Did AI Overviews Cut Your Traffic? Diagnose the Loss Before You Change Your Content

Your rankings held, but your clicks did not.

Nothing obvious changed on the site. There was no redesign, no manual action, no clear tracking failure, and no meaningful loss of links. Every article you have read since offers the same explanation: Google answered the question inside an AI Overview, so the searcher never reached your page.

That may be what happened. The graph does not prove it.

Stable rankings beside falling clicks can also come from lower demand, seasonality, a changed query mix, stronger ads or search features above the organic result, a weaker title or snippet, or a reporting problem. Our diagnostic for ranking and traffic drops covers how to separate those causes.

This page begins after that diagnostic. The affected rankings broadly held, demand did not collapse, and the lost clicks are concentrated on pages and queries now exposed to AI-generated results.

The remaining question is economic: what were those clicks worth, which pages still perform a business function, and where should the next content dollar go?

The National AI Overview Percentage Is Not Your Percentage

AI Overview studies often appear to contradict one another. Sometimes the apparent contradiction exists inside research published by the same company.

Conductor’s September 2025 monthly analysis examined 118 million actively tracked keywords and found an AI Overview on 16% of them. A separate Conductor benchmark examined approximately 21.9 million unique U.S. Google searches between September 15 and October 12, 2025. Nearly 5.5 million of those searches generated an AI Overview, or approximately 25%.

Conductor later analyzed more than 274 million U.S. searches over several months. Within that research framework, AI Overview coverage rose from 23% in September 2025 to 47% in January 2026, then fell to approximately 34% in February.

Those figures do not mean one report was truthful and the others were defective. They came from different Conductor indexes, collection windows, query universes, and methodologies.

That distinction is the central point. A prevalence figure can be accurate inside its own panel and still be useless for your budget.

National research tells Rank Nashville what deserves investigation. It does not replace the queries that produced a client’s traffic, consultations, bookings, or revenue.

What the Conductor figures actually describe

Dataset	Scope	Finding	Safe interpretation
Monthly AIO analysis	118 million actively tracked keywords, September 2025	16% generated an AI Overview	Coverage inside Conductor’s monthly tracked-keyword index
AEO/GEO benchmark	21.9 million unique U.S. searches, September 15 to October 12, 2025	Nearly 5.5 million, approximately 25%, generated an AI Overview	Coverage inside a separate Conductor search index
Volatility study	More than 274 million U.S. searches, September 2025 to February 2026	23% → 47% → 34%	AI Overview exposure changed sharply inside one research framework

The useful question is not:

How common are AI Overviews nationally?

It is:

How common are they on the searches that mattered to this business?

Search Console Starts the Investigation but Cannot Finish It

Search Console provides query, page, impression, click, click-through-rate, position, country, and device data in its standard Search performance report.

Google’s newer Generative AI performance report is more specific, but still limited. It is currently rolling out to a subset of website owners and reports impressions from AI Overviews and AI Mode. The available dimensions are page, country, date, and device. It does not provide a query dimension or a separate click metric.

That means Search Console cannot currently export a definitive list saying:

These 47 queries displayed an AI Overview, and these 19 lost clicks because of it.

The diagnosis has to combine several forms of evidence.

The diagnostic sequence

1. Export the pages that lost the most clicks.

2. Open each affected page’s query data.

3. Separate genuine position loss from stable-position click loss.

4. Check whether impressions and broader demand changed.

5. Compare against an appropriate seasonal or historical period.

6. Identify the pages and query groups previously associated with qualified leads or assisted customer journeys.

7. Observe or track the current search results for those queries across representative devices, locations, and dates.

8. Record whether each result contains an AI Overview, local pack, ads, featured result, or another element capable of changing click behavior.

A rank tracker that records AI Overview presence can make this faster. Repeated manual observation can still help when the affected query set is small.

Neither method reveals Google’s internal mechanism, and one personalized search is not a measurement system. The objective is to assemble enough evidence to make AI substitution a credible diagnosis, not to manufacture certainty the available tools cannot provide.

Intent Changes Exposure, but No Funnel Stage Is Permanently Safe

Whitespark’s local-search research is useful because it separates queries by intent.

The study checked 540 searches across plumbers, personal injury lawyers, dentists, optometrists, medical clinics, and real estate agents in Houston, Phoenix, and Denver.

Across the entire sample:

• AI Overviews appeared on 68% of queries.

• Local packs appeared on 39% of queries.

The intent-level differences were much larger:

Query type	AI Overview appeared	Local pack appeared
Simple local intent	15%	93%
Informational intent	92%	6%
Hybrid intent	97%	17%

Simple local intent included searches such as “personal injury lawyers in Phoenix.” Informational intent included questions such as “how much do personal injury lawyers charge in Phoenix?” Hybrid intent included searches such as “should I get a lawyer after a car accident in Phoenix?”

Whitespark found no reliable overall correlation between AI Overview and local-pack prevalence across the complete limited dataset. At the intent level, however, it observed a strong inverse pattern: local packs dominated simple local queries, while AI Overviews dominated informational and hybrid queries.

This does not prove that bottom-of-funnel searches are safe.

Hybrid searches can carry substantial commercial intent. A person asking what a furnace replacement costs, whether an accident requires a lawyer, or how a medical procedure works may be close to choosing a provider.

The responsible conclusion is narrower:

Query intent appears to influence AI Overview exposure, but only the business’s current search results and customer evidence can show which commercially valuable searches were affected.

What the Click Studies Actually Establish

Pew Research Center tracked the browsing behavior of 900 U.S. adults across 68,879 Google searches in March 2025.

When an AI summary appeared, users clicked a traditional search result in 8% of visits. When one did not appear, they clicked a result in 15% of visits. Users clicked a source cited inside the AI summary in approximately 1% of visits.

That is meaningful evidence that AI summaries were associated with less outbound clicking in Pew’s observed sample.

It does not prove that every business should expect its organic click-through rate to be cut in half. The study reflects one user group, one period, and one query mix. Position, brand recognition, search intent, summary length, citation design, and competing search features can all affect the result.

Click behavior has also continued to move. Seer Interactive observed organic click-through rate on AI Overview queries in its tracked cohort rising from 1.3% in December 2025 to 2.4% in February 2026. That was a rebound within Seer’s data, not proof that the market reached a permanent floor or that older click levels would return.

The useful planning assumption is simpler:

Historical click-through rates are no longer reliable enough to justify a content budget without current query-level and business-level evidence.

Decide What You Lost Before You Decide What to Cut

A decline in traffic can represent four different economic events.

You lost traffic that produced qualified demand

The page generated consultations, calls, bookings, purchases, or another outcome the business valued.

If its rankings held and the loss aligns with AI Overview exposure, the business has suffered a real acquisition loss. The response may involve improving the page, protecting adjacent queries, testing paid coverage, building demand elsewhere, or reallocating investment toward searches where the user still needs to visit.

You lost traffic that assisted later demand

The page rarely converted on the first visit but appeared earlier in customer journeys.

A prospective customer may have discovered the business through an educational page, returned through branded search, checked reviews or a directory profile, and contacted the company days later.

A last-click report gives the conversion to branded search and makes the page that introduced the business appear worthless.

Attribution is not a perfect reconstruction of human memory. The responsible approach is to triangulate analytics, call tracking, landing pages, CRM information, intake responses, and assisted journeys where the evidence exists.

You lost traffic that supported distribution or commercial pages

A page can earn relevant links, citations, mentions, branded searches, internal-link value, or visibility inside an AI answer without generating a directly tracked lead.

These benefits should not be invented to rescue every article. They should be verified where possible.

A page with fewer clicks may still justify maintenance when it continues to:

• Earn links or references

• Support high-value commercial pages

• Appear in customer research paths

• Produce branded demand

• Establish defensible topic ownership

• Surface in relevant AI results

You lost traffic that was never worth its cost

Some pages generate sessions and little else.

They produce no qualified demand, show no meaningful assisted value, attract no useful links or references, support no commercial page, and require continuing production or maintenance expense.

In that situation, AI Overviews may not have destroyed a valuable asset. They may have exposed an economic weakness that was already there.

Recent Customers Are Evidence, Not a Verdict

Reviewing recent customers can reveal patterns that a traffic report hides. A small sample cannot prove that a page has no value.

The result can be distorted by seasonality, service mix, one referral relationship, cross-device journeys, cookie loss, untagged links, customer memory, and untracked AI exposure.

Use recent customers as one evidence source. For every affected page, review:

• Direct qualified leads

• Assisted customer journeys

• Landing-page conversions

• Relevant links and mentions

• AI citations or other search visibility

• Branded-query relationships

• Internal support for commercial pages

• Production and maintenance cost

• Strategic topic ownership

• Opportunity cost

The page survives when its measurable and strategic value justifies its cost. It is improved, merged, redirected, or stopped when it does not.

What to Do With the Pages AI Overviews Answered

The decision is not limited to “keep” or “delete.”

Keep and maintain

Use this when the page still performs a measurable direct, assisted, distribution, or commercial-support role.

Do not rewrite it merely because an AI Overview exists.

Improve its information value

Use this when the query still matters but the page has become interchangeable with dozens of summaries.

The objective is not to “write for AI.” It is to publish what a generic answer cannot reproduce:

• First-hand experience

• Local procedures and constraints

• Original data

• Named expert judgment

• Real photographs and examples

• Current pricing methodology

• Clear comparisons

• Documented limitations and tradeoffs

Google’s current guidance says its generative Search features use the existing Search index and core ranking and quality systems. It emphasizes useful, non-commodity content rather than a separate AI writing format.

Merge or redirect

Use this when several pages compete for the same intent, repeat the same commodity information, or divide authority without serving distinct user needs.

The decision should be based on page role and query overlap, not an arbitrary word-count threshold.

Stop funding the format

Use this when the business publishes broad educational content by habit and cannot identify direct demand, assisted value, links, citations, commercial support, or strategic ownership after a meaningful review period.

That does not mean informational content is dead. It means this particular publishing model failed its economic test.

Technical Eligibility Must Be Checked Platform by Platform

There is no universal “AI crawler” control governing Google AI Overviews, ChatGPT Search, Perplexity, model training, and every future assistant.

Google AI Overviews and AI Mode

Google’s generative Search features rely on the Google Search index. A page must be indexed, eligible to appear with a snippet, and included in Search generative AI features to be eligible.

Google says the same technical Search requirements continue to matter:

• Googlebot must be able to crawl the page.

• Important content must be accessible and indexable.

• The page must be eligible to appear with a Search snippet.

• The site must be included in generative AI features through Search Console.

Google also says sites do not need a special AI schema, an llms.txt file, artificial content chunking, or an AI-specific writing style. Google can process JavaScript content when it is accessible and not blocked, although JavaScript sites add crawling and rendering complexity.

Google-Extended is not the control for Google Search inclusion. Googlebot governs Search access. Google-Extended controls certain uses in Gemini and Vertex AI and does not affect inclusion or ranking in Google Search.

ChatGPT Search

OpenAI identifies OAI-SearchBot as the crawler that helps public content become discoverable, linked, and cited in ChatGPT Search.

GPTBot is a separate control associated with potential model-training use. Blocking GPTBot should not be treated as the same decision as blocking ChatGPT Search discovery.

Perplexity

Perplexity identifies PerplexityBot as the crawler used to surface and link websites in Perplexity search results.

It separately documents Perplexity-User, which supports user-triggered page visits. These agents have different purposes and should not be collapsed into one crawler decision.

Rendering

Do not assume every search or assistant system renders client-side JavaScript in the same way.

Keep critical business information and core page content available in accessible HTML where practical. Then test the platforms the business actually wants to reach instead of applying one rendering theory to every AI system.

AI-Assisted Local Discovery Is Real, but the Benchmarks Have Boundaries

BrightLocal’s 2026 U.S. consumer research found that 45% of respondents had used an AI tool to find local-business recommendations, up from 6% in its prior survey. The same research reported Google review use falling from 83% to 71%.

That demonstrates rapid adoption among the surveyed consumers. It does not prove that 45% of a particular Nashville company’s prospective customers use AI during their buying journey.

SOCi’s 2026 Local Visibility Index found ChatGPT recommending 1.2% of the locations in its sample, compared with a 35.9% appearance rate in Google’s local three-pack. The study covered more than 350,000 locations across 2,751 multi-location brands.

That demonstrates how selective ChatGPT recommendations were for the multi-location brands SOCi studied. It is not a benchmark for a single-location Nashville plumber, law firm, clinic, or restaurant.

These studies tell a local business what to investigate:

• Are customers using AI assistants in this category?

• Which prompts resemble real buying decisions?

• Is the business named, cited, omitted, or described incorrectly?

• Which sources does the assistant rely on?

• Are those sources accurate and consistent?

• Does AI visibility contribute visits, calls, branded searches, or assisted demand?

They do not provide a universal AI budget.

How Rank Nashville Measures the Loss

No reporting platform captures the complete customer journey.

Search Console does not connect an individual query to a named customer. Analytics loses journeys across devices, privacy controls, apps, and untagged links. Call tracking identifies some interactions, not every influence. Intake recollection is useful but imperfect.

The answer is triangulation.

A Rank Nashville visibility assessment compares:

• Search Console page and query trends

• Position, impression, click, and click-through changes

• Generative AI impression data when available

• Representative SERP observations or tracking

• Landing-page and conversion data

• Call-tracking records

• CRM and intake-source information

• Branded-search changes

• Direct and assisted page value

• The cost of producing and maintaining the affected content

The goal is not to claim that one query conclusively produced one customer.

The goal is to determine:

1. Whether AI substitution is a credible explanation for the loss

2. Which page and query groups were affected

3. Whether those groups produced qualified business value

4. Which pages still perform a defensible role

5. What should be maintained, improved, merged, stopped, or replaced

One business may discover that it lost thousands of sessions and almost no qualified demand. Another may discover that a small set of hybrid searches produced consultations before AI Overviews compressed the clicks.

Those businesses need opposite recommendations.

That is where our visibility assessment begins. We use the available evidence across Search Console, analytics, calls, intake, customer outcomes, and current search results to identify what most plausibly changed, how confident the diagnosis is, and which investment decision the evidence supports.

The Numbers, With Their Boundaries Attached

Evidence	Scope	Finding	Safe use
Conductor monthly analysis	118 million tracked keywords, September 2025	16% generated an AI Overview	Measure coverage inside that tracked-keyword index
Conductor AEO/GEO benchmark	21.9 million unique U.S. searches, September 15 to October 12, 2025	Approximately 25% generated an AI Overview	Measure coverage inside a separate Conductor search index
Conductor volatility study	274 million-plus U.S. searches, September 2025 to February 2026	23% → 47% → 34%	Understand instability inside one research framework
Whitespark local-search study	540 queries, six sectors, three U.S. cities	AIO 68% overall; local pack 39% overall	Form intent hypotheses, not permanent Nashville benchmarks
Pew browsing study	900 U.S. adults, 68,879 Google searches, March 2025	Traditional-result clicks were 8% with an AI summary and 15% without one	Understand observed click behavior in Pew’s sample
Seer CTR tracking	Seer’s tracked query cohort	AIO-query organic CTR rose from 1.3% to 2.4% between December 2025 and February 2026	Avoid treating one observed low point as permanent
BrightLocal consumer research	U.S. consumer panel, 2026	45% reported using AI for local recommendations	Test whether AI-assisted discovery matters to the actual audience
SOCi Local Visibility Index	350,000-plus locations across 2,751 multi-location brands	ChatGPT 1.2%; Google three-pack 35.9%	Understand selectivity inside SOCi’s enterprise sample

Every figure has an expiry date, an inclusion rule, and a boundary.

Your decision should survive after the headline number changes.

Frequently Asked Questions

Should we stop publishing educational content?

Not because traffic fell.

Stop or reduce production when the content cannot demonstrate direct demand, assisted value, meaningful links or citations, strategic topic ownership, commercial support, or another defensible business function relative to its cost.

Continue when the content still contributes to customer acquisition, brand evaluation, commercial pages, references, or qualified visibility.

The unit of judgment is the page and its role, not informational content as a category.

Can we optimize for AI Overviews?

You can improve eligibility and usefulness. You cannot guarantee citation.

For Google, continue doing the work that supports ordinary Search: make the page crawlable and indexable, publish specific and useful information, use clear authorship, maintain accurate business details, and avoid commodity content that adds nothing beyond what already exists.

Google does not require special AI markup, llms.txt, artificial chunking, or a separate AI-writing style.

Does being cited restore the lost traffic?

Not necessarily.

A citation can create visibility and may influence trust, branded search, or later clicks. Citation status can also correlate with stronger rankings, better-known brands, better snippets, and more useful content.

Treat citation as a form of visibility to measure, not a guaranteed replacement for the traffic an AI answer removed.

Are local transactional searches safe?

No query category is permanently safe.

Whitespark’s limited study found much lower AI Overview prevalence on simple local-intent searches than on informational and hybrid searches. It also found very high exposure on hybrid questions with commercial relevance.

Audit the searches that produce customers instead of relying on a national intent label.

Should we block AI crawlers?

Decide platform by platform.

For Google AI Overviews, ordinary Google Search crawling, indexing, snippet eligibility, and the Search Console generative-AI control matter.

For ChatGPT Search, review OAI-SearchBot.

For Perplexity search visibility, review PerplexityBot.

Training controls, Search crawlers, user-triggered fetchers, and Google Search controls are not interchangeable.

Why do AI Overview studies disagree so much?

Because they do not measure the same search universe.

They use different dates, keyword sets, sectors, countries, devices, intent mixes, collection systems, and weighting methods. Even two studies from the same company can report different figures when they use different indexes or collection windows.

Use national research to decide what to test. Use your own query and customer evidence to decide what to fund.

How long will this page remain current?

The individual percentages will age quickly.

The method should last longer:

• Separate exposure from causation

• Separate clicks from business value

• Segment queries by intent

• Respect each study’s sample

• Distinguish platform-specific crawler controls

• Measure direct and assisted content value

• Make budget decisions from your own evidence

The market will keep moving. Your evidence standard should not move with every headline.

Written by Nick Rizkalla, founder of Rank Nashville and a search strategist with more than 14 years of experience across medical, legal, and e-commerce accounts. He is based in West Nashville. Connect on LinkedIn.