Scrubsy AI Visibility Case Study: What Changed Across ChatGPT, Gemini and Perplexity

Transparency & limitations

This is a real-world measurement case study. It reports observed before-and-after results, but it does not claim that Digital Platform 271 caused the changes. Implementation was not verified complete, and model updates, retrieval sources, marketplace data, media coverage and other external authority signals may also affect AI visibility.

Measurement design

• 20 identical frozen prompts tested across ChatGPT, Gemini and Perplexity = 60 total runs.
• 14 unbranded prompts per engine provide the cleanest before-and-after denominator: 42 unbranded runs across three engines.
• Six branded/comparison prompts per engine were also reviewed for retrieval, entity/category understanding and recommendation behavior.

Unbranded visibility: baseline vs retest

• ChatGPT recommendations: 0/14 (0%) → 3/14 (21.4%)
• Gemini recommendations: 3/14 (21.4%) → 3/14 (21.4%)
• Perplexity recommendations: 4/14 (28.6%) → 1/14 (7.1%)
• All-engine recommendations: 7/42 (16.7%) → 7/42 (16.7%)
• All-engine mentions: 8/42 (19.0%) → 7/42 (16.7%)

What changed

• Aggregate unbranded recommendation visibility remained unchanged at 16.7%.
• ChatGPT improved from 0/14 to 3/14 unbranded recommendations.
• Gemini held at 3/14 unbranded recommendations.
• Perplexity declined from 4/14 to 1/14 unbranded recommendations.
• When Scrubsy was named, the engines generally understood it as a foam-led specialist for kitchen grease, stubborn cleaning and bathroom/hard-water use cases.
• Broad category visibility, family-safe/child-pet-safe contexts, lower-harshness positioning and consistent foam ownership remained weaker areas.

Branded and comparison understanding

Across 18 branded/comparison runs, current entity/category resolution was correct in 17/18. Perplexity misresolved Scrubsy in P17 as an on-demand cleaning service. P17 asks for alternatives to Scrubsy, so recommendation is not scored for that prompt.

Implementation status

The implementation period was not verified as fully complete. Developer QA was incomplete and some items remained client-dependent, awaiting client action or pending audit. This should therefore be read as an observed retest after the implementation period, not as a controlled full-implementation causal study.

What this case study shows

The most important result is not a single growth percentage. Different engines moved in different directions, while the aggregate unbranded recommendation rate stayed flat. That makes engine-level reporting essential.