AI Visibility Insights — October 8, 2026

1. ChatGPT Is Making Source Decisions Before Search Results Arrive

Finding

Recent research suggests that ChatGPT is increasingly using targeted searches rather than relying exclusively on broad web queries.

Several independent datasets show this direction:

• Peec AI: The share of search fan-out queries containing the site: operator reportedly increased from approximately 0.3% to 23%. The number of retrieved sources also increased from roughly 12 to 24.
• Nectiv: Across approximately 4,000 prompts tested through the API, average fan-out queries per prompt increased from 2.17 to 7.61, with site: appearing in 64% of queries.
• Resoneo: The number of unique cited domains per answer reportedly declined from 19 to 15, suggesting that broader retrieval does not necessarily translate into broader citation diversity.

These studies used different methodologies, environments, and observation periods. Nectiv's sample provides a meaningful basis for identifying patterns, while the exact sample sizes behind some other reported figures are less clear. The numbers should not be treated as measurements from one controlled experiment.

A separate investigation by Suganthan Mohanadasan examined 57 conversations and 3,554 retrieved pages.

In 21 of 27 initial search queries, ChatGPT introduced brand names that users had never mentioned. Brands included in fan-out queries were associated with a reported 68.9% citation rate, compared with 2.1% for pages that were merely fetched.

Those percentages describe different units of analysis and do not establish a causal relationship.

The most revealing observation is simpler: ChatGPT sometimes introduces specific brands into its search queries before receiving the results of those searches.

This is not evidence that ChatGPT maintains a fixed, preselected list of preferred businesses collected through previous crawling. OpenAI operates web crawlers, but crawling, model training, and retrieval during answer generation are different processes.

The experiment demonstrates a behavior during query generation. It does not establish whether the model's prior knowledge, conversation context, or another mechanism determines which brands are selected.

An October 6 observation from Lily Ray adds another dimension: increased domain-restricted searching may make subdomains, including documentation and support centers, more visible as answer sources. This remains a practitioner observation rather than a quantified finding.

My Take

The important distinction is between being discoverable during retrieval and being considered as a candidate before retrieval begins.

Imagine someone asking ChatGPT to recommend an Iranian lawyer in Toronto.

One possible approach is to search broadly, retrieve relevant pages, and identify lawyers from those results.

Another is for ChatGPT to introduce particular law firms, directories, or brands directly into its initial search queries.

Those approaches can produce different discovery opportunities.

This does not mean an unfamiliar business is automatically excluded. A business can still be discovered through broad searches and third-party sources.

However, it suggests that AI visibility may involve an earlier stage than the retrieval and citation processes we typically measure.

Why It Matters

For local businesses, having a crawlable website with accurate information remains important, but it may not fully explain whether a business enters an AI-generated recommendation.

It is useful to distinguish three stages:

1. Candidate and source selection: Which entities or domains does the system decide to investigate?
2. Retrieval: Which documents and pages are actually discovered?
3. Answer generation: Which businesses are mentioned, recommended, or supported with citations?

A business can perform well at one stage without necessarily succeeding at the others.

What to Watch or Test

The most interesting measurement opportunity is whether a brand appears in AI-generated search queries before retrieval.

For a defined set of local service queries, I would examine whether particular businesses or directories repeatedly enter the initial fan-out queries, then compare those observations with actual mentions and recommendations.

This could help distinguish candidate visibility from conventional citation visibility.

However, repetitions alone would not establish why a brand was selected, and any monitoring method must account for differences between the ChatGPT interface, API environments, and available search traces.

Source: Lily Ray — What We Can Learn From Evolving ChatGPT Search Behavior

2. Google's AI Local Pack Shows Why Traditional Rankings Are No Longer Enough

Finding

Google is introducing AI-powered local result formats in parts of the US mobile search experience.

According to Sterling Sky and Places Scout observations, these AI local packs appeared for approximately 7% of tracked keywords in June, rising to around 8% by September 21.

Unlike the traditional Local 3-Pack, which commonly displays three businesses, the AI format often presents only one or two.

More importantly, the businesses selected are not necessarily the same as those appearing in the conventional local results.

Across 322 markets, researchers observed:

The AI format contained approximately 68% fewer unique businesses in this comparison.

This describes the observed sample, not a measured 68% decline in impressions, traffic, or leads for individual businesses.

The new format also sometimes omits prominent call buttons.

Separately, Jepto data covering 179 Google Business Profiles across 34 US law firms showed declining call clicks over a two-year period, while website clicks held up better.

That pattern is worth investigating, but it does not prove that AI local packs caused the decline.

Another finding concerns entity matching. Sterling Sky observed cases where businesses appeared in AI local results without being correctly connected to their Google Business Profiles, leaving them represented by basic map pins without the expected reviews or business details.

At an October 7 conference, Joy Hawkins recommended making business names consistent between websites and Google Business Profiles, including capitalization.

The recommendation is operationally reasonable, but there is not yet published controlled evidence that exact capitalization improves AI local rankings or resolves matching problems.

My Take

What stands out to me is that AI can change the businesses users see even within the same search platform.

Google may return one set of businesses through its traditional Local Pack and another through its AI-powered format for a similar query.

That is significant because we are not comparing Google with ChatGPT or Gemini. We are observing differences within Google's own search experience.

However, different outputs do not necessarily mean completely different search architectures.

Both formats may rely on overlapping local indexes, Google Business Profile information, and other underlying data, while applying different selection or reranking processes.

The evidence establishes a difference in results, not a complete separation of retrieval systems.

My broader interpretation is that as AI becomes more involved in search result presentation, businesses cannot assume that existing visibility automatically carries over to new answer formats.

At the same time, I would not interpret these findings as a reason to abandon traditional Local SEO.

Why It Matters

A business ranking first in the conventional Local 3-Pack may not appear in an AI-powered local result.

This creates two distinct problems:

Visibility: Does the AI format include the business at all?

Entity resolution: If included, does Google correctly associate it with the business's official profile and supporting information?

There is also a potential concentration effect. When a result format displays fewer businesses, fewer positions are available for exposure.

Whether those changes ultimately improve or reduce a particular business's leads remains an empirical question.

For measurement, tracking traditional local ranking positions alone is becoming insufficient.

Local visibility assessments increasingly need to distinguish between conventional rankings, AI inclusion, business recognition, and actual customer actions.

What to Watch or Test

I would prioritize comparing traditional Local Pack and AI Local Pack appearances for the same local queries, locations, and devices.

The questions worth answering are:

• How frequently do the selected businesses differ?
• Are established high-ranking businesses excluded, and which businesses replace them?
• When entity matching fails, does correcting business information resolve the problem?

The last question requires before-and-after evidence rather than treating name consistency as a confirmed ranking factor.

Sources: Sterling Sky — AI Is Quietly Taking Over Local SEO and The State of Local SEO in 2026.

3. Social Media Visibility Is Engine-Dependent, but Citations Are Not Recommendations

Finding

A SE Ranking study examined approximately 100,000 keywords across 20 niches and measured how frequently AI-generated answers linked to social platforms.

The reported results varied considerably.

YouTube was particularly prominent in AI Overviews.

Facebook, YouTube, and Reddit were notable sources within AI Mode, while Reddit and LinkedIn were among the leading social sources in ChatGPT, although social citations overall were uncommon there.

The study also reported that 76.73% of Google's conventional top-ten result sets included a social or community source.

The dataset is substantial, but the available reporting does not establish all relevant sampling details, including geographic distribution and collection dates.

More importantly, the research measured links to social platforms, not whether social activity caused a business to receive more AI recommendations.

My Take

This finding reinforces something that has become increasingly clear across recent GEO research: different AI engines do not necessarily use the same source categories in the same proportions.

A platform that frequently appears in Google's AI-generated answers may play a much smaller role in ChatGPT.

That makes broad recommendations such as "publish more on social media to improve AI visibility" too imprecise.

But I also see a longer-term strategic problem with optimizing exclusively around today's engine-specific percentages.

If ChatGPT currently cites relatively few social sources, that does not mean it will continue doing so.

Search architectures, source selection, product interfaces, and user behavior can all change.

A strategy built entirely around one platform's current retrieval behavior may become fragile as those conditions evolve.

Why It Matters

We need to distinguish source citation from business visibility.

An AI answer citing a YouTube video is not necessarily recommending the company that published it.

Likewise, low social citation frequency in ChatGPT does not prove that social presence has no indirect relationship with brand discovery or recognition.

For local businesses, the practical question is not simply which platforms receive more citations.

It is whether visibility through a particular source translates into relevant brand mentions, recommendations, and customer discovery.

That relationship needs to be measured separately for each AI engine.

What to Watch or Test

A useful experiment would compare the performance of official websites, relevant third-party profiles, and social content across the same business-specific query set.

I would measure source citations and business recommendations independently rather than assuming that one produces the other.

Sources: SE Ranking study coverage — Social Media Today via Yahoo and SE Ranking.

4. My Broader Conclusion: Durable AI Visibility Requires a Diversified Digital Presence

Looking at these findings together, I keep returning to one strategic conclusion.

We are still learning how different AI engines discover entities, retrieve information, select sources, and generate recommendations.

Research continues to reveal meaningful differences across engines, industries, queries, and result formats.

Even detailed measurements of today's behavior cannot guarantee that the same source preferences will remain stable.

That makes me skeptical of strategies built around a single citation source, platform, or current algorithmic pattern.

I see three complementary foundations for a more durable AI visibility strategy.

1. Owned Presence

A business needs an accurate, accessible, well-structured official website.

This is where it can explain its services, expertise, locations, pricing, and other important attributes directly.

The objective is not simply to add structured data. It is to make authoritative business information easy for both people and retrieval systems to understand.

2. Third-Party Presence

A business should establish a credible presence across relevant external sources: professional directories, review platforms, independent publications, comparisons, and industry-specific resources.

Not every external mention is earned media. Some profiles are business-managed, while others reflect independent editorial coverage or customer contributions.

These distinctions matter, as do relevance, accuracy, and source credibility.

3. Social and Community Presence

A business should participate in social platforms and communities that genuinely matter to its customers.

YouTube, Reddit, Facebook, and LinkedIn serve different purposes, and their relevance varies by industry.

The objective should be meaningful, consistent participation and useful content rather than publishing everywhere simply because a platform occasionally receives AI citations.

Diversified Presence, Selective Investment

I do not think these three foundations should receive equal investment.

A law firm may benefit more from professional directories and strong explanatory content, while a restaurant may have greater opportunities through reviews, community discussions, and visual content.

The appropriate balance depends on the industry, location, customer behavior, and evidence from AI visibility measurement.

My preferred approach is therefore:

Build a credible baseline across all three foundations, then allocate greater effort where relevance and measured opportunity justify it.

This is not a claim that diversified presence guarantees inclusion in AI-generated answers.

It is a way to reduce dependence on any single platform or source-selection pattern while preserving the ability to adapt.

There is one more principle connecting all three foundations: entity consistency.

The official website, business profiles, independent mentions, and social accounts should communicate a coherent identity.

Consistent business names, services, locations, and other identifying details can make the digital footprint clearer, although the precise effects on AI recommendations are not yet established.

Finally, these foundations need to be evaluated through outcomes, not merely existence.

I would distinguish:

• Presence: Is the business represented accurately across relevant sources?
• Retrieval and citation: Can AI systems find those sources, and which ones do they reference?
• Mention and recommendation: Does the business actually appear in answers, and is it recommended when relevant?

These are related but different measurements.

The Direction I Want to Follow

I see AI visibility becoming less about securing a particular ranking position and more about building a recognizable, trustworthy, accessible entity across the information environments AI systems use.

Traditional SEO remains part of that foundation. But as search interfaces evolve, ranking performance alone cannot describe the full picture.

The long-term objective should not be to predict which source ChatGPT, Gemini, or Google AI Mode will favor next.

It should be to build a strong enough digital presence to remain discoverable across changing systems, while continually measuring where that presence actually translates into visibility.

The strategy should be diversified. The investment should be selective. And the measurement should remain engine-specific.