The Odyssiant Product AI Visibility Benchmark
A monthly benchmark tracking how products appear across AI-generated answers, citations and buyer journey stages.
AI-led discovery is changing how buyers research products, services and suppliers. The Odyssiant Product AI Visibility Benchmark tracks whether brands and products are mentioned, recommended, ignored or supported by credible sources across AI-generated answers.
Benchmark snapshot
In the June benchmark, product-level AI visibility weakened even as the dataset grew. The tracker analysed 4,623 product answers across four AI engines, but 57% of answers still did not mention the product being tested. The average visibility score fell to 26/100, down from 31/100 in the previous period. The commercial risk remains clear: AI systems are often able to answer buyer questions without including the product at all.
Methodology note
The June 2026 benchmark is based on 4,623 product-level AI answers from the Odyssiant AI Search Tracker dataset. The dataset covered four engines: OpenAI GPT-4o, Anthropic Claude 4 Sonnet, Google Gemini 2 Flash and Perplexity Sonar. Product audits accounted for 100% of the analysed answers. For citation league tables and source analysis, audited company domains were excluded so the benchmark could show the wider evidence environment rather than simply surfacing the companies being tested.
Verdict distribution
| Verdict | June count | June share |
|---|---|---|
| Not Mentioned | 2,656 | 57% |
| One of Several | 706 | 15% |
| Primary Recommendation | 693 | 15% |
| No Answer | 560 | 12% |
| Mentioned Only | 8 | 0% |
The visibility challenge remains weighted towards absence. In 57% of product-level answers, the audited product was not mentioned at all. Where products did appear, they were almost equally likely to be included as one option among several or positioned as the primary recommendation. The high No Answer share also shows why visibility needs to be tracked by engine as well as in aggregate.
What the benchmark shows
1. Most product-level AI answers still do not mention the product
In June 2026, 2,656 out of 4,623 product-level answers were classified as Not Mentioned. That means 57% of answers did not include the product being tested. The issue is not negative sentiment. It is absence.
2. Product visibility weakened month on month
The average visibility score fell from 31/100 to 26/100. The dataset grew, but increased measurement did not translate into stronger product presence.
3. Engine behaviour is fragmented
OpenAI GPT-4o averaged 32, Perplexity Sonar averaged 29, Anthropic Claude 4 Sonnet averaged 22 and Google Gemini 2 Flash averaged 0 in this dataset. A product’s AI visibility cannot be understood from one engine alone.
4. Early-stage visibility still matters
Frame & Clarify and Explore Landscape prompts accounted for 57% of the benchmark. This means many visibility gaps happen before the buyer reaches a formal comparison or decision stage.
5. Citation coverage remains high, but evidence is uneven
80% of answers included citations. However, citation presence does not guarantee product visibility. AI systems may produce sourced, useful answers that still omit the product being tested.
Where AI answers get their evidence
Across the June tracker, AI-generated answers drew on a broad evidence layer, with 24,870 citations extracted and 80% of answers including citations. For the benchmark source analysis, audited company domains were excluded so the league tables show the wider evidence environment rather than the companies being tested. Vendor and owned-source material still made up the largest share, but documentation, regulatory, social, analyst, encyclopaedic, publisher and community sources were all highly visible.
The implication for marketers is clear: AI visibility cannot be improved through website content alone. Brands need a wider evidence strategy across content, PR, proof, listings, reviews, communities and credible third-party sources.
Recommended actions
Audit product visibility across multiple engines, not just one AI system.
Measure absence as a commercial risk, not just sentiment or ranking.
Build content for early-stage buyer questions, especially problem framing and category exploration.
Strengthen proof assets: case studies, quantified outcomes, compliance evidence, security documentation and third-party validation.
Map the wider citation environment after excluding owned and audited company domains.