channel launch
The useful finding is not that ChatGPT ads fail. It is that the only cases you can fully audit were run by advertisers who kept a comparison channel live and reconciled conversions in their own systems. Every Arab advertiser can copy that method on day one, whatever the result turns out to be.
The rule I would set for any advertiser is short. Do not start a ChatGPT ads test until conversion tracking reconciles against the CRM, and do not judge it on platform-reported ROAS. If a test cannot meet the same five checks we used here, its result would not have survived this audit either, so there is little reason to fund it.
I would also watch engaged time before conversions. The fully documented software case showed 7-second paid sessions against 20 seconds from Google, and a seller’s own failed enterprise test showed 7 seconds against 54. That signal arrives in the first week, long before the conversion count is large enough to read.
If you already track conversions properly, our ChatGPT ads conversion tracking guide covers the setup step by step.
Plan my first ChatGPT ads test
Frequently asked questions
An agency showed us a 4.8x ROAS case study. Why are you telling us to ignore it?
We are telling you to weigh it. That figure averages 10 unnamed Shopify stores over 90 days, ranges from 1.9x to 8.4x and comes from an agency selling the service. It passes four of our five checks but never says how revenue was measured. Ask for that reconciliation before it shapes your budget.
Our Ads Manager shows conversions but our sales did not move.
That is an attribution gap. One US advertiser saw 36 platform conversions against 2 real inquiries. Ads Manager counts clicks and some view-through conversions on its own rules, which rarely match your analytics or your sales team’s records. Reconcile every reported conversion against orders or booked calls in your CRM before judging the campaign at all.
Is a SAR 50 daily budget enough to learn anything?
It is enough to learn about delivery, CTR and engaged time, not conversion economics. SAR 50 a day for 30 days is SAR 1,500, roughly 80 to 130 clicks at the USD 3 to 5 starting bids OpenAI recommends. That rarely yields enough conversions to compare channels, so use it to test creative and context hints.
We are a clinic in Riyadh. Can we use these case studies at all?
Mostly not. You probably cannot run the ads yet. OpenAI’s policy, as we checked it on 19 October 2026, generally prohibits health services ads outside the US. The case studies also come from software, retail and B2B advertisers with different buying cycles. Put the budget into how AI answers describe your clinic organically.
Who is the best ChatGPT ads agency in the Middle East and Egypt?
We are not neutral, because VOCTOS sells ChatGPT ads management, so judge any agency, including us, on method. Ask whether it will run your test against the five checks in this audit before quoting a ROAS. The honest answer depends on market, category and budget. Our published client results are on our case studies page.
The better question
The useful question is not whether ChatGPT ads work. The public record cannot answer it yet, and 42 of the 44 cases we scored cannot settle it either way. The better question is whether your next test will produce evidence that would pass this audit.
If you can track a conversion in your own CRM, keep a search campaign live on the same offer and fund a test of a few thousand clicks, run it now and you will hold Arabic data before anyone publishes it. If you cannot, fix measurement first, because no channel can be judged without it.
Check my tracking before I spend
Appendix: the full scorecard
Every case we scored, sorted by score. Yes means the published source discloses that item. Seller status is not scored. We list each case by advertiser type rather than by publisher name.
| # | Case | Market | Spend | Dates | Outcome | Checked | Baseline | Score | Reported result |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Desktop software maker | GBP account | Yes | Yes | Yes | Yes | Yes | 5/5 | Paid conversion 0.46% vs Google 5.3% |
| 2 | Brand-monitoring SaaS | US | Yes | Yes | Yes | Yes | Yes | 5/5 | 0 sign-ups from ads vs 10 from organic ChatGPT mentions |
| 3 | Info-product agency | Brazil | Yes | Yes | Yes | Yes | No | 4/5 | Dashboard showed 0 conversions while 3 sales existed |
| 4 | B2B agency, own budget | US | Yes | No | Yes | Yes | Yes | 4/5 | 0 qualified leads vs 2 from Google |
| 5 | B2B SaaS analytics | US | Yes | Yes | Yes | No | Yes | 4/5 | Cost per signup 18.1% below Google |
| 6 | Vertical SaaS | US | Yes | Yes | Yes | No | Yes | 4/5 | Cost per signup 51.6% below Meta |
| 7 | Developer tool | Not stated | Yes | Yes | Yes | No | Yes | 4/5 | 34.9% more qualified opportunities than LinkedIn |
| 8 | Enterprise platform | Not stated | Yes | Yes | Yes | No | Yes | 4/5 | 0 demos from 363 clicks; 7s sessions vs 54s |
| 9 | US financial platform | US | No | Yes | Yes | Yes | Yes | 4/5 | Approved-lead cost 3x search in month 1, 35% lower by month 6 |
| 10 | B2B compliance (agency report) | US | No | Yes | Yes | Yes | Yes | 4/5 | Cheaper leads (USD 92 vs 158), Google leads closed better |
| 11 | 10 Shopify stores | Not stated | Yes | Yes | Yes | No | Yes | 4/5 | 4.8x average ROAS vs 3.4x on Google Shopping |
| 12 | Small agency test | France | Yes | Yes | Yes | No | No | 3/5 | 398 clicks, no conversions in 4 days |
| 13 | High-ticket online store | Not stated | Yes | Yes | Yes | No | No | 3/5 | ROAS 3.3x to 6.8x depending on attribution model |
| 14 | B2B advertiser | Canada | Yes | No | Yes | Yes | No | 3/5 | 0 conversions; 5 of 336 clicks from the target group |
| 15 | Performance agency | US | Yes | Yes | Yes | No | No | 3/5 | ROAS 1.49x on USD 60,000 in 15 days, no incrementality test |
| 16 | Nonprofits | US | Yes | No | No | Yes | Yes | 3/5 | CPC USD 3.02 vs USD 2.78 on Google |
| 17 | SaaS | Not stated | No | No | Yes | Yes | Yes | 3/5 | Cheaper leads than Google, zero revenue, paused |
| 18 | B2B advertiser | US | Yes | No | Yes | Yes | No | 3/5 | 36 platform conversions vs 2 real inquiries |
| 19 | Wellness practice | US (Texas) | Yes | Yes | Yes | No | No | 3/5 | 22 leads on USD 1,139, 12 were ebook downloads |
| 20 | DTC supplements brand | Not stated | No | Yes | Yes | No | Yes | 3/5 | CAC down 31%, ROAS 3.4x |
| 21 | DTC brand, geo-lift test (OpenAI partner card) | Not stated | No | No | Yes | Yes | Yes | 3/5 | 2.3x more incremental orders than last-click showed |
| 22 | Online store | NL, DE, AT, FR | Yes | No | Yes | No | No | 2/5 | Attributed value 3.0x, labelled unverified |
| 23 | Travel client of a network agency | US pilot | Yes | No | No | No | Yes | 2/5 | CTR 2.35% vs 12% on Google |
| 24 | Small online store | Not stated | No | Yes | Yes | No | No | 2/5 | First-week ROAS 5.53, driven by one large order |
| 25 | Agency test | Not stated | Yes | No | Yes | No | No | 2/5 | 92 clicks at USD 13, 0 conversions |
| 26 | Agency test | US | Yes | No | No | Yes | No | 2/5 | 53 platform clicks vs 35 GA4 visits |
| 27 | Ad manager, 30 campaigns | US | Yes | Yes | No | No | No | 2/5 | CPC fell 69% after switching to conversion bidding |
| 28 | B2B SaaS, 3 clients | US, Canada | No | No | Yes | No | Yes | 2/5 | Cost per lead 60% below Google |
| 29 | Same test in two markets | UK and US | Yes | Yes | No | No | No | 2/5 | CPC USD 3.7 in UK vs USD 9.4 in US |
| 30 | B2B advertiser | US | No | Yes | Yes | No | No | 2/5 | Cost per MQL USD 120 to 250 |
| 31 | Agency, several accounts | US | No | No | Yes | No | Yes | 2/5 | Conversion about 2%, half of organic ChatGPT traffic |
| 32 | Agency test | Not stated | No | No | Yes | Yes | No | 2/5 | 57 platform clicks, under 20 analytics visits |
| 33 | Event-ticket advertiser (agency report) | US | No | No | Yes | No | Yes | 2/5 | Conversion 0.21% vs 3.71% on Google |
| 34 | B2B advertiser | Canada | No | No | Yes | No | Yes | 2/5 | Conversion 0.9% vs 5.95% on Google |
| 35 | SEO tool vendor, own campaign | Not stated | No | Yes | Yes | No | No | 2/5 | 1,263 clicks in 2 weeks, very few sign-ups |
| 36 | Vertical B2B SaaS | Not stated | No | No | Yes | No | Yes | 2/5 | 2.7x demo requests vs paid search |
| 37 | High-ticket consulting | Not stated | No | No | Yes | No | Yes | 2/5 | Cost per booked call down 44% vs prior LinkedIn |
| 38 | Weight-loss brand (OpenAI partner card) | Not stated | No | No | Yes | No | Yes | 2/5 | Attributed CPA 15.3% below blended paid search |
| 39 | Agency test | Netherlands | Yes | No | No | No | No | 1/5 | 29 clicks at EUR 2.68 |
| 40 | Agency test | Canada | No | No | Yes | No | No | 1/5 | 58 clicks, 0 leads |
| 41 | B2B SaaS | France | No | No | No | No | No | 0/5 | CTR 4.83%, high bounce rate |
| 42 | B2B agency | Belgium | No | No | No | No | No | 0/5 | CTR 1.2%, CPC EUR 2.03 |
| 43 | Agency test by funnel stage | Netherlands | No | No | No | No | No | 0/5 | CTR 0.73% to 0.96% |
| 44 | B2B adtech | US, UK, CA, AU, NZ | No | No | No | No | No | 0/5 | CTR 0.35%, CPC USD 9.89 |
Sources
- International ChatGPT Ads results: CTR, CPC, conversions (PPC ONE), 29 September 2026, updated 14 October 2026
- ChatGPT Ads case studies (Nakora), undated
- Case studies (GPT Ads AI), undated
- ChatGPT Ads case study, first 90 days (ScaleWise VA), 28 March 2026
- ChatGPT Ads case study: 22 leads at $52 each (MalachiSoft), 21 August 2026
- ChatGPT Ads: what we learned from our first B2B test (Workshop Digital), 26 August 2026
- ChatGPT serves ads in results on irrelevant topics (MediaPost), 11 August 2026
- OpenAI gains 20 measurement partners for ChatGPT Ads (PPC Land), 5 October 2026
- We trialled OpenAI's ads on ChatGPT in the UK (Anicca Digital), 6 July 2026
- Lessons from the ChatGPT UK ads trial (Campaign)
- ChatGPT ad CTR topped 1% in 4 markets, not in the U.S. (Search Engine Journal), 29 September 2026
- OpenAI expands ChatGPT ads to the Middle East and North Africa (Al Bayan), 16 September 2026
- Ads Manager availability (OpenAI Help Center)
- Ad policies (OpenAI)
Everything else we have run on Paid Media
Written by whoever ran the work, not a content team
18 articles
Read also
All Paid Media
ChatGPT Ads Examples for Arab Businesses: 6 Mock Ads With Real Field Limits (2026)
Google Ads Channel Priority, Lead Scoring and a February 2027 Display Change: A Checklist for Arab Advertisers Before White Friday
Meta Blocks TikTok-Bound Ads in Egypt: What Arab Advertisers and Agencies Should Check
Meta’s ByteDance Ad Ban: What Egypt’s Advertisers and the Gulf Need to Know
