The platform difference was large: Perplexity returned structured Citations in every successful response, Gemini in 68.8%, and ChatGPT in 31.2%. These are route- and dataset-specific observations, not permanent product scores.
Citation rate definition
Citation Rate = successful evaluable responses with at least one structured Citation
/ successful evaluable responses on that platform or question setFailed, missing, or non-evaluable collection units are excluded from both numerator and denominator. When a successful route returned a valid answer with no structured Citation, that unit remained in the denominator as a no-Citation observation.
Platform benchmarks
| Platform route | Planned units | Successful units | Successful units with Citation | Citation rate | Citation facts | Unique canonical URLs |
|---|---|---|---|---|---|---|
| Perplexity | 48 | 45 | 45 | 100.0% | 450 | 141 |
| Gemini | 48 | 48 | 33 | 68.8% | 447 | 63 |
| ChatGPT | 48 | 48 | 15 | 31.2% | 153 | 46 |
| Google AI Mode | 48 | 0 | — | Not available | 0 | 0 |
| Google AI Overviews | 48 | 0 | — | Not available | 0 | 0 |
| Measured total | 144 | 141 | 93 | 66.0% | 1,050 | 231 across the combined dataset |
The measured-total row excludes the 96 planned Google units because the collection tasks failed. Reporting them as 0% would falsely turn missing measurement into platform behavior.
Citation rates by question category
| Concept entity | Successful responses | Responses with Citation | Citation rate | Citation facts |
|---|---|---|---|---|
| LLM Visibility & Optimization | 9 | 9 | 100.0% | 123 |
| AI Search Optimization | 26 | 20 | 76.9% | 227 |
| AEO | 44 | 32 | 72.7% | 335 |
| AI Visibility | 27 | 18 | 66.7% | 240 |
| AI SEO | 18 | 9 | 50.0% | 75 |
| GEO | 17 | 5 | 29.4% | 50 |
The category rates should not be read as a universal hierarchy. The Panel contains different numbers and types of questions in each category, and a single observation window can be sensitive to wording and platform behavior.
What changed by question task
Product- and service-selection questions were the most consistently cited in this sample. Three Panel questions reached a 100% observed Citation rate across their successful units:
- “What are the best Answer Engine Optimization solutions for AI platforms?”
- “Which Answer Engine Optimization service providers are top-rated?”
- “Which AI search optimization startups have the strongest visibility metrics?”
Definition and broad educational questions were less consistent. “What is Generative Engine Optimization?” returned a Citation in 25.0% of successful units, while “What is Answer Engine Optimization?” returned one in 66.7%.
This difference matters for content planning. Commercial-comparison pages compete in a source-heavy answer environment. Definition pages still need trustworthy evidence, but a missing Citation in one answer should not automatically be treated as a page failure.
Why platform rates differ
Platforms use different answer products, search routes, grounding interfaces, and source-display policies. A Citation rate therefore reflects both the question and the observed route. It is not a direct measure of answer quality.
For example, a response may be useful without exposing a structured URL, while a heavily cited response may still use irrelevant or weak sources. Citation analysis must inspect the exact page and its role, not stop at the count.
Methodology
The study used a fixed Panel of 16 US English questions and three repetitions on each planned platform route on July 23, 2026. The 240 planned units produced 141 successes and 99 failures or missing results. The successful responses produced 1,050 structured Citation facts and 231 unique canonical URLs.
The UnderAI baseline preserved:
- exact Prompt ID, text, business task, and Panel layer;
- platform, run, timestamp, status, and returned model slug when available;
- full response text and response hash;
- structured Citation URL, title, source name, and response relationship;
- raw and canonical URLs;
- publisher classification;
- strict UnderAI mention and official-site Citation state.
The baseline found zero strict UnderAI mentions and zero UnderAI official-site Citations. That result is reported separately from Citation availability: a platform can cite sources while still omitting UnderAI.
How to use the benchmark
Use these rates to design measurement, not to set a universal target. A useful brand baseline should separate:
- response success rate;
- answers with any structured Citation;
- answers citing the brand’s official site;
- answers citing credible external evidence about the brand;
- answers that mention or recommend the brand accurately.
The Most Cited Domains study shows which sources filled the measured answers. AI Citation Tracking explains the response-level data model, and AI Search Visibility Metrics keeps Citation rates separate from brand mention and business outcomes.
Benchmark your own question set
UnderAI GEO Workspace preserves the Prompt, Answer Snapshot, competitor state, and Citation evidence behind each aggregate. A brand-specific benchmark should use the questions that influence its own category, comparison, implementation, and purchase decisions rather than adopting this research Panel unchanged.
Research disclosure
This report uses UnderAI’s July 2026 baseline study, collected July 23, 2026. The rates can be reproduced only with the same Prompt Panel, platform routes, market, date window, repetitions, status rules, and structured Citation extraction method.