SEO in the age of AI Overviews and AI Mode: how to increase your chances of being cited by Google AI and protect organic traffic in 2026

Google Search is changing faster than the traditional SEO market became accustomed to stable rules.

In June 2026, Google said AI Overviews had more than 2.5 billion monthly active users, while AI Mode had surpassed one billion monthly users.^[1] In May, Google also said AI Mode queries had more than doubled every quarter since launch.^[17]

For site owners, this changes the search journey.

The classic path was often:

query
→ ten blue links
→ click
→ website

A growing share of search journeys now looks more like:

query
→ AI answer
→ query fan-out
→ multiple sources
→ supporting links / citations
→ follow-up question
→ possible click

That creates two simultaneous effects.

The opportunity is that Google can surface a page as a source even when it is not the number-one classic organic result for the exact wording of the user's query.

The challenge is that the AI answer may satisfy some of the user's intent without a visit to a website.

As a result, SEO in 2026 cannot be evaluated only through:

position
CTR
clicks

It increasingly requires understanding:

visibility in AI answers
grounding sources
fan-out queries
brand visibility
Preferred Sources
generative AI impressions
click quality

The most important caveat is this:

Google has not announced a separate “GEO ranking system” that replaces SEO.

Google's official guidance is explicit: the same SEO foundations remain relevant for AI Overviews and AI Mode. There is no special required schema, magic AI file or mandatory markup that makes a page eligible for generative Search.^[2][3]

This article separates what is documented, what is third-party research and what is still speculation.

Information status: August 31, 2026.

TL;DR

Question	Verified answer
Does SEO still matter?	Yes. Google says SEO remains foundational for AI Overviews and AI Mode
Is there a separate GEO ranking system?	Google does not describe one that replaces SEO
Is llms.txt required?	No, not for Google Search
Does llms.txt improve Google AI visibility?	Google says no
Is special AI schema required?	No
Must the page be indexed?	Yes, to be eligible as a supporting link
Must the page be eligible for a snippet?	Yes
Does crawlability matter?	Yes
Can JavaScript pages appear?	Yes, if content is accessible and JS SEO best practices are followed
Does structured data matter?	Yes as ordinary SEO when it matches visible content; there is no special AI schema
Does Google use query fan-out?	Yes
Does TOP 10 guarantee a citation?	No
Does a citation guarantee a click?	No
Can sites opt out of generative Search features?	Google is testing a dedicated Search Console control
Does Google-Extended remove a site from AI Overviews?	No
Does nosnippet affect AI Overviews and AI Mode?	Yes
Is there an AI report in Search Console?	Yes, rolling out to a subset of properties
Does Preferred Sources work with AI Overviews and AI Mode?	Yes where those features are available
Is there an official Preferred Sources button?	Yes, Google documented it in August 2026
Do AI Overviews reduce CTR?	Independent studies find substantial reductions for some query sets; Google reports stable aggregate organic clicks and higher click quality
Can citation be guaranteed?	No
Best strategy	original source-quality content + technical SEO + brand strength + first-party evidence + good UX + measurement

AI Overviews and AI Mode are no longer edge experiments

Scale changes the SEO equation.

Google reported in June 2026:

AI Overviews: >2.5B monthly active users
AI Mode: >1B monthly users

^[1]

Google also said that one year after AI Mode launched, queries in the mode had more than doubled every quarter since launch.^[17]

For many query classes, generative AI is now part of the mainstream Google Search interface.

AI Overview vs AI Mode

AI Overview is an AI-generated summary embedded in the conventional search results page.

AI Mode is a more conversational Search experience where users can continue with follow-up questions.

Google says AI Overviews and AI Mode may use different models and techniques, which means they may show different answers and different source links.^[3]

SEO practitioners often call those links:

citations
AI citations
AI sources

Google itself more often uses terms such as:

supporting links
web resources
sources

The most important technical rule: you still need to qualify for Google Search

Google states that a page must be:

indexed
+
eligible to appear in Google Search with a snippet

to be eligible as a supporting link in AI Overviews or AI Mode.^[3]

There are no additional technical eligibility requirements specifically for these AI features.

That keeps the classic technical foundation critical:

• crawlability,
• indexability,
• canonicalization,
• correct status codes,
• internal linking,
• accessible main content,
• JavaScript rendering,
• page experience,
• compliance with Search policies.

“GEO” cannot rescue a page that cannot be crawled or indexed correctly.

Google's official position: AEO and GEO are still SEO

Google's 2026 optimization guide addresses:

AEO = Answer Engine Optimization
GEO = Generative Engine Optimization

and says that from Google Search's perspective, optimizing for generative AI Search is still optimizing for Search, which means SEO.^[2]

User behavior has changed dramatically.

But Google has not announced a replacement protocol for the technical and quality foundations of Search.

How does Google build an AI answer?

Google documents two especially important mechanisms.^[2]

Retrieval-Augmented Generation

The system can use Google's ranking and retrieval systems to fetch relevant and current pages from the Search index, then ground the generated response in information from those pages.

Query fan-out

The model can generate multiple related searches around subtopics and retrieve results in parallel.

A query such as:

best hosting for an online store

may imply subtopics such as:

WooCommerce hosting performance
ecommerce backup
uptime
Redis object cache
PCI DSS
CDN

Google does not expose the exact fan-out queries for every AI response.

Query fan-out changes keyword research

Traditional SEO often starts with:

one keyword
→ one SERP
→ analyze top 10

A better mental model for AI Search is:

one user intent
→ many subproblems
→ many fan-out queries
→ multiple SERPs
→ a pool of possible sources

This is not a reason to create hundreds of thin pages for every imagined variation.

Google explicitly warns that mass-producing pages for query variations primarily to manipulate rankings or generative AI responses may violate its scaled content abuse policy.^[2][10]

The safer strategy is:

topic depth
+
clear information architecture
+
original evidence
+
strong internal relationships between subtopics

Ranking in the top 10 does not guarantee citation

Ahrefs published a March 2026 study based on roughly 863,000 SERPs and 4 million AI Overview URLs.^[14]

It reported that around:

37.9%

of cited URLs also appeared in the first ten result blocks of the same SERP.

When Ahrefs considered standard blue links, it reported roughly:

37.1% of cited URLs ranked in top 10
26.2% ranked 11–100
36.7% did not rank in top 100 for the exact query

^[14]

This is third-party research, not Google data.

It is nevertheless consistent with Google's documented query fan-out behavior: sources may come from related retrieval paths rather than the exact-query SERP alone.

Do not optimize only for the exact keyword

A strong page often needs to address not only:

what?

but also:

why?
when?
for whom?
how?
how much?
what are the limitations?
what changes the answer?
what if it fails?
what are the alternatives?

That does not mean artificially increasing word count.

Google explicitly says it has no preferred word count and recommends content that leaves the user satisfied.^[9]

The strongest content advantage in 2026 is information that is hard to commoditize

Google's updated AI optimization guidance emphasizes:

unique
valuable
non-commodity content

^[2]

Commodity content is easy for anyone to reproduce from the same public facts.

Examples of weak differentiation:

10 SEO tips
what is hosting?
how SSL works

when they contain no original evidence or expertise.

Stronger differentiation includes:

• proprietary data,
• an original benchmark,
• a reproducible test,
• shared methodology,
• code and observed results,
• real screenshots,
• deployment lessons,
• incident analysis,
• experiments,
• expert trade-off analysis,
• a unique comparison table.

Citation-worthiness starts with being a source, not a summary

If an article merely rewrites ten other pages, there is little reason for a retrieval system to treat it as the best primary source.

Google's people-first guidance asks whether content contains:

• original information,
• reporting,
• research,
• analysis,
• substantial value beyond existing results.^[9]

A useful editorial pattern is to include sourceable units such as:

definition
methodology
result
table
specific number
diagram
checklist
procedure
comparison
conclusion based on data

Do concise answers under headings help?

Google does not publish a rule such as:

40–60 word answer = AI citation

No such guarantee exists.

However, clear information architecture is still strong SEO and good UX.

For example:

## Does llms.txt improve visibility in Google AI?

No. Google says Search does not require llms.txt and the file has no positive or negative impact on Google Search visibility or rankings.

The explanation can then continue below.

This is useful because it answers the reader quickly, not because it is a secret AI-ranking trick.

llms.txt: one of the biggest AI SEO myths

In June 2026 Google clarified that llms.txt:

• is not required for Google Search,
• does not improve visibility,
• does not reduce visibility,
• can still be maintained for other systems that use it.^[16]

Google also says AI Overviews and AI Mode do not require:

new machine-readable files
AI text files
special markup
special schema.org data

^[2][3]

Therefore:

llms.txt = potentially useful for other services
llms.txt ≠ Google AI ranking factor

Structured data still matters, but it is not a ticket into AI Overviews

Google recommends that structured data accurately match visible page content.^[3]

Schema can help Search understand a page and make it eligible for supported rich-result features.

But:

Article schema
Organization schema
Product schema

are not special AI citation tags.

There is no required:

AIOverview schema
GEO schema
Citation schema

for Google Search.

Use structured data for its real documented purpose.

FAQ schema is no longer a route to a Google FAQ rich result

Many old SEO checklists still recommend FAQ schema for additional SERP real estate.

Google retired FAQ rich results from Search starting May 7, 2026.^[16]

An FAQ section may still be excellent for users.

It should not be justified by a promise of a classic FAQ rich result.

Crawlability matters more than AI hacks

Google's current AI optimization guide again emphasizes crawlability.^[2]

Audit:

robots.txt
CDN / WAF
HTTP status
canonical
internal links
rendering
sitemap
noindex
HTTP headers

Google can process JavaScript content when it is not blocked, but JavaScript SEO remains more complex than static HTML and should be tested carefully.^[2]

Google-Extended does not control inclusion in Google Search

This is a common misunderstanding.

Google-Extended is a robots.txt product token controlling whether content Google crawls may be used for certain Gemini uses, including future model training and specified grounding uses outside standard Google Search.^[8]

Google explicitly says:

Google-Extended does not impact inclusion in Google Search
and is not a ranking signal in Google Search.

^[8]

Therefore:

User-agent: Google-Extended
Disallow: /

does not mean:

exclude my pages from AI Overviews

How can you restrict direct use in AI Overviews?

Google documents multiple controls.

nosnippet

<meta name="robots" content="nosnippet">

prevents a text snippet and also prevents page content from being used as direct input for AI Overviews and AI Mode.^[7]

max-snippet

<meta name="robots" content="max-snippet:160">

limits the amount of text that can be used in snippets and as direct input.^[7]

data-nosnippet

This can exclude selected page fragments from snippets.

These controls have broader Search consequences, so they should not be deployed mechanically.

The new generative AI setting in Search Console

In 2026 Google began testing a dedicated Search Console setting for participation in generative AI Search features.^[1][5]

It covers:

• AI Overviews,
• AI Mode,
• generative AI features in Discover.

Sites are included by default.

If excluded, site links and content should no longer appear in those generative features and the property should no longer receive their associated impressions or traffic.^[5]

Google also clarifies that this setting:

• is not a ranking signal for other Search features,
• does not control AI training,
• does not replace noindex.

The control is rolling out to a subset of properties.

Search Console now has a dedicated generative AI report

Google launched a:

Generative AI performance

report for Search and Discover.^[4]

The Search report currently covers:

• AI Overviews,
• AI Mode.

It includes dimensions such as:

• impressions,
• pages,
• countries,
• devices,
• dates.

The data also remains part of the overall Search Performance report.^[4]

The dedicated view is still being rolled out and is not available to every property.

Important August 31 caveat: AI impression logging is currently undercounted

Google's Search Console Data Anomalies page says that from August 13, 2026, a logging issue has caused undercounting of impressions in the generative AI Search performance report.^[18]

Google says this is a reporting issue, not a serving issue.

Therefore, an apparent August drop in AI impressions should not automatically be interpreted as a real visibility loss.

This matters for any August 2026 SEO reporting.

How Search Console counts AI Overviews and AI Mode

Google documents the methodology.^[4]

AI Overviews

An external link click counts as a click.

Standard impression rules apply.

An AI Overview occupies a single Search position, and all links within that overview are assigned that same position.

AI Mode

External link clicks count as clicks.

A follow-up question is effectively treated as a new query, so impressions, positions and clicks in the follow-up response are counted against that new query.^[4]

Preferred Sources is one of the most practical 2026 changes

Google lets users select publications as:

Preferred Sources

^[6]

For users who select a domain, its content may:

• appear more prominently in Top Stories,
• receive a “Preferred” badge,
• be highlighted in AI Overviews,
• be highlighted in AI Mode.

This is not a global ranking boost for all users.

It is user-level personalization.

Google now offers an official Preferred Sources button

On August 20, 2026, Google documented a publisher-facing interactive Preferred Sources button.^[6][16]

Basic implementation:

<script async src="https://news.google.com/swg/js/v1/publisher.js"></script>

<div google-add-preferred-source-btn></div>

A dark theme can be selected:

<div google-add-preferred-source-btn data-theme="dark"></div>

Google also provides a custom JavaScript integration and a deeplink approach.

For publishers with returning audiences, this is one of the clearest official mechanisms for turning brand loyalty into personalized Search visibility.

A citation is not the same as a brand mention

Being used as a source and having the brand visibly named in the AI answer are separate outcomes.

Measure them separately:

citation / source appearance
brand mention
click
conversion

A citation can create visibility without a click.

But from a brand-building perspective, it is stronger when the user also recognizes the source.

Original datasets, named methodologies and first-party studies can make information more attributable to a specific publisher, although Google does not guarantee a brand mention.

Do AI Overviews reduce clicks?

There is no single universal number because studies use different samples and methodologies.

Pew Research Center

Pew analyzed browsing behavior from 900 U.S. adults and 68,879 unique Google searches from March 2025.^[12]

It found:

traditional-result click when AI summary appeared: 8% of visits
traditional-result click without AI summary: 15% of visits
click on a source inside the AI summary: 1% of visits

^[12]

Ahrefs

Ahrefs analyzed 300,000 keywords and aggregated Google Search Console data.

Its February 2026 update estimated that in December 2025 the presence of an AI Overview correlated with roughly:

58% lower CTR

for the number-one organic result relative to its modeled no-AI-Overview scenario.^[13]

That is an Ahrefs study, not an official Google metric, and it should not be applied as a guaranteed 58% loss to every site.

Google reports a different traffic picture

In August 2025 Google said total organic click volume from Search to websites was relatively stable year over year and that average click quality had increased.^[15]

Google described “quality clicks” as visits where users do not quickly return to Search.

At first glance this may look inconsistent with Pew and Ahrefs.

The methodologies are not equivalent.

Google is describing an ecosystem-wide aggregate.

Pew measured the behavior of a specific panel.

Ahrefs modeled CTR across selected query classes and rankings.

For a publisher, first-party data remains the most important evidence.

SEO KPIs for 2026

Traditional KPIs remain:

• organic clicks,
• impressions,
• CTR,
• position,
• conversions,
• revenue.

Useful additional metrics include:

AI impressions
AI source appearances
brand mentions in AI answers
share of cited pages
citation-to-click rate
conversion rate from AI-origin traffic
Preferred Sources adoption
branded search growth

Not all of these are available directly from Search Console.

Some require third-party or internal monitoring.

How should an article be structured for both Search and AI Search?

A useful pattern is:

clear H1
→ concise introduction
→ TL;DR
→ logical H2 sections
→ definitions
→ primary evidence
→ methodology
→ tables
→ examples
→ limitations
→ checklists
→ FAQ
→ accurate update date
→ footnotes

Google does not require this template.

It works because it improves human readability and makes claims easier to connect with evidence.

What content increases your “citation surface”?

The strongest content is sourceable.

Original research

sample size
methodology
results
raw observations

Benchmarks

same dataset
same prompts
same hardware
same scoring criteria

Technical tutorials

problem
configuration
code
result
edge cases

Incident analysis

timeline
root cause
IOC
remediation
lessons learned

Comparisons

Not:

A is better than B

but:

A wins scenario X
B wins scenario Y
methodology
cost
limitations

Multimedia is increasingly valuable

Google explicitly recommends supporting text with relevant, high-quality images and video when appropriate.^[2]

Generative Search can surface visual media as well.

Useful assets include:

• original diagrams,
• screenshots,
• benchmark charts,
• short demos,
• YouTube videos,
• transcripts,
• descriptive alt text.

Ahrefs' external research also found meaningful YouTube representation among AI Overview sources, but this is an observation rather than a Google ranking rule.^[14]

AI-generated content is allowed, but scaled low-value production is risky

Google does not ban content simply because AI assisted its creation.

It does warn that using automation to mass-produce pages without added user value can violate scaled content abuse policies.^[10][11]

The important questions remain:

Is it accurate?
Is it original?
Was it verified?
Does it add value beyond its sources?
Does it solve the reader's problem?

Technical publishing also benefits from clear authorship, methodology and citations.

SEO checklist for AI Overviews and AI Mode

Technical SEO

• Page returns a correct 200.
• It is not blocked in robots.txt.
• There is no accidental noindex.
• Canonical points to the correct URL.
• Main content is accessible to Googlebot.
• JavaScript does not hide critical content.
• Internal links connect the page appropriately.
• Sitemap contains the canonical URL.
• CDN/WAF does not block Googlebot.
• Mobile UX works correctly.

Content

• The article adds more than a generic web summary.
• It contains original experience, analysis or data.
• Material claims have sources.
• Tables explain their methodology.
• Headings reflect real user questions.
• Answers are direct and avoid filler.
• Limitations and exceptions are explicit.
• Author and publisher are clear.
• Update date is accurate.
• Date changes correspond to real content updates.

AI Search

• We do not treat llms.txt as a Google ranking factor.
• We do not implement fictional AI schema.
• We analyze topics, not only exact-match keywords.
• We cover natural sub-intents and fan-out angles.
• We do not generate hundreds of thin fan-out pages.
• We monitor AI impressions in Search Console when available.
• We account for the August 13, 2026 reporting issue.
• We evaluate Preferred Sources for loyal audiences.
• We distinguish citation, mention and click.
• We do not promise guaranteed citations.

Brand and distribution

• We publish original assets.
• We develop video where the topic benefits from it.
• We build legitimate brand mentions outside our own domain.
• We publish source-quality research and data.
• Organization information is consistent.
• Local businesses maintain Business Profile data.
• Ecommerce businesses maintain Merchant Center data.
• We build a returning audience, not only one-off search clicks.

Measurement

• Track organic clicks.
• Track impressions.
• Track CTR.
• Track conversions.
• Track branded search.
• Track landing pages shown in AI reporting.
• Compare AI-visible pages with non-visible pages.
• Separate SERP changes from technical issues.
• Annotate core updates and Search changes.
• Avoid conclusions from one week of data.

A 30-day action plan

Days 1–5: technical audit

Review:

indexability
robots
canonical
status codes
rendering
internal linking
page experience
Search Console

Days 6–10: identify source-worthy topics

Choose topics where you can contribute:

first-party data
benchmark
test
case study
code
experience

Days 11–15: improve the best existing articles

Add:

clear definitions
tables
sources
methodology
limitations
FAQ
multimedia

Days 16–20: strengthen brand visibility

If eligible, implement:

Preferred Sources

and develop real distribution channels.

Days 21–25: monitoring

Compare:

Search Console
Generative AI performance
organic landing pages
AI citation monitoring
branded queries

Days 26–30: iterate

Do not automatically create 100 new pages.

Choose a small set of pages that already have demand, broad intent coverage and the potential to become a source of original information.

POLPROG verdict

SEO did not die because of AI Overviews and AI Mode.

The economics of visibility did change.

Classic Search could often be simplified as:

rank → click

Generative Search increasingly looks like:

crawl
→ index
→ relevance
→ fan-out
→ source selection
→ citation
→ brand recognition
→ click
→ conversion

The biggest mistake in 2026 is to search for shortcuts such as:

llms.txt
magic schema
LLM keyword stuffing
1,000 fan-out pages

Google explicitly says those special requirements do not exist.^[2][3][16]

The most defensible long-term strategy is:

technically healthy website
+
original expertise
+
first-party evidence
+
well-organized content
+
credible authorship and brand
+
multimedia
+
real usefulness
+
measurement

AI Search increases the value of information that is difficult to commoditize.

If every competitor can generate an identical article with one prompt, the SEO moat is weak.

If you publish original benchmarks, current evidence, rigorous technical guides, incident analyses, reproducible tests and transparent sources, you are creating something a user, a search engine and a grounded generative system can all use.

That is the most truthful answer in 2026 to:

How do I get cited by Google AI?

There is no “cite me” button.

But you can systematically increase the probability that your page becomes the best available source for part of an answer.