AI Overviews vs. AI Mode: how do the two search experiences differ?
AI Overviews and AI Mode are both part of Google Search’s development, but they are not the same search experience.
AI Overviews appear within traditional Google results when Google’s systems determine that a generative summary can provide additional value beyond the standard search results.[1]
They are designed, among other things, to help users understand the main points of a more complex topic and provide links for further exploration.
AI Mode, by contrast, is a dedicated search mode designed particularly for more complex questions, comparisons, deeper exploration and follow-up questions.[1]
AI Mode has been available in Poland since October 2025.[3]
| Feature | AI Overviews | AI Mode |
|---|---|---|
| Location | part of traditional Google results | dedicated search mode |
| Typical use | a quick synthesis of a more complex topic | deeper exploration, comparisons and follow-up questions |
| Links to websites | yes | yes |
| Query fan-out | may be used | may be used |
| Models and techniques | may differ from AI Mode | may differ from AI Overviews |
The final point is particularly important.
Google explicitly states that AI Overviews and AI Mode may use different models and techniques.
As a result, the responses and links shown in the two experiences may differ.[1]
Query fan-out: how Google expands one question into multiple searches
One of the most important changes in generative Search is query fan-out.
Google defines it as a set of concurrent, related queries generated by the model to retrieve additional search results needed to answer the user’s question.[2]
For AI Mode, Google has also described the process as breaking a question into subtopics and issuing multiple searches at the same time.[4]
A simple query fan-out example
Imagine that a user asks:
A traditional SEO analysis might focus primarily on the main phrase:
A generative system, however, may need information from a much broader set of areas.
Illustrative related searches could concern:
- ecommerce SEO audits;
- an agency’s ecommerce experience;
- indexation problems on large product catalogues;
- link building for ecommerce;
- measuring SEO performance;
- ecommerce migrations;
- category and product optimisation.
We do not know the exact related searches generated for a particular user.
The example simply demonstrates why answering a complex question may require information from several areas.
User question
related search A
+ related search B
+ related search C
retrieval of relevant information from sources
generative response + links to websites
Do not turn this into a technical diagram claiming to describe Google’s complete internal architecture.
The full internal pipeline is not public.
The diagram should explain the concept of query fan-out, not pretend to document Google’s systems.
How does Google select sources for a generative response?
Google does not, of course, publish the complete algorithm used to select sources for AI Overviews or AI Mode.
We can, however, work from information Google has officially disclosed.
Google confirms that its generative Search features are rooted in its core Search ranking and quality systems.[2]
Its official guidance also describes retrieval-augmented generation, or RAG.
In this process, Search systems retrieve relevant, up-to-date pages from Google’s Search index and use information from those retrieved pages to generate a response.[2]
Query fan-out extends that process because the system can search not only for the original question, but also for additional related queries.
What can we reasonably conclude?
- Google’s traditional Search index still matters;
- ranking systems are involved in retrieval;
- one question can trigger more than one search;
- different parts of a response can rely on different sources;
- AI Overviews and AI Mode can show different links.
What can we not conclude?
- that we know the exact weight of individual signals;
- that there is a public “AI citation algorithm”;
- that ranking first automatically produces a citation;
- that every response uses the same set of related searches;
- that we know every internal stage of source selection.
Traditional rankings and AI citations: what does the data actually show?
If Google uses its core ranking systems to retrieve information, a natural question follows:
does a high organic ranking increase the probability of being included in an AI Overview?
There appears to be a relationship, but it is not simple and it does not mean that traditional ranking directly determines citation.
Interesting data was published at the end of 2025.
Search Engine Land reported on a Surfer SEO analysis of 10,000 keywords.[6]
Within that sample:
- 76% of the analysed keywords triggered an AI Overview;
- approximately 33,000 potential fan-out queries were reconstructed using Gemini;
- researchers found a strong correlation between the number of fan-out queries for which a page ranked and the likelihood of being cited;
- the reported Spearman correlation was 0.77.[6]
At the same time, the analysis revealed an important limitation of trying to explain citations using traditional rankings alone.
Around 68% of cited pages did not rank in Google’s Top 10 for either the primary query or any of the analysed fan-out queries.[6]
For the three most visible citations, that share was lower, but still around 46%.[6]
How should we interpret this?
Not as a new ranking factor.
The analysis identifies correlation, not causation.
The fan-out queries used in the study were also reconstructed using Gemini.
They were not taken from Google’s internal query logs.
The report additionally noted that only around 27% of the reconstructed fan-out queries remained consistent across repeated runs.[6]
This illustrates how dynamic the process can be.
Do keywords still matter?
Yes.
Query fan-out does not make traditional keyword research irrelevant.
It does, however, change how we should interpret an individual query.
Keywords still help us understand:
- the language people use;
- common problems;
- search intent;
- how products and services are described;
- demand around particular topics.
Problems arise when an SEO strategy is reduced to:
one keyword = one URL.
A complex user question may contain several different aspects of a problem, and query fan-out allows Google to search for those aspects in parallel.
Google also notes that its systems have become better at understanding page relevance even when there is no exact match between every query variation and the page’s primary content.[2]
A better model for content planning
Instead of asking only:
“What keyword should this page rank for?”
also ask:
- what problem is the user trying to solve;
- what questions might come next;
- what information is needed to make a decision;
- which topics genuinely require a separate page;
- which are simply parts of a broader topic;
- how related pages should connect to each other.
Do you need more content for AI search?
Query fan-out can create a tempting idea:
if Google runs many additional searches, why not create a separate page for every possible subquery?
That is the wrong approach.
Google’s current guidance for generative Search explicitly warns against creating separate content for every possible search variation primarily to manipulate rankings or generative AI responses.[2]
Doing so may fall under Google’s scaled content abuse policy.
A high number of pages does not automatically make a website a better source.
Develop the topic, not just the URL count
A more sensible approach is to:
- define the primary intent of each page;
- expand important resources with relevant subtopics;
- create a separate URL only when there is a genuinely distinct user need;
- reduce duplication;
- connect related resources through internal links;
- update useful existing pages instead of automatically creating new ones.
This approach makes sense not only because of AI Search, but also because it aligns with the fundamentals of well-planned SEO.
How are AI Overviews changing user behaviour on search results pages?
The changes concern more than the way Google retrieves information.
They may also affect how people interact with the search results page.
Useful observational data was reported by Search Engine Land in May 2026.[7]
Eric Van Buskirk of Clickstream Solutions analysed anonymised clickstream data provided by Surfer SEO.
The sample included approximately 846,000 US-based Google search sessions collected in February and March 2026.[7]
One pattern was that users more frequently returned to parts of the SERP they had already viewed.
Among sessions where users reversed their scrolling direction, the median share of scrolling back upwards was around 47.5% on SERPs containing an AI Overview.
Without an AI Overview, the equivalent figure was around 27%.[7]
The analysis interprets this as a possible signal that an AI Overview SERP can act more like a comparison and reconsideration environment.
That does not mean that every user behaves in this way.
What might this mean for SEO?
It becomes increasingly important not only to gain visibility, but also to consider how a brand and result appear alongside other information.
Factors that still matter include:
- a clear title;
- a precise description of the topic;
- brand recognition;
- consistency between the result and the destination page;
- genuine value after the user clicks.
This does not mean that there is one universal “AI Overview title formula”.
How can you measure AI Overviews and AI Mode visibility in Search Console?
Until recently, one of the largest challenges in generative Search SEO was the lack of dedicated visibility reporting.
On 3 June 2026, Google began rolling out dedicated Generative AI performance reports in Search Console.[5]
The reports cover generative features in:
- Google Search, including AI Overviews and AI Mode;
- Discover.
Google currently provides:
- impressions – how often URLs from the website appeared in generative AI features;
- pages – which URLs appeared;
- countries – where the visibility occurred;
- devices – available for Search;
- dates – performance over time.[5]
| Data | Available in the current report |
|---|---|
| Impressions | yes |
| URLs / pages | yes |
| Countries | yes |
| Devices | yes, for Search |
| Dates | yes |
| Exact fan-out queries | no |
| Individual citation CTR | no |
| Complete generation path | no |
Google is rolling these reports out gradually to a subset of websites.
If the report does not appear in a particular Search Console property, that does not necessarily mean the website never appears in generative Search features.[5]
What do AI Overviews, AI Mode and query fan-out change in SEO strategy?
The most useful conclusion is not:
“we need different SEO for AI”.
A better conclusion is:
we need to look at search visibility more broadly than one keyword and one ranking position.
1. Keep investing in traditional SEO fundamentals
Google itself states that generative Search features are rooted in its core ranking and quality systems.[2]
Technical SEO, indexing, content and website architecture have not stopped mattering.
2. Analyse the topic, not only one query
Query fan-out shows that one user question can lead to several related searches.
Understanding the full user problem therefore matters alongside tracking the most popular keyword.
3. Do not create a page for every subquery
Google explicitly advises against this approach.[2]
A separate URL should exist because there is a genuine user need, not simply because a tool generated another search variation.
4. Build logical information clusters
Service pages, articles, FAQs, examples and related resources should form a coherent information system.
This does not mean mass-producing content.
It means giving important parts of a topic a clear place within the website.
5. Measure visibility at several levels
- traditional rankings;
- impressions;
- landing pages;
- visibility in generative features;
- website traffic;
- leads;
- sales or other conversions.
One metric no longer describes the entire search journey.
Summary
AI Overviews and AI Mode do not replace the fundamentals of SEO.
Google officially confirms that its generative Search features are rooted in its core Search ranking and quality systems.[2]
What is changing is the process of finding information.
With query fan-out, one user question can lead to multiple concurrent searches, and a generative response can use different websites for different parts of the problem.[1][2]
This means that a ranking for one primary keyword increasingly provides only part of the visibility picture.
That does not mean the end of keywords, traditional rankings or SEO.
It means combining them with analysis of:
- intent;
- the broader topic;
- relationships between pages;
- generative visibility;
- real user behaviour;
- business outcomes.
Query fan-out is also not a reason to produce dozens of additional pages.
A better approach remains well-organised, useful content that genuinely helps people solve their problems.
Read also: How to check if your website is ready for AI search: a practical SEO checklist
Frequently asked questions
Do AI Overviews and AI Mode use the same sources?
Not necessarily.
Google states that AI Overviews and AI Mode may use different models and techniques, so the responses and links they show can differ.[1]
Do you need to rank in Google’s Top 10 to appear in an AI Overview?
No.
Strong organic visibility may be associated with a greater chance of a page being retrieved, but it is not a requirement for citation.
In the Surfer SEO analysis reported by Search Engine Land, around 68% of cited pages did not rank in the Top 10 for either the primary query or the analysed fan-out queries.[6]
This was a correlational study, and the fan-out queries were reconstructed rather than taken from Google’s internal data.
Does query fan-out mean keywords no longer matter?
No.
Keywords still help with understanding language, demand and search intent.
Query fan-out simply demonstrates that one user question can involve multiple related aspects of a topic.
Should you create a separate page for every fan-out query?
No.
Google explicitly advises against creating large amounts of separate content for every possible search variation primarily to manipulate rankings or generative responses.[2]
A separate page should exist because it serves a distinct user need.
Can you measure AI Overviews and AI Mode in Google Search Console?
Partially.
Since June 2026, Google has been gradually rolling out dedicated Generative AI performance reporting to a subset of websites.[5]
The reports include impressions, pages, countries, devices and performance over time.
They do not provide the exact fan-out queries or the complete path used to generate an AI response.