BrandRadar’s new UAE study ran the same 50 commercial prompts through ChatGPT and Google AI Mode and found that citation sources could change even when the questions did not. The test covered 25 Toyota UAE automotive prompts and 25 Sukoon Insurance motor-insurance prompts through BrandRadar V2 in August 2026.
The clearest split came from Toyota. Toyota’s own domain supplied 9.6% of ChatGPT citations but 1.8% of Google AI Mode citations for the same 25 prompts. In the other direction, YouTube produced 23 citations in Google AI Mode and one in ChatGPT. Sukoon’s owned-domain share was almost unchanged at 4.2% versus 4.4%.
That contrast is the useful part for B2B teams. When IVRIS reported earlier on cross-engine citation divergence, the evidence was a broad multi-study index. BrandRadar’s smaller experiment holds the prompt set and market more tightly. It does not prove a universal platform preference, but it tests whether one source strategy can be assumed to transfer between engines.
Direct answer — do ChatGPT and Google AI Mode rely on the same sources for the same commercial questions?
In BrandRadar’s UAE automotive and motor-insurance experiment, not consistently. Toyota showed a large source split: toyota.ae represented 9.6% of ChatGPT citations versus 1.8% in Google AI Mode, while YouTube appeared once versus 23 times. Sukoon did not reproduce the owned-domain gap. The result supports engine-specific measurement in this sample, not a universal rule about either platform.
Key Takeaways
- BrandRadar tested 50 UAE prompts: 25 for Toyota UAE and 25 for Sukoon Insurance, with each prompt run on both engines in August 2026.
- For Toyota, toyota.ae’s citation share was 7.8 percentage points higher on ChatGPT, or about 5.3× the Google AI Mode share.
- YouTube’s Toyota citation share was about 0.36% on ChatGPT and 6.78% on Google AI Mode, a 6.43-point gap.
- Sukoon’s owned-domain share barely changed: 4.2% on ChatGPT versus 4.4% on Google AI Mode.
- Exact model versions, rerun count and duplicate-URL counting rules are not disclosed.
What BrandRadar Actually Tested
The 50 prompts were split evenly between Toyota UAE and Sukoon Insurance. Each brand’s set covered five clusters: best-of/category, direct comparisons, problem or need-based, local or location-specific, and long-tail category questions. BrandRadar logged 281 ChatGPT and 339 Google AI Mode citations for Toyota, plus 337 and 475 for Sukoon.
The paper calls this a two-category, one-market snapshot. Testing occurred in August 2026, and the Toyota comparison is described as same-day, but no calendar date is given. BrandRadar also does not disclose sign-in status, exact model versions, whether prompts were repeated, duplicate-URL treatment or a formal source taxonomy. Those gaps rule out probability or stability claims.
The Source Differences Are Large in Toyota’s Sample
Toyota supplies the strongest source-divergence evidence because BrandRadar publishes engine totals, owned-domain shares and a raw YouTube count. The table keeps citation share separate from response presence.
| Observed metric | ChatGPT | Google AI Mode | IVRIS calculation | Scope |
|---|---|---|---|---|
| Toyota-owned domain citation share | 9.6% | 1.8% | ChatGPT +7.8 pp; 5.3× as large | 25 Toyota prompts; 281 vs 339 citations |
| Toyota YouTube citation share | 1/281 = 0.36% | 23/339 = 6.78% | AI Mode +6.43 pp; 19.1× as large | Normalized from published counts |
| Sukoon-owned domain citation share | 4.2% | 4.4% | AI Mode +0.2 pp | 25 insurance prompts; 337 vs 475 citations |
| Policybazaar response presence | 8/25 = 32% | 18/25 = 72% | AI Mode +40 pp | Response coverage, not citation share |
| InsuranceMarket.ae response presence | 9/25 = 36% | 17/25 = 68% | AI Mode +32 pp | Response coverage, not citation share |
IVRIS calculation based on BrandRadar’s published sample. For Toyota’s owned domain, 9.6% minus 1.8% is 7.8 points; 9.6 divided by 1.8 is 5.33. So the observed owned-domain citation share was about 5.3× as large in ChatGPT’s Toyota sample. That is not the same as being 5.3× more likely to cite Toyota.
BrandRadar’s reported Toyota top-four lists also barely overlap. ChatGPT names toyota.ae, uae.autotraders.ae, consumerreports.org and reddit.com; AI Mode names reddit.com, youtube.com, yallamotor.com and arabwheels.ae. Reddit is the only shared domain, giving 14.3% Jaccard overlap across the seven-domain union. That applies only to the published top-four lists, not every citation.
Insurance Shows Why the Toyota Result Cannot Become a Rule
Sukoon is the built-in counterexample. Its own-domain citation share moved only 0.2 points, from 4.2% on ChatGPT to 4.4% on AI Mode. Larger differences appeared in comparison-site response coverage: Policybazaar was present in 8 of 25 ChatGPT responses versus 18 of 25 AI Mode responses; InsuranceMarket.ae appeared in 9 versus 17.
Those counts are response presence, not citation share. The distinction matters because brand mentions and source citations are different measurements. Toyota was mentioned in 84% of ChatGPT responses and 72% of AI Mode responses even though AI Mode logged more citations. Different source mixes appeared here, but their direction and size changed by vertical and metric.
The B2B GEO Implication: Measure Source Mix Per Engine
A practical GEO program should treat source composition as an engine-level diagnostic. Keep prompt wording, market and schedule fixed, then record citations, mentions and domains separately. IVRIS’s fixed-prompt measurement framework uses the same principle: changing the prompt universe while comparing results changes the denominator.
Our read: BrandRadar’s study is more useful as a measurement warning than an optimization recipe. Measure where each engine sources answers before reallocating content or distribution effort. Citation share describes recorded citations; it does not by itself measure ranking, recommendation probability, clicks, traffic, conversion or revenue.
Frequently Asked Questions
No. BrandRadar reports 50 prompts in total: 25 for Toyota UAE automotive questions and 25 for Sukoon Insurance motor-insurance questions. Every prompt was run on both ChatGPT and Google AI Mode through BrandRadar V2 in August 2026. The paper does not disclose the exact number assigned to each of its five prompt categories.
In BrandRadar’s 25-prompt Toyota UAE sample, yes. toyota.ae accounted for 9.6% of ChatGPT citations and 1.8% of Google AI Mode citations. The observed gap was 7.8 percentage points, and the ChatGPT share was about 5.3× as large. That result should not be generalized to other brands, categories or markets.
Yes in this sample. BrandRadar counted 23 YouTube citations across 339 Google AI Mode citations for Toyota, versus one across 281 ChatGPT citations. IVRIS calculates those shares at about 6.78% and 0.36%, a 6.43-point gap. This is an observed citation-share difference, not a probability of future citation.
BrandRadar’s UAE test suggests teams should not assume clean transfer. Toyota’s owned-domain and YouTube patterns differed sharply by engine, while Sukoon’s owned-domain share did not. The safer approach is to hold prompts constant and measure source composition per engine before deciding which owned, third-party or video surfaces deserve more investment.
