Research Note 001 · Full report
What AI Actually Recommends for Lifecycle Marketing
A 432-response pilot across 24 buyer scenarios and three leading AI models.
432
total responses
417
accepted
24
scenarios
3 × 2 × 3
models · modes · reps
Executive read
The recommendation funnel has two layers.
Braze owns unaided recall. It received 91 of 206 accepted open-mode recommendations, a 44.2% share. Salesforce Marketing Cloud was a distant second at 19.4%.
Adobe wins after exposure. When neutral product profiles were supplied in eligible scenarios, Adobe Journey Optimizer won 60 of 114 opportunities, or 52.6%.
This is not a contradiction. Unaided prompts measure which brands are available in a model’s learned consideration set. Controlled prompts test preference among a balanced, scenario-specific shortlist. For marketing leaders and vendors, both signals matter.
Spontaneous default
44.2%
Braze, open mode
More than twice the unaided recommendation share of the next platform.
Informed preference
52.6%
Adobe, controlled mode
A 46.3-point lift over its unaided recall share when scenario-appropriate profiles appear.
Results
Recall and preference produce different market pictures.
Unaided recall
Share of 206 accepted open-mode recommendations. All platforms share the same denominator.
Equal-exposure preference
Selection rate when eligible and exposed. Denominators vary by scenario fit.
Why the denominators differ
Controlled candidate sets were constructed for each buyer situation. Adobe, for example, appeared in 114 accepted opportunities; Braze appeared in 185. Comparing raw controlled win counts would therefore be misleading, so the report uses selection rate when exposed.
Buyer context
The brief changes the winner.
The most useful result is not one universal leaderboard. It is the pattern of fit that appears when company scale, channels, data environment, technical capacity, and governance needs change.
Ecommerce growth
Klaviyo and Omnisend become most competitive in lean, commerce-native programs.
Klaviyo won all 9 eligible runs for a scaling Shopify Plus brand; Omnisend won 8 of 9 for a seasonal specialty retailer.
Mobile-first products
Braze is the clearest default for high-volume, behavior-triggered mobile engagement.
Braze won all 9 eligible runs in both the consumer subscription app and global mobile gaming scenarios.
Flexible midmarket stacks
Customer.io gains ground when data flexibility and implementation control matter.
Customer.io won every accepted eligible run for online education and 8 of 9 for a consumer-services scale-up.
Regulated and enterprise
Adobe strengthens sharply when governance, identity, and orchestration enter the brief.
Adobe won every accepted eligible run for regulated consumer finance and healthcare member engagement.
Model behavior
The labs do not see the market the same way.
Claude Sonnet 5
Recalled Salesforce in 31.3% of open responses, but Adobe in 0 of 67.
After exposure, Claude selected Adobe in 19 of 36 eligible runs.
134 / 144
accepted
GPT-5.6 Terra
Produced the strongest controlled preference for Adobe: 26 of 39 eligible runs.
Its unaided Salesforce recall was only 5 of 67 open responses.
139 / 144
accepted
Gemini 3.1 Pro Preview
Was the only model with a 100% contract-valid response rate in this pilot.
Its open recommendations were less concentrated on Braze than Claude or GPT.
144 / 144
accepted
Implications
What marketing leaders and vendors should take from it.
For marketing leaders
Treat AI recommendations as a starting set, not due diligence.
- Ask the model to state the buyer situation and constraints before naming a platform.
- Provide balanced candidate evidence when the decision matters.
- Compare recommendations across more than one model.
For vendors
Measure both consideration and conversion after exposure.
- Unaided recall reveals whether the brand enters AI’s default consideration set.
- Controlled preference reveals whether the product story converts once it is seen.
- Segment-level evidence is more actionable than one pooled visibility score.
Limits
What this pilot does not prove.
- It is not a product-quality ranking. The study measures model recommendation behavior, not implementation success or business outcomes.
- Three repetitions are exploratory. Broad patterns are informative; narrow differences and exact ordering remain unstable.
- Training familiarity may influence recall. Open recommendations can reflect how often a brand appears in learned material, not only fit.
- Controlled profiles simplify reality. They were balanced and scenario-specific, but procurement, pricing, services, and implementation detail were outside scope.
- The model panel is intentionally narrow. It represents three major AI labs, not every available model.