The five AI engines recommend brands five different ways

The engines are sorted by how often they actively recommend a tracked brand. Click an engine to jump to its field guide.

What we observed

1. 1Names 5.9 brands per answer on average, the highest of the five. Claude prefers menu-style answers over single-brand picks.
2. 2Most stable engine under persona signals (74% mean top-3 overlap vs baseline). Persona signals reorder Claude's top 5 less than they reorder ChatGPT or Gemini.
3. 3Most resilient at the Comparison phase (69% leader-held under professional persona, the most stable cell in the entire panel). Feature-comparison rankings on Claude are particularly durable.
4. 4Different brands hold across personas on Claude than on other engines. Carvana on used auto, Lowe's on home retailers. Neither matches the ChatGPT or Gemini leader.

Content patterns we observed in cited brands. Long-form, structured comparison content with explicit feature lists. Claude's broad-recommender bias rewards content that frames brands as part of a category set, not as standalone winners.

What we observed

1. 1Cites the brand in 25% of buyer-intent answers, 2nd highest of the five engines. Only 10% of its visibility is mention-without-recommendation, the lowest of the five. ChatGPT recommends rather than hedges.
2. 2Most stable engine for commerce queries when buyer signals are absent. Baseline rankings are durable.
3. 3The kingmaker for default-choice brand discovery. If a brand holds #1 on ChatGPT, it usually holds across personas too. Chime in fintech is the only segment-leader that survived all 10 buyer personas in our panel.
4. 4ChatGPT does NOT change its mind on premium-coded categories (luxury goods, hotels, makeup) without a strong premium signal. Generic shopping framing produces the 'safe default' answer, not the premium answer.

Content patterns we observed in cited brands. Strong entity authority and consistent third-party citations. The brands ChatGPT cites tend to have well-structured entity data on their primary domains and dense external mentions on aggregator sites.

What we observed

1. 1Names 4.5 brands per answer on average, the lowest of the five. Grok prefers naming a few brands and committing.
2. 2Cites the brand in 18% of answers, 3rd highest of the five, and the brands Grok does cite tend to be category-defining.
3. 3Surprisingly persona-stable for commerce queries (70% mean top-3 overlap, on par with ChatGPT). Grok's persona behavior is more like ChatGPT than its small answer set would suggest.
4. 4Locks in on home retailer (Home Depot) and used auto retailer (CarMax) baseline leaders across all 10 buyer personas in our panel, the most-stable engine for those two specific segments.

Content patterns we observed in cited brands. Brand authority signals that read as 'category default'. Grok's narrow-scope picks tend to be the brands consumers would name without thinking. Short answer set means top-of-mind matters most.

What we observed

1. 1Cites the brand in 15% of answers, 4th highest of the five, with 4.9 brands per answer on average. Gemini is one of the most selective recommenders on the panel.
2. 2Most personalization-sensitive engine (66% mean top-3 overlap, 8 percentage points below Claude). The same buyer signal moves Gemini rankings the most.
3. 3Different baseline leaders than the other engines in 6 of 25 segments. Gemini's defaults are not the panel's defaults.
4. 4Premium signal has its biggest effect on Gemini at Discovery (35% leader-held, the lowest cell in our panel). If your buyer is premium-coded, your Gemini Discovery rankings are wide open.

Content patterns we observed in cited brands. High Google-aligned signals: Knowledge Graph entries, structured data on the primary domain, and presence in Google's AI Overviews retrieval set. Gemini's training data weights Google's surface signals heavily.

What we observed

1. 122% of Perplexity's visibility is mentions without a recommendation, the highest of the five engines. Perplexity hedges: it names brands without explicitly recommending them.
2. 2Web-grounded retrieval matters here. Brands with strong recent press, partner integrations, or sustained social signal punch above their weight.
3. 3Surprisingly stable under persona signals (73% top-3 overlap), comparable to Claude. The intuition that web grounding amplifies personalization was wrong in our data.
4. 4Has the cleanest baseline-leader story in some categories (hi-retailers all 5 personas held Lowe's). Where Perplexity locks in, it locks in hard.

Content patterns we observed in cited brands. Recent, authoritative third-party citations. A category review article published in the last 90 days moves Perplexity citations more than the same content published 18 months ago.

The most volatile (persona × phase) cell in our panel was premium × evaluation at 32% leader-held. When a premium-coded buyer asks AI which brand to actually buy, the rankings reshuffle hardest. The most stable cell was professional × comparison at 68%: feature-comparison rankings hold under professional buyer signals. The full persona × phase heatmap is in Finding 10.

Three ways to act on this guide

1. Find your category. See which engines cite which brands in your space on the research panel.

2. Test your buyers. Open the Persona Explorerto see how your brand's citation rate moves under each buyer persona.

3. Check your brand. Run a free AI visibility check to find out which engine cites you most and which is your biggest blind spot.