Google runs two AI surfaces on the same query — and most Shopify merchants optimise for only one of them. AI Overviews sit on top of the SERP, answer in 40-60 words, and close themselves in under 12 seconds of user attention. AI Mode is a full conversational destination: multi-turn, citation-heavy, with product cards rendered inside the answer. The optimisation playbook diverges sharply, and the brand that wins one can easily lose the other.
Why Google is running two AI surfaces at once
AI Overviews launched as an in-SERP overlay: Google’s generative answer gets pinned above the organic listings, with a handful of citation chips, and the user either accepts the summary or scrolls past it to the classic blue links. AI Mode is the other half of the bet — a Gemini-powered conversational surface where the entire page is the AI answer, retrieval happens per turn, and the user can follow up indefinitely. Same query, two fundamentally different rendering surfaces, two different citation economies.
Internally, Google is hedging. AI Overviews protect the existing ad-loaded SERP from ChatGPT-style flight; AI Mode is the beachhead for a fully generative future where the ten blue links matter less. That split is relevant to you because the two surfaces pull from overlapping but non-identical pipelines, and they reward different content structures.
Anatomy of each surface
AI Overviews: fast, shallow, chips-based
The answer is short by design — Google’s internal guidance caps most Overviews around 40-60 words for commerce queries, with a handful of pull-outs as bulleted lists. The citation UI is a row of 3-5 small chips (favicon + domain), each linked to the source. Chip real estate is scarce and click-through is low: our cohort of Shopify merchants shows chip CTR in the 7-11% range, and most users never expand the full citation drawer. What matters is chip position and brand-name legibility, not long-form content.
AI Overviews retrieval is tuned for high-confidence, schema-validated answers. Product + Offer JSON-LD is the single strongest signal for commercial queries — miss it and Overviews falls back to informational pages like Wirecutter and Reddit, locking out direct merchants. FAQPage schema helps here too, because Overviews often lifts a Q&A pair verbatim when the query is question-shaped.
AI Mode: longer, deeper, multi-turn
AI Mode answers are 150-400 words, with citations appearing both inline (bracketed numerics in the prose) and in a right-rail sidebar. Product cards render inside the answer body for commercial queries — brand name, image, price, review rating, and a Buy button. Follow-up prompts are pre-suggested at the bottom: “how long do wool rugs last?” becomes a one-tap path into turn 2, where retrieval re-runs with accumulated context.
Retrieval is different too. AI Mode pulls wider source sets per turn (we’ve seen 12-18 sources consulted per answer vs. 4-7 for Overviews), weights brand corroboration across Reddit + editorial media more heavily, and rewards freshness. The multi-turn structure means a single query can produce 3-4 separate citation opportunities — turn 1 about “best wool rug,” turn 2 about “how long they last,” turn 3 about “care instructions.” Each turn has its own retrieval pass.
The optimisation matrix
Six levers matter across both surfaces, but the priority ordering flips depending on which surface you’re optimising for. A brand investing equally in both should treat these as a matrix, not a checklist.
Schema: medium for Overviews, critical for Mode
AI Overviews can and often does answer a commercial query without a valid Product + Offer block — it falls back to third-party editorial. AI Mode’s product cards require valid schema to render at all. No Product node with a complete Offer — no card. The card grid is roughly half the answer’s visual real estate, so missing it effectively means your brand gets cited in prose but never shown with price + Buy button. This is the single biggest divergence between the two surfaces.
Quotability: critical for Overviews, medium for Mode
Because Overviews answers are 40-60 words, the generator is effectively stitching 3-5 quotable sentences together from source documents. Pages written in long flowy prose with no extractable claims get skipped. Short, factual, self-contained sentences (“Wool rugs typically last 20-30 years with proper care”) are what Overviews pulls. AI Mode synthesises more and quotes less, so this lever matters less there.
Reddit corroboration: medium for Overviews, critical for Mode
AI Mode weights corroboration across independent sources much more heavily than Overviews does. When users ask “is brand X worth it,” AI Mode cross-checks your product page against Reddit threads, YouTube reviews, and editorial media before synthesising. A brand that only shows up on its own domain gets framed as under-reviewed and de-prioritised. Overviews cares less because it’s not synthesising opinion — it’s pulling a factual answer.
Follow-up / FAQPage coverage: low for Overviews, critical for Mode
AI Mode’s follow-up prompts are a huge unclaimed citation lane. When turn 1 returns “best wool rug,” the suggested follow-ups are typically “how long do they last,” “how to clean,” “are they pet-safe.” If your product + content pages cover those downstream questions (ideally with FAQPage schema), you re-appear as a citation on turn 2 and turn 3. If not, some other brand wins the follow-up. Overviews doesn’t have a follow-up mechanic, so this lever is dead weight there.
Freshness: high for both
Both surfaces discount stale pages. The threshold is looser for Overviews (18-24 months) and tighter for Mode (6-12 months for commercial queries, especially price-anchored ones). A product page with a dated price mention from 2024 reads as stale to AI Mode’s retriever and drops in rank. Rotate updated timestamps honestly (re-publish dates, updated-at markers) and keep price copy current.
Image quality: low for Overviews, high for Mode
Overviews rarely renders product imagery — it’s a text-first surface with occasional thumbnails. AI Mode renders product cards with images as primary visual anchors, and images below ~800×800 with busy backgrounds get down-ranked visually. The card engine prefers clean-background PNG/WebP at 1200×1200 or larger. Fixing this is a 2-hour exercise for most stores and meaningfully changes whether your card wins real estate.
How to measure both surfaces without blurring them
The most common measurement mistake we see: merchants pool “Google AI visibility” as a single number. That number moves when either surface moves, so the team can’t tell whether a schema fix actually helped AI Mode or whether a fresh FAQ moved Overviews. Split the tracking.
- Track AI Overviews separately. Capture chip appearance, chip position (1/2/3/other), and whether your brand name is the full brand or a domain truncation. Noise reduction: re-query three times per week and count presence, not exact phrasing.
- Track AI Mode separately. Capture whether your brand is cited in-prose (inline bracket), cited in the sidebar, and whether a product card renders. All three are distinct signals with different causes. Multi-turn: track turn 1 citation plus follow-up turn citations as separate rows.
- Map each query to a surface-trigger class. Some queries almost always trigger Overviews (simple factual, top-of-funnel), some trigger Mode more often (exploratory, multi-turn-shaped). Measure only queries that trigger the surface you’re optimising for, or you’ll flatten your own signal.
- Tag conversions by surface. Plausible or PostHog’s referrer handling can distinguish Overviews click-throughs from AI Mode click-throughs via the referrer URL pattern. Our cohort shows AI Mode converting at roughly 2x the rate of AI Overviews on identical product pages.
- Run comparison reviews monthly, not weekly. Both surfaces have re-ranking cycles in the 10-30 day range. Weekly snapshots are mostly noise. Monthly matches the actual cadence of meaningful change.
What to actually do this quarter
For Shopify merchants reading this with finite engineering and content hours, here’s the ruthless priority order. If AI Mode isn’t rendering your product cards, fix schema first — it’s the single biggest unlock, costs 2-3 dev days, and doubles your surface area on commercial queries. Next, write FAQPage-shaped answers to the three most common follow-up questions on your top 20 products. That’s maybe 60 Q&A pairs, which a content lead can ship in two weeks. Then, seed Reddit corroboration through honest community participation — not spam, not paid shills; just answering questions in your category subreddits when you have genuine expertise.
If you’re already winning AI Mode and your Overviews chip appearance is low, the priority flips: ship quotable-sentence factual claims in your product descriptions and category pages, add FAQPage schema with explicit short answers, and make sure your brand name is legible (not domain-truncated) in the chip UI. The two surfaces share a long tail of shared best practice, but the top-priority moves diverge enough that you should plan them separately.