Google AI Overviews & AI Mode Optimization

AI Overviews vs AI Mode – why I treat them as two separate surfaces

AI Overviews and AI Mode give similar answers but mostly cite different URLs, so I measure and optimize them separately. In Ahrefs' analysis from December 2025 (540,000 US query pairs, September 2025 data) only 13.7% of citations overlapped between the two, even though the answers were 86% semantically similar on average.

Both are widely available. Google's October 2025 expansion added more than 40 new countries and territories and over 35 languages, bringing AI Mode to over 200 countries and territories, including much of Europe. In most markets you can measure both surfaces on local-language queries today.

The practical takeaway: a page cited in the AI Overview for the main query isn't guaranteed a place in AI Mode, and vice versa. In the query map I record each surface in its own column, and in reports I never add them up into a single number.

An AI Overviews query map – how I build it and choose which pages to work on

The query map is a table of queries from your industry showing whether each triggers an AI answer, who Google cites there, where your page stands and what the query is worth to the business. From it I usually pick 10–30 pages to rework first.

Where the queries come from

• Search Console – queries your site already gets impressions for, especially those where CTR dropped while the position held steady. That's a common sign an AI Overview has appeared.
• SEO tools with SERP feature filters – Ahrefs, Semrush and similar tools let you filter queries where they track an AI Overview. I treat them as a candidate list, not an oracle, because AI answers come and go.
• Manual checks – I check key queries in a private window with the target market's location and language, because tools often measure from elsewhere.
• Customer questions – from sales calls and support, because in AI Mode people type full questions that aren't in keyword databases. The method is in the post on keyword research for AI.

What I record for each query

How I set priorities

1. First, queries with an AI answer where my page is in the top 10 but not cited. A passage rework is usually enough here, because Google already considers the page relevant.
2. Then comparison and pricing queries with high business value, even if the page ranks lower. According to Ahrefs (March 2026), 31% of URLs cited in AI Overviews weren't in the top 100 for that query, so a low position doesn't rule out a citation.
3. Last, purely informational queries with low value. I often drop them on purpose, because even a citation rarely turns into a customer.

Analysing the pages cited in AI Overviews – what I check on competitors' pages

For each query on the map I look at exactly which passage Google cited and why it may have picked it over your page. I don't copy competitors; I look for the missing element: a direct answer, data, format or freshness.

When you click a link in an AI answer, Google often scrolls to the cited passage and highlights it. That shows whether the citation is a one-sentence definition, a table row or a list item. It's the most useful piece of the whole analysis, because it shows which passage shape wins for that question.

Checklist for a cited page

• Answer placement – is the cited passage at the start of a section, under a question-style heading?
• Passage shape – a one-sentence definition, a list of steps, a table, a number with a unit?
• Specifics – does the passage contain a figure, a price range, a time frame, a standard or a regulation?
• Source and date – does the figure name a publisher and year, does the page show an update date?
• Author – is it clear who wrote it and where their experience comes from?
• Page type – guide, store category, forum, video, manufacturer page?
• Follow-up questions – which fan-out questions does this page answer that yours doesn't?

To compare content I use my own Content Intelligence tool, which sets your page against the cited ones and shows missing topics and questions. When a YouTube video gets cited, the signal is clear: according to Ahrefs (2026), 5.6% of all AI Overview citations are YouTube URLs from outside the top 100. That's when I consider video, as described in the post on YouTube and AI visibility.

Query fan-out in practice – covering follow-up questions without multiplying pages

Query fan-out is a set of concurrent, related queries the model generates to gather more sources for an answer. Google's own example: for "how to fix a lawn that's full of weeds", fan-out queries might include the best herbicides for lawns, removing weeds without chemicals and preventing weeds. Your page can be cited for answering a question the user never typed.

How I collect follow-up questions

1. I run the main query in AI Mode and note which subtopics appear in the answer and which follow-ups Google suggests.
2. I collect questions from "People also ask" and search autocomplete.
3. I generate a list of likely fan-out queries with Gemini through the API, prompting it to break the question into subtopics the way a search engine would. It's a simulation, not Google data, so I always check the output against what AI Mode actually shows.
4. I group the questions by intent and remove duplicates in meaning.

Section or separate page – how I decide

That boundary matters. Google's guide to optimizing for generative AI features (updated July 2026) says outright that creating separate content for every query variation, including fan-out queries, mainly to manipulate rankings or AI answers violates its scaled content abuse policy. So I cover follow-ups mostly within existing pages and internal links, and create new pages only where there's a distinct, real intent.

Reworking a page for AI Overviews – what I change and what I leave alone

I only rework passages with a real chance of being cited: a direct answer at the start of the section, a specific detail with a source and a clear structure. I don't chop whole articles into tiny pieces, because the same 2026 Google guide says there's no such requirement and no need to write in a special way for AI.

Those two positions don't conflict. The page should work for people, and a direct answer at the top of a section is simply good editing that AI finds easier to use. Patterns for these paragraphs are in the post on how to write content AI cites; here I focus on what's specific to Google.

The change list I go through for every page

• Answer in the first sentences of the section – especially for map queries where the page is in the top 10 but not cited.
• Follow-up questions as headings – based on the fan-out coverage map, without adding sections "just in case".
• A table where the question is about comparisons or specs – in HTML, not in an image.
• A first-hand element – your own test, a photo from a real project, data from an implementation. Google's guide encourages non-commodity content based on the author's own knowledge rather than rehashing common information.
• Images and video – where they support the text, following the image and video SEO practices Google points to in the same guide.
• Author, update date and structured data that match the visible content – Google stresses that markup has to reflect what's on the page. I cover the implementation in the Schema.org implementation service.
• Internal links – from the follow-up section to the page that expands on it.

Stores and local businesses

Google's AI answers can include products and information about local businesses. Google says Merchant Center product feeds and Google Business Profiles help with that. For stores and local businesses I therefore check those data sources as well as the site. Details are in the posts on e-commerce SEO and GEO and local SEO in the AI era.

Technical requirements and controlling your presence in AI Overviews – the 2026 picture

Any page that's indexed and eligible to appear in Google with a snippet can show up in AI Overviews and AI Mode. Google requires no extra files or special markup, and since 2026 it gives site owners a dedicated switch in Search Console.

The "Search generative AI" control in Search Console

Search Console settings now include a control covering AI Overviews, AI Mode and generative AI features in Discover. According to Search Console Help, it has been available to all websites worldwide since 31 August 2026. Sites are included by default. Excluding a site means its content isn't shown or linked in those features and isn't used to ground answers, and the site loses the traffic and impressions they would bring. The change usually takes effect within 1–2 days, and Google says the setting isn't a ranking signal for the rest of Search. It works at Search Console property level (child properties can inherit it), not per page.

What I check on the technical side

• Whether key pages are indexed and free of an accidental nosnippet or a too-low max-snippet (common in CMS templates).
• Whether the Search Console control has been switched to exclude, for instance by someone on the team after reading an article about protecting content from AI.
• Whether the content is in the HTML without JavaScript rendering, or renders correctly in the URL Inspection tool. Google recommends following JavaScript SEO practices; details are in the post on JavaScript SEO.
• Whether structured data validates and describes what's visible on the page.
• Whether duplicates are splitting signals across several similar URLs.

How I measure results: Search Console's Generative AI report, citation monitoring and conversions

I measure on three levels: impressions in AI features from the new Search Console report, citations on selected queries from monitoring, and clicks and conversions from organic traffic. None of these sources gives the full picture on its own.

The Generative AI report in Search Console

In June 2026 Google launched performance reports for generative AI features in Search Console, separately for Search and Discover, and since 31 August 2026 they've been available to all websites. The report shows how often URLs from your site appeared in AI features such as AI Overviews and AI Mode, broken down by page, country, device and date. The same data is still included in the overall Performance report too.

Google's announcement lists impressions as the report's metric and says more will be added over time, so I still analyse clicks and queries in the regular Performance report. In practice the Generative AI report answers "which pages does Google show in AI answers, and are there more after the rework", not "how many customers did that bring".

The metrics I report

Context matters when reading the data. In a 2026 Ahrefs study (300,000 keywords, December 2023 compared with December 2025), the presence of an AI Overview was associated with a 58% lower average CTR for the top result. A drop in clicks with stable rankings doesn't necessarily mean something is wrong with the page, and a citation won't bring back pre-AI traffic levels. That's why I set goals in conversions and citation share, not clicks alone. The GA4 and server-log setup is in the post on measuring traffic from ChatGPT, Perplexity and AI Mode, and the strategy for falling clicks in the piece on zero-click search.

Scope and timeline for AI Overviews and AI Mode optimization

The first phase usually takes 8–12 weeks: analysis and the query map, rework of selected pages, then measurement and fixes. Pricing starts from €1,200 for the analysis and rework of key pages; ongoing monitoring and updates are billed monthly.

If you first want to know how AI describes your brand across all assistants, start with an AI visibility audit. Optimization for ChatGPT, Perplexity and Copilot, which rely on other indexes, is covered by Perplexity, Copilot and ChatGPT Search optimization, and getting Google to recognise the company as an entity by knowledge graph engineering. For the wider context of Google's changes, see the post on AI Overviews and AI Mode.