The two ingredients, briefly
Every answer about your brand comes from training data (the frozen snapshot of the web the model memorized) or live retrieval (RAG — the engine searches, reads current pages, and grounds the answer in them). Memory moves in quarters. Retrieval moves in days. Knowing which ingredient produced an answer tells you how fast you can change it.
Engine by engine
Perplexity: the citation machine
Perplexity retrieves for nearly every answer and shows its sources. That makes it the most transparent engine to optimize for — the cited pages are the to-do list. If a comparison post you're absent from gets cited for your category question, you know precisely which door to knock on. It's also the fastest engine to reflect new content, often within days of a page being crawled.
ChatGPT: the mood-dependent researcher
ChatGPT answers from memory when it feels confident and searches when the question is current, commercial, or specific. In practice, category recommendation questions increasingly trigger search. The catch: it synthesizes sources without always showing which one carried the answer, so attribute movement by re-asking the same question over time rather than by reading footnotes.
Gemini: the index native
Gemini sits closest to a search index and behaves like it — grounded answers with heavy weight on pages that already rank. If your Google SEO is strong, Gemini is usually your friendliest engine; if it isn't, Gemini will faithfully reproduce that weakness. The most SEO-correlated of the four.
Claude: the careful one
Claude is the most conservative about recommendations — more hedging, more "it depends on your needs," fewer confident brand lists. When it does name brands, memory plays a large role, which means entity consistency and a broad, corroborated footprint matter disproportionately. Slow to change, and steady once changed.
What this means for effort
- Chasing a fast win? Start with Perplexity: fix or win the pages it cites, and measure again in two weeks.
- Strong SEO already? Gemini is likely your best engine — check whether the others lag it, because that gap is your memory problem in miniature.
- Absent from Claude and uncited ChatGPT answers? That's footprint work: sustained coverage, consistent facts, entity SEO. Quarters, not sprints.
