Visibility by industry
Explicit recommendation rate and average score.
IndustryRateScore
Energy storage17%48.86
Industrial robotics14%47.08
Life sciences services7%46.29
Cross-border logistics5%42.80
Nutraceuticals3%43.79
Published August 14, 2026 · Version 1.1 · Reviewed September 6, 2026
2026 B2B Brand AI Search Visibility Benchmark
Synta tested 50 brands across five China-export industries in 750 standardized, web-grounded answers from Perplexity, Qwen and DeepSeek. The result: being retrievable does not guarantee being recommended.
Author
Synta Research
Method review
Synta internal methodology review
Test window
August 14, 2026
Current version
1.1 · September 6, 2026
The review is an internal methods check, not independent peer review. Version 1.1 adds visible research governance, public example test cells, and a clearer separation between this API benchmark and Bing/Copilot observations. View change log.
60%
never recommended
66.7%
retrieved but not recommended
13.67
point China-origin gap
3,032
parsed citations
Executive finding
Retrieval presence is not shortlist inclusion
Thirty of the 50 brands received no explicit recommendation in any unbranded opportunity. Their websites may exist, rank or appear in retrieval, but the models did not have enough comparative evidence to place them on the buyer shortlist.
Among 33 brands whose official domains appeared in the study's shared retrieval evidence packs, 22 still received zero explicit recommendations. This 66.7% gap is a reproducible retrieval-visibility proxy, not a claim about Google or Bing organic rankings.
China-origin brands averaged 41.67 out of 100, compared with 55.33 for international benchmarks. The underlying unrounded gap was 13.67 points.
Visibility by model family
The platform gap shows why one-engine monitoring is incomplete.
Model familyRateScore
Perplexity11%46.29
DeepSeek10%46.86
Qwen6%44.14
Every answer used the same evidence pack for its prompt cell, helping separate model behavior from source-set variation.
Public test examples
What one standardized test cell looked like
These examples disclose the question structure and counting rule without presenting a controlled API answer as a consumer-product result. Brackets represent the capability or buying constraint defined for that industry.
Life sciences services
“Which providers should a US biotechnology buyer evaluate for [capability]?”
10 fresh unbranded runs per model family. Public outputs are aggregate recommendation rates and source counts; complete provider responses remain internal for reproducibility and provider-policy compliance.
Nutraceutical ingredients
“Which suppliers fit a US formulator seeking [ingredient or application]?”
10 fresh unbranded runs per model family. Public outputs are aggregate recommendation rates and source counts; complete provider responses remain internal for reproducibility and provider-policy compliance.
Bing/Copilot is measured separately
This benchmark does not measure Bing or Copilot. Synta's separate measurement guide uses Bing Webmaster Tools citation pages and grounding queries, and does not merge those observations into the 750-answer API dataset.
Read the Bing/Copilot measurement guideComplete brand attachment
Top 10 of 50 benchmarked brands
| Rank | Brand | Industry | Score | Recommendation rate |
|---|---|---|---|---|
| 1 | Tesla Energy | Energy storage | 72.77 | 63% |
| 2 | Fluence | Energy storage | 69.79 | 57% |
| 3 | FANUC | Industrial robotics | 67.74 | 57% |
| 4 | Universal Robots | Industrial robotics | 66.98 | 45% |
| 5 | Lonza | Life sciences services | 64.26 | 38% |
| 6 | ShipBob | Cross-border logistics | 61.31 | 27% |
| 7 | ABB Robotics | Industrial robotics | 59.25 | 35% |
| 8 | SIRIO Pharma | Nutraceuticals | 58.33 | 17% |
| 9 | Sungrow | Energy storage | 56.79 | 17% |
| 10 | BYD Energy Storage | Energy storage | 55.31 | 20% |
The public attachment includes all 50 brands, model-level scores, recommendation rates and citation totals. Full provider answers are retained for reproducibility and are not published. Rankings are research signals, not procurement endorsements.
Version history
Version 1.1 · September 6, 2026
Added visible authorship and review status, public example prompt cells, and explicit separation from Bing/Copilot reporting. No benchmark scores or rankings changed.
Version 1.0 · August 14, 2026
Initial publication of the completed report, methodology, 50-brand diagnostic appendix and CSV data attachment.
Open research package
Download the report and auditable data
Use the executive report for decision-makers, the diagnostic appendix for brand-level review, and the data attachment for further analysis.
English executive report
A five-page summary for global marketing and commercial leaders.
PDF · 5 pages50-brand diagnostics
One-page score, evidence and action summary for every benchmarked brand.
PDF · 52 pages50-brand data attachment
Complete rankings, platform scores, recommendation rates and citation totals.
CSV · all 50 brandsFrequently asked questions
What does AI search visibility measure?
It measures whether a brand is discovered, cited, compared and explicitly recommended in AI-assisted buyer research. It is distinct from traditional search ranking.
Does the benchmark evaluate product quality?
No. The study measures machine-visible evidence and recommendation behavior. It is not a product certification, procurement endorsement or market-share ranking.
Are these results the same as consumer chat applications?
No. The results come from standardized model API tests using a controlled protocol. Consumer interfaces may differ because of personalization, location, model routing and product features.
What is included in the data attachment?
The public attachment includes the complete 50-brand ranking, model-level scores, recommendation rates and citation totals. Full provider answers are retained for internal reproducibility and are not published.
Is your brand retrievable but still not recommended?
Request a manual snapshot covering your company, two competitors and five proposed buyer prompts on one agreed testing surface. Feasibility and scope are confirmed before testing.