AI Overviews Explained: How They Work and How to Get Cited

AI Overviews are the AI-generated answers Google shows at the top of many search results, built by Gemini from a handful of web pages it picks and cites as sources. They appear on a large share of informational searches in 200+ countries, they take a real bite out of clicks, and they have their own report in Google Search Console. This guide explains how AI Overviews work, what they do to your traffic, how to see which of your pages already show up in them, and how to earn citations, first by hand and then faster with Claude or ChatGPT.

Key takeaways

• AI Overviews are a summary with sources, not a new ranking system. Gemini fans your search out into several sub-queries, pulls pages from Google's normal index, and writes an answer with links. If you don't rank for the sub-queries, you can't be cited.
• Clicks drop, but citations soften the blow. Ahrefs measured a 58% CTR drop for the #1 result on keywords with an AI Overview (December 2025 data). Seer Interactive found cited pages get roughly double the CTR of uncited pages on the same results page.
• Search Console now shows it. Since mid-2026 there's a dedicated generative AI performance report with impressions by page, country, date and device for AI Overviews and AI Mode. There's also an opt-out toggle.
• Shopping searches are no longer exempt. AI Overviews show on about 14% of product-intent queries, and they tend to cite reviews, comparisons and buying guides instead of product pages.
• The playbook is answer-first content with original information, plus clean technical access, plus measuring what actually converts rather than what merely gets impressions.

What AI Overviews are and how they work

An AI Overview is a block at the top of a Google results page with a few paragraphs or bullets written by Gemini, plus links to the pages it used. Google launched them in the US in May 2024, expanded them to 200+ countries and 40+ languages during 2025, and in January 2026 made Gemini 3 the default model behind them globally. Google says they reach more than 2 billion people a month.

Under the hood, the process looks like this:

• Query fan-out. Your search gets split into several related sub-queries. "best running shoes for flat feet" might fan out into "running shoes overpronation," "stability vs motion control shoes," "flat feet arch support," and so on.
• Retrieval from the normal index. Each sub-query pulls candidate pages from Google's regular search index. There is no separate "AI index," which is why standard crawlability and ranking still gate everything.
• Grounding and synthesis. Gemini reads the retrieved passages, writes an answer, and attaches links to the pages that support each part of it.
• Decision to show. Google only shows the Overview when it decides generative help adds value; navigational, very simple and sensitive queries often get none.

Since January 2026 you can also tap a follow-up question inside an AI Overview and continue the conversation in AI Mode, carrying the context along.

AI Overviews vs. featured snippets vs. AI Mode

• Featured snippet: one source, quoted verbatim, usually above the blue links. Easiest to attribute.
• AI Overview: multiple sources, synthesized text, shown on a standard results page with ads and organic results underneath.
• AI Mode: a separate tab or follow-up flow, multi-turn, heavier fan-out, far fewer outbound clicks per search. Reported together with AI Overviews in Search Console.

Where AI Overviews show up (and where they don't)

Trigger rates vary by study and by month because Google keeps tuning them. What's consistent:

• Informational, question-style queries ("how to," "what is," "why does," "best way to") trigger them most often. Longer queries trigger more than short ones.
• Commercial research queries ("X vs Y," "best X for Y," "is X worth it") trigger them increasingly. Visibility Labs analyzed 20.9 million shopping SERPs in early 2026 and found AI Overviews on 14% of product-intent keywords, up from 2.1% in November 2025.
• Navigational and brand queries rarely trigger them.
• Sensitive health, finance and news topics get them less often or with heavier caveats.

If your traffic is mostly brand and transactional, the direct impact is smaller than the headlines suggest. If you live on "how to" and "best" content, you need a citation strategy, not just a ranking strategy.

What AI Overviews do to your traffic

The best available data:

• Ahrefs (300,000 keywords, December 2025 data): on informational keywords with an AI Overview, the CTR of the #1 organic result fell from 0.073% to 0.016% between December 2023 and December 2025, the "58% drop" you'll see quoted everywhere.
• Seer Interactive (53 brands, 5.47M queries, Jan 2025 to Feb 2026): organic CTR was about 3.3% on searches with no AI Overview, 2.1% when there was an Overview and your page was cited, and 0.9% when there was an Overview and you weren't cited. Overview CTR bottomed in December 2025 at 1.3% and climbed back to 2.4% by February 2026. Paid CTR barely moved.
• ISB / Carnegie Mellon randomized field experiment (1,065 Chrome users, 2026): showing an AI Overview cut clicks to publishers by roughly 40%, with the effect concentrated in informational queries and no measurable effect on transactional ones.

The takeaway isn't "SEO is dead." Clicks are being redistributed: cited pages keep a meaningful share, uncited pages lose most of theirs, and searches without an Overview have become more valuable. Your job is to win citations on the queries that matter and to know, in dollars, which pages still bring buyers.

How to see your AI Overviews data in Search Console

The manual way

Google added a standalone generative AI performance report to Search Console in June 2026 and rolled it out worldwide by late summer. How to read it:

1. Open Search Console and pick your property.

2. In the left menu, under Performance, open the generative AI report (there's one for Search, covering AI Overviews and AI Mode together, and one for Discover).

3. You'll see impressions (how often your URLs appeared inside an AI feature), broken down by page, country, date, and, for Search, device. Note: there's no click or query data in this report yet.

4. Sort by page. The pages at the top are the ones Gemini already trusts as sources. These are your starting point for expansion.

5. Cross-reference with the regular Search results report, filtered to the same pages and date range. High AI impressions with falling clicks means you're being summarized; high AI impressions with stable clicks means you're cited in a way that still drives visits.

6. Export both reports as CSV (the export button is top right).

Caveat: the regular performance report still lumps AI Overview clicks in with everything else, so there's no clean "clicks from AI Overviews" number yet. Treat the AI report as a visibility signal, not a revenue report.

The faster way with AI

Give the two CSV exports to Claude or ChatGPT with a prompt like this:

```

I'm attaching two Search Console exports for the last 3 months:

1) "ai-pages.csv" – impressions inside AI Overviews / AI Mode, by page.

2) "search-pages.csv" – total impressions, clicks, CTR and position by page.

Join them on the URL. Then:

- List the 20 pages with the most AI impressions and show their total clicks and CTR next to them.

- Flag pages where AI impressions are high but CTR is under 1% (summarized, not visited).

- Flag pages that rank in the top 5 but have zero AI impressions (ranking, but not cited).

- Group the URLs by type (blog, collection, product, other) based on the path.

Give me the result as a table, then 5 bullet points on what to prioritize.

```

To make it repeatable, use Claude Code: drop the CSVs in a folder, ask it to write a Python script that joins the exports and outputs the three lists, and rerun it monthly with new files.

How to optimize for AI Overviews

There's no AI-specific markup that gets you in. Google's own guidance is that what ranks you in normal search gets you cited: helpful, unique, crawlable content. What changes is the shape of the content, because Gemini extracts passages rather than reading pages top to bottom.

Step 1: Pick the queries worth winning (manual)

1. From the regular Search Console report, export your top 200 queries with impressions, clicks and position.

2. Search each important query in an incognito window and note whether an AI Overview appears, which pages are cited, and whether yours is one of them.

3. Keep three lists: cited (protect), ranking top 10 but not cited (fix), and AI Overview appears but you don't rank (long-term).

4. For each "fix" query, open the cited pages and note what they have that yours doesn't: a direct answer up top, a table, a named expert, original numbers, steps.

Step 2: Rewrite for extraction (manual)

For each page on the "fix" list:

• Answer in the first 40 to 60 words under the H1 and under each H2 that targets a sub-question. No warm-up paragraphs.
• Mirror the fan-out. The sub-queries Gemini generates become your H2s and H3s. If the query is "best espresso machine under $500," sections like "What you give up under $500," "Single boiler vs. heat exchanger," and "Grinder or not?" are the sub-questions it will go looking for.
• Add something that doesn't exist elsewhere: your own test data, a customer survey, a price table you maintain, an expert quote with a name and title. Google's guidance to focus on unique, non-commodity content is the biggest differentiator between cited and uncited pages on competitive queries.
• Use plain structure: short paragraphs, bullets, tables for comparisons, dates on anything that changes.
• Keep the technical basics clean: indexable, fast, no key text loaded behind JavaScript tabs, no `nosnippet` on pages you want cited. Blocking `Google-Extended` in robots.txt does not remove you from AI Overviews; the Search Console opt-out does (more below).

Example: a Shopify store selling cast-iron cookware ranks #3 for "how to season cast iron" but isn't cited. The cited pages all open with a 4-step summary and a specific oven temperature; the store's page opens with 300 words of brand history. Moving the steps to the top and adding a table of oils with smoke points is the fix; the history can stay, lower down.

Step 3: Do it faster with AI

Prompt for Claude or ChatGPT (one page at a time):

```

You are an SEO editor. Target query: "how to season cast iron".

Here's the current page content: [paste]

Here are the three pages currently cited in Google's AI Overview for this query: [paste the main text of each]

1. List the sub-questions a searcher would expect answered (think like Google's query fan-out: 5–8 related questions).

2. Say which of those my page answers, which it doesn't, and which it answers too late in the page.

3. Rewrite the opening so the first 50 words directly answer the main question.

4. Propose an H2/H3 outline that covers every sub-question, with a 40–60 word direct answer under each heading.

5. Suggest one piece of original data or expertise I could add that none of the cited pages have.

Don't invent facts; mark anything you're unsure about with [VERIFY].

```

Claude Code flow for a whole site section (blog or collection pages in a Shopify theme or any CMS export):

1. Export your pages to a folder of Markdown or HTML files, one per URL, plus a CSV with the target query per URL.

2. Ask Claude Code to: read every file, check whether the first 60 words answer the target query, list headings as questions, flag missing sub-questions, and write a `report.csv` with a 0–5 "extraction score" and the top 3 fixes per page.

3. Review the report, then ask it to apply the fixes to the files for the 10 lowest-scoring pages, leaving the rest untouched.

4. Paste the revised content into your CMS, publish, and check the Search Console AI report 4 to 8 weeks later.

Keep a human at step 3: the model will happily "improve" a page by making it more generic, which is exactly what gets you uncited.

AI Overviews for ecommerce and Shopify stores

Product pages almost never get cited on shopping queries. When an AI Overview appears for "best weighted blanket for hot sleepers," the sources are typically video reviews, forum threads, editorial comparisons and buying guides. Product data reaches Google's AI shopping surfaces through your Merchant Center feed, not a crawl of your product page. So a Shopify store has three jobs:

• Feed quality. Accurate titles, GTINs, prices, availability and shipping in Merchant Center. If the feed is wrong, the AI answer is wrong or your product isn't in it.
• Buying-guide content on your own domain. Collection pages with a real guide above or below the grid, and blog posts that answer "X vs Y," "how to choose," and "is X worth it" with your own data (return rates, sizing data, customer survey results). These are the pages that get cited.
• Reviews and comparisons off-domain. Encourage reviews on the platforms Gemini already cites in your category, with your product correctly named.

Then measure the right thing. An AI impression is not a sale. The question for a store is which organic pages still produce orders, and with what margin, so you double down on the guides that actually pay.

Tools to track AI Overviews

• Google Search Console (free). The only source of first-party data: AI impressions by page, country, date and device. No clicks, no queries. Best for: knowing which of your pages are already cited.
• Semrush and Ahrefs (paid). Both flag AI Overviews as a SERP feature in rank tracking and show which domains are cited for tracked keywords. Pros: keyword-level and competitor-level data. Cons: sampled, and a real cost for a small store.
• Dedicated AI visibility trackers (Otterly, Peec AI, Profound and similar). They re-run prompts across AI Overviews, AI Mode, ChatGPT and Perplexity and log citations. Pros: cross-engine view. Cons: another subscription, and synthetic prompts rather than real users.
• Manual incognito checks plus a spreadsheet. Free and accurate for your exact queries, painfully slow past 50 keywords. Fine for a first audit.

If you only do one thing: check Search Console monthly and run the join prompt above. Add a paid tracker when you have 100+ queries to defend.

Should you opt out of AI Overviews?

Since June 2026, Search Console has a toggle to exclude your site from AI Overviews and AI Mode without affecting regular rankings. The older levers, `nosnippet`, `data-nosnippet` and `max-snippet`, still work and apply to both AI Overviews and classic snippets.

When opting out makes sense:

• Your business model is the page view itself (ad-supported publishing) and the data says Overviews are summarizing you without sending visits.
• You publish gated research or proprietary methodology and don't want it paraphrased for free.

When it doesn't:

• You sell something. Seer's data says cited pages get more than twice the CTR of uncited pages when an Overview is present. Opting out doesn't remove the Overview; it just removes you from it, and your competitor becomes the source.
• Your traffic is mostly brand or transactional, where Overviews barely show anyway.

A middle path: keep buying guides and blog posts eligible, and apply `max-snippet` only on the few pages you don't want quoted at length.

How savvy.so does it

savvy.so connects your Shopify store to Search Console and shows, for each organic landing page, the orders it produced and the real profit of those orders (price minus product cost, shipping, payment fees and expenses), so you can see whether a page that's losing clicks to AI Overviews is losing money or just impressions. Its AI catalog tool writes collection descriptions, product copy and blog articles from keyword research in the answer-first format described above, directly in Shopify. It doesn't track AI Overview citations; use Search Console or one of the trackers for that. There's a free trial.

FAQ