AI Recommendation Benchmark
performance management software

Which brands do AI assistants recommend when buyers ask, and which sources feed those answers. 50 buyer-intent prompts × 8 independent runs per prompt · 400 observations · generated 2026-08-09 · methodology v0.1.

Pilot data. This run uses a simulated engine to demonstrate the instrument and report format. Production benchmarks sample all major AI engines live; numbers below illustrate the methodology, not real market shares.

Share of Recommendation

Share of answers in which the brand appears in the recommended set. Whiskers are 95% Wilson intervals. Overlapping whiskers mean the ranking between those brands is not statistically settled.

Lattice
73%
15Five
72%
Leapsome
59%
Culture Amp
55%
Betterworks
44%
PerformYard
34%
WorkTango
25%
Trakstar
18%
Workleap
11%
Quantum Workplace
10%
View as table
BrandShare95% CIRuns
Lattice73.0%68.4% - 77.1%292/400
15Five72.0%67.4% - 76.2%288/400
Leapsome59.0%54.1% - 63.7%236/400
Culture Amp55.2%50.4% - 60.0%221/400
Betterworks44.0%39.2% - 48.9%176/400
PerformYard33.8%29.3% - 38.5%135/400
WorkTango24.8%20.8% - 29.2%99/400
Trakstar18.0%14.5% - 22.1%72/400
Workleap11.2%8.5% - 14.7%45/400
Quantum Workplace10.5%7.9% - 13.9%42/400

Citation share: where the answers come from

Share of answers citing each domain. A small set of pages does most of the work (concentration Gini 0.30). This is the supply-side map of the category.

g2.com
60%
capterra.com
54%
selectsoftwarereviews.com
38%
peoplemanagingpeople.com
31%
reddit.com
27%
forbes.com
24%
pcmag.com
22%
business.com
13%
techradar.com
12%
zapier.com
11%
View as table
DomainShare95% CIRuns
g2.com59.5%54.6% - 64.2%238/400
capterra.com53.5%48.6% - 58.3%214/400
selectsoftwarereviews.com38.2%33.6% - 43.1%153/400
peoplemanagingpeople.com31.0%26.7% - 35.7%124/400
reddit.com27.3%23.1% - 31.8%109/400
forbes.com23.5%19.6% - 27.9%94/400
pcmag.com22.5%18.7% - 26.8%90/400
business.com13.0%10.1% - 16.7%52/400
techradar.com12.2%9.4% - 15.8%49/400
zapier.com11.2%8.5% - 14.7%45/400

How to read this

Every number is a proportion over repeated cold-session runs, with its interval. A single snapshot supports ranking, not trend claims. Movement claims require consecutive windows and the market-shift control described in the methodology. Panels are versioned and frozen before running; extraction code is published.