That finding describes a defined AEO and AI-visibility dataset. It is not a universal ranking of the internet. The questions, platforms, market, collection route, and counting rule all affect which sources appear.
Key findings
- The study planned 240 response units: 16 fixed English prompts × five platform routes × three repetitions in the US market.
- 141 response units succeeded and 99 failed or were missing. Failed units were excluded rather than counted as answers with zero Citations.
- Successful responses produced 1,050 Citation facts and 231 unique canonical Citation URLs.
- Semrush was the leading domain with 60 Citation facts across 46 responses, 14 prompts, and 12 unique URLs.
- UnderAI received zero strict brand mentions and zero official-site Citations in the 141 successful answers. That is the baseline this content and source program is intended to change.
- External publishers supplied 77.3% of Citation facts; competitor official sites supplied 14.7%; community and social sources supplied 8.0%.
Top cited domains in this dataset
| Rank | Domain or subdomain | Citation facts | Responses citing it | Prompts covered | Unique cited URLs | Main source type |
|---|---|---|---|---|---|---|
| 1 | semrush.com | 60 | 46 | 14 | 12 | Competitor official |
| 2 | tryprofound.com | 50 | 26 | 7 | 5 | Competitor official |
| 3 | reddit.com | 48 | 27 | 7 | 14 | Community or social |
| 4 | blog.hubspot.com | 25 | 19 | 6 | 6 | External publisher |
| 5 | medium.com | 24 | 3 | 1 | 1 | Community or social |
| 6 | nogood.io | 21 | 15 | 4 | 3 | External publisher |
| 7 | dageno.ai | 21 | 9 | 2 | 2 | External publisher |
| 8 | meltwater.com | 21 | 3 | 1 | 1 | External publisher |
| 9 | builtin.com | 20 | 8 | 3 | 2 | External publisher |
| 10 | roirevolution.com | 20 | 5 | 1 | 1 | External publisher |
The table keeps subdomains separate because the canonical URL’s host is the counting unit. For example, blog.hubspot.com and hubspot.com are separate rows. A future root-domain view may combine them, but it must be labeled as a different aggregation.
What the leaders tell us
Official product evidence can scale across many questions
Semrush appeared across 14 of the 16 prompts and Profound across seven. Their advantage was not a single high-performing article. Multiple official pages supplied definitions, product facts, comparisons, and category language that could support different questions.
Independent publishers remain the majority source layer
More than three quarters of all Citation facts came from external websites. A brand cannot address that gap by publishing only on its own domain. It needs clear official evidence and credible independent corroboration.
Communities can win concentrated source demand
Reddit covered seven prompts with 14 unique URLs. Medium produced 24 Citation facts from one URL and one prompt. Those patterns are different: Reddit showed distributed community coverage, while Medium showed concentrated adoption of one source. Domain totals alone would hide that distinction.
Methodology
Question set
The fixed Panel contained 16 questions covering AI visibility, AEO, GEO, AI search optimization, LLM visibility, and AI SEO. It included definition, strategy, product-selection, service-selection, and business-value tasks.
Collection design
| Field | Study setting |
|---|---|
| Market and language | United States; fixed English prompts |
| Observation date | July 23, 2026 |
| Repetitions | Three per prompt and platform route |
| Planned platforms | ChatGPT, Perplexity, Gemini, Google AI Mode, Google AI Overviews |
| Planned response units | 240 |
| Successful response units | 141 |
| Failed or missing units | 99 |
| Citation rule | Structured grounding Citation records only |
| URL rule | Remove tracking parameters and normalize to a canonical URL before aggregation |
ChatGPT completed 48 units, Perplexity 45, and Gemini 48. Google AI Mode and Google AI Overviews returned no successful units in this collection route, and three Perplexity units failed. Those 99 units are missing measurement, not zero-Citation answers.
Counting rules
A Citation fact is one structured response-to-URL relationship. The same URL can create more than one Citation fact when it appears in different responses or more than once under the provider’s structured output. The study therefore reports Citation facts alongside cited responses, covered prompts, and unique URLs.
Publishers were classified as competitor official, external website, or community/social. UnderAI official URLs were tracked separately. URL and publisher classification were reviewed from the preserved response-level evidence.
Limits
- The Panel covers the AEO and AI-visibility category, so the ranking should not be generalized to unrelated markets.
- It represents one US English observation window, not a permanent platform preference.
- Platform routes expose Citations differently. This study counts the structured Citation records actually returned by the collection route.
- Google AI Mode and Google AI Overviews are not represented in the ranking because their planned units failed.
- Citation presence does not prove endorsement, factual agreement, or traffic to the cited page.
What brands should do with this result
Start with the response, not the domain leaderboard. For each important prompt, identify the exact page cited, what evidence it supplies, whether it is an official or external source, and which page your brand needs to strengthen.
AI Citation Tracking explains the response-to-URL data model. AI Source Gap Analysis turns the observed source difference into a page or publisher action. AI Citation Rate Benchmarks compares how frequently each platform returned structured Citations.
How UnderAI supports the analysis
UnderAI GEO Workspace connects Prompt Groups, Answer Snapshots, competitors, and Citation records inside a market-specific Tracking Project. UnderAI Citation Strategy then verifies the exact source page, classifies its role, assigns an official or external action, and retests the same fixed Panel after publication.
Research disclosure
This page was prepared from UnderAI’s July 2026 baseline study, collected July 23, 2026. The underlying audit preserves every planned response unit, success or failure state, raw answer path, canonical URL, and Citation fact ID. Reproducing the result requires the same Panel, platform routes, market, date window, repetitions, and normalization rules.