Google did not ship one AI answer to the world; it shipped three. AI Overviews summarizes above the results, AI Mode replaces the results page with a conversation, and Gemini is the assistant in its own app. They share plumbing and reward different things.
Here is each game's rules, how sources surface in each, and the single strategy that plays all three.
| AI Overviews | AI Mode | Gemini | |
|---|---|---|---|
| Lives | Above classic results | Its own search tab | Standalone assistant |
| Appears | On queries Google picks | When users choose it | When users choose it |
| Sources show as | Citation links and cards | Links woven into the answer | Grounding links, sometimes none |
| Click behavior | Absorbs simple-query clicks | Fewer, deeper clicks | Fewest clicks, strongest advice |
| The game | Rank + be extractable | Cover the subtopics | Be remembered + be indexed |
Game 1: AI Overviews, the summary box
AI Overviews triggers per query, mostly informational ones, and assembles a short answer with citations. The pages it cites overwhelmingly share two traits: they already rank well for the query, and they contain a block that answers it directly, the same extractability that wins featured snippets.
Its business effect is absorption: on queries a paragraph can settle, the paragraph settles them above your listing. The response is the familiar one: win the citation where you can, and weight your portfolio toward the query depths a summary cannot finish.
Game 2: AI Mode, the conversation
AI Mode is search rebuilt as chat, and its defining mechanic is fan-out: your question gets decomposed into multiple sub-searches, whose results the model reads and synthesizes into one answer with follow-ups.
That mechanic changes what wins. A page that covers a topic's subquestions comprehensively can surface for angles no one typed, because the fan-out typed them instead. Depth of coverage, the related questions, the comparisons, the edge cases, becomes retrievable surface area rather than decoration.
Game 3: Gemini, the assistant
Gemini plays closest to ChatGPT's rules: a blend of model memory and live grounding through Google Search. Recommendations draw on what the model learned about brands, the mention currency, checked against what retrieval finds right now.
Which makes it the fullest test of the two doors at once: index presence gets you retrieved, reputation across the readable web gets you named, and neither substitutes for the other.
One strategy, three expressions
The corollary is a warning about effort allocation: chasing surface-specific tricks means re-optimizing every time Google reshuffles the surfaces, which it does continuously. The foundation transfers; the tricks expire.
Measuring across the three
Measurement is unglamorous here. Search Console folds AI Overview appearances into ordinary impressions with no separate report, so AIO impact reads as CTR movement on affected queries rather than a labeled number. Gemini sends identifiable referrals; AI Mode's activity lives inside Google's totals.
Practical setup: your AI referral segment catches the assistant traffic, quarterly SERP checks on money keywords log which now carry Overviews, and prompt tracking covers the recommendation layer across engines, Google's included.
The one-line takeaway: Overviews reward rank plus extractability, AI Mode rewards covering the question's whole fan-out, Gemini rewards memory plus index presence. Build the one foundation, express it three ways, and skip the tricks that expire with the next reshuffle.