From Search Box to Answer Engine: What the Shift Means for Local Businesses

For twenty years, the canonical act of local discovery was a search query followed by a list of ten blue links. A consumer typed a question; a search engine returned a ranked set of web pages; the consumer clicked, compared, and chose. The mental model was familiar enough that most local business owners could reason about it intuitively. Rank higher, get seen more. The mechanics were opaque, but the objective was clear.

That model is being displaced. Not suddenly, and not completely — but with a directional clarity that makes the trajectory difficult to argue with. AI-powered answer engines have begun to answer the question directly rather than returning a list of places to look. When they do, they typically name one, two, or three businesses. The rest of the local market is not ranked lower. It is simply absent from the answer.

This piece examines what that shift means structurally, what determines which businesses get named, and why the signals that govern AI citability bear a striking resemblance to the reputation hygiene that local businesses should already be investing in.

A note on epistemic honesty before we begin: AEO — answer engine optimisation — is a young and rapidly evolving discipline. Some of what is written here rests on established, published fact. Some of it is directional inference drawn from how these systems work at a technical level. Where the distinction matters, we say so explicitly. The reader should treat the sections marked as inference as informed analysis, not settled science.

I. The channel shift — what is actually changing

Google launched AI Overviews — the AI-generated summary block that appears at the top of many search results pages — to general availability in the United States in May 2024, following a year of testing under the name "Search Generative Experience."¹ By mid-2024, AI Overviews were appearing on a significant proportion of US search queries, with rollout expanding to the UK, India, Japan, and other markets through late 2024 and 2025.² These are documented facts.

What the numbers mean for local business discovery is less precisely settled, because Google has not published category-level data on how often local queries trigger AI Overviews versus traditional Local Pack results. Independent research and industry monitoring suggest that informational and navigational queries ("best dentist in [city]," "where to get a massage near me") are more likely to trigger AI-generated summaries than transactional queries with clear commercial intent, but the pattern is not consistent across query types or geographies.³

Perplexity — an AI-native search engine that generates cited, conversational answers from web sources — has reported reaching 100 million monthly active users by early 2025, with a substantial share of queries being local in nature.⁴ ChatGPT's web-browsing capability, available to paying subscribers, handles a growing volume of what would previously have been Google queries. Apple Intelligence, which began surfacing in iOS 18 in late 2024, routes some Siri queries through AI synthesis rather than returning a traditional web search. The aggregate share of local discovery that flows through AI-mediated channels is not precisely quantifiable from public data — but it is growing, and across multiple platforms simultaneously.

The scale of this shift is now measurable in consumer behaviour. The BrightLocal Local Consumer Review Survey 2026 (n=1,002) found that consumer use of AI for local business recommendations rose from 6% in 2025 to 45% in 2026, making AI the third most-used local discovery channel — behind Google and Facebook, but ahead of every other digital surface.⁷ The query-level picture is similarly significant: an independent audit of 540 queries across three cities and six verticals by Whitespark (May 2025) found that 68% of local searches now surface a Google AI Overview — though the figure varies sharply by intent type, from 15% for transactional "near me" queries to 92–97% for informational and cost-comparison queries.⁸

The practical consequence for a local business is this: there is now a meaningful and growing class of potential customers who, when they look for a service, will receive an AI-generated answer that either names the business or does not. The traditional search result, even at rank one in the Local Pack, may not appear in that answer at all.

The competitive stakes of that binary outcome are stark. The SOCi 2026 Local Visibility Index — which analysed over 350,000 locations across 2,751 brands — found that ChatGPT recommends only 1.2% of local business locations, compared with 35.9% appearing in Google's Local 3-Pack; only 45% of brands winning traditional local search also win AI recommendations, meaning the two surfaces are selecting largely different winners.⁹ The traffic economics reinforce the urgency: research by Ahrefs (February 2026, 300,000 keywords) found that AI Overviews cut the click-through rate to the top organic result by 58% — but businesses actually cited within an AI answer see organic clicks rise 35% and paid clicks rise 91%, because the citation functions as an endorsement rather than a bypass.¹⁰

II. What makes a business citable by an AI system

To understand what governs AI citability, it helps to understand, at a structural level, how these systems retrieve and use local business information. The three dominant AI discovery surfaces — Google AI Overviews, Perplexity, and ChatGPT Browse — each work differently in detail, but share a common logical architecture.

All three retrieve information from external sources — either the live web, indexed web content, or structured data sources — and synthesise it into a response. None of them operates purely from a static knowledge base when answering local queries, because local business information (hours, reviews, services, locations) changes frequently. For a query like "best physiotherapist in Singapore," a modern AI search system will retrieve and synthesise data from sources that include Google Business Profiles, review platforms (Google Reviews, Yelp, Trustpilot), the business's own website, and — in Google's case — structured data within its Knowledge Graph.

From this architecture, the selection logic becomes clearer. A business is cited when:

First, it is retrievable — meaning its information exists in the sources the AI system queries, is crawlable, and is not blocked by robots.txt or other access restrictions. A business with no website, a sparse Google Business Profile, and no indexed reviews is structurally invisible to a system that can only cite what it can retrieve.

Second, it is extractable — meaning the information about it is specific and structured enough for the AI to pull out concrete facts. A profile that says "we offer high-quality services in a friendly environment" gives an AI system nothing to extract. A profile that says "specialist in deep-tissue massage, serving Notting Hill and surrounding areas, open Monday to Saturday, 4.7 stars from 186 Google reviews" gives the system multiple extractable facts with which to construct a recommendation.

Third, it is trustworthy in the system's terms — meaning the signals that the AI uses as proxies for quality and relevance are favourable. This is the least precisely documented part of the selection logic, and the claims that follow should be understood as directional inference rather than established fact.

What we can reason about, from the way large language models work and from the limited public guidance that AI search platforms have published: review volume, recency, and consistency are likely strong citability signals, because they are the clearest machine-readable proxy for genuine customer experience. A business with 200 recent reviews that consistently mention specific services is providing a language model with high-quality training signal for what that business does and how well it does it. A business with 12 reviews, three years old and generically positive, provides much weaker signal.

The data now supports a sharper version of this inference. The SOCi 2026 Local Visibility Index identified effective rating floors for AI citation: ChatGPT tends to recommend businesses rated 4.3★ or above; Perplexity 4.1★; Gemini 3.9★.⁹ Review freshness compounds the floor: BrightLocal's 2026 survey found that 74% of consumers only trust reviews from the last three months, and 32% only those from the last two weeks — a consumer expectation that likely shapes which reviews AI systems weight most heavily when synthesising recommendations.⁷ Engine-level sourcing diverges further than most practitioners expect: a multi-million-citation analysis by Yext found that ChatGPT draws approximately 49% of its local citations from third-party directories such as Yelp, while Gemini draws approximately 52% from brand-owned sites — with only 11–25% citation overlap between engines.¹¹ A business that is well-represented on directories but has a thin website, or vice versa, is therefore citable on some AI surfaces and invisible on others.

Schema markup — structured data embedded in a website's HTML that labels entity types, opening hours, services, and geographic area — makes business information unambiguously machine-readable in a way that removes the need for inference. Google's Search Central documentation explicitly notes that structured data helps search systems understand page content;⁵ by extension, it reduces the risk of AI systems misattributing or omitting a business's information. That said, the evidence for schema as an independent AI-citation driver is more modest than vendor commentary often implies: a controlled difference-in-differences study by Ahrefs (1,885 pages) found that schema markup moved AI citation rates by only −4.6% to +2.2% — a range that overlaps zero.¹² Schema remains best practice for information accuracy, but the primary citation drivers — rank, reviews, and directory presence — appear to carry considerably more weight.

E-E-A-T (Experience, Expertise, Authoritativeness, Trustworthiness) — Google's published quality framework for evaluating web content — is not specific to AI Overviews, but it shapes the pool of content from which AI Overviews are drawn. A business whose website demonstrates genuine domain expertise and whose online presence is consistent and corroborated across multiple platforms starts the citability competition ahead of one that has only a sparse business listing.

III. The local business gap — why most profiles are not structured for extraction

The standards described above — structured, consistent, recent, specific — sound straightforward. In practice, the gap between what AI systems need to extract reliable recommendations and what the typical local business profile actually provides is substantial.

Most local business profiles, when examined honestly, fail the extractability test. A typical mid-market business in any sector — a physiotherapy clinic in Dallas, a hair salon in Singapore, a solicitor's practice in Manchester — has a Google Business Profile with basic information filled in, fewer than forty reviews posted over several years, no schema markup on its website beyond what a generic web builder adds automatically, and website copy that describes its qualities rather than its specifics. This is not negligence; it is the natural equilibrium for a business whose owner is focused on delivering the service rather than architecting the digital record of it.

The gap matters more now than it did two years ago because the consequence of a thin profile has changed. Under the old model, a thin profile meant lower rank in a list. Under the AI answer model, it increasingly means absence from an answer entirely. The floor has risen.

IV. What citability means in practice — the overlap with reputation best practice

Here is the point where the analysis becomes practically useful rather than merely alarming. The signals that appear to govern AI citability are not new requirements that demand entirely new behaviour. They are, in large part, the same signals that a well-managed local business presence has always sought to generate — and that most businesses are under-producing not because they do not understand their value, but because they lack the systems to produce them consistently.

Review volume and recency

A language model synthesising a local recommendation has, in practice, more evidence to work with about a business that has 150 reviews than one that has 15. More reviews means more specific factual claims that can be extracted: the service types mentioned, the staff names that recur, the location descriptors, the quality indicators. Review recency matters because language models — like human readers — weight recent signal more heavily than stale signal when assessing current quality. A business with 15 reviews published in the last six months is providing fresher, more reliable signal than one with 80 reviews, the most recent of which is two years old.

Consumer expectations reinforce the machine logic. BrightLocal's 2026 survey found that 74% of consumers only trust reviews from the last three months, and 32% exclusively from the last two weeks — making review velocity as important as review volume for any business that wants its reputation to read as current.⁷ Organic search rank compounds the effect: an Authoritas study found that a business in the #1 organic position has a 53% chance of being cited in an AI Overview, compared with 37% for the #10 position — confirming that the traditional signals of search authority are also the primary predictors of AI inclusion.¹³

This is directional inference grounded in how retrieval-augmented generation systems work, not a published specification from Google or Perplexity. But it is well-supported inference, and it aligns with what local SEO practitioners are observing empirically as they monitor which businesses appear in AI Overviews across different categories.⁶

Profile accuracy and completeness

An AI system that retrieves information about a local business and finds conflicting data — different phone numbers on the website and the Google Business Profile, incorrect opening hours, a category listing that does not match the services described on the website — has been given noise rather than signal. The system must either synthesise a best guess (which may be wrong) or deprioritise the business in favour of one about which it has cleaner information.

NAP consistency — the alignment of Name, Address, and Phone number across all online citations — has been a local SEO best practice for years, precisely because it reduces signal noise. Its importance for AI citability is, if anything, higher, because AI systems are synthesising rather than simply ranking: an inconsistency that might merely affect a traditional Local Pack result can cause an AI system to attribute incorrect information or to avoid citing the business entirely.

Schema markup

LocalBusiness schema — the structured data format that identifies a web page as describing a specific type of business, its services, location, hours, and price range — is the most direct way to make a business's information machine-readable in a format that AI systems can parse without inference. Google's own published guidance on structured data explicitly notes that it helps Google understand page content and can enable "rich results" in search.⁵ The extension to AI Overviews is not stated in Google's documentation, but it is the logical consequence: a system that needs to cite a physiotherapy clinic in Chicago is better equipped to do so accurately if the clinic's website contains a structured data block that labels it unambiguously as a physiotherapy clinic in Chicago.

Content specificity

A business website that articulates specific services, named practitioners, real patient outcomes (where privacy permits), and distinctive features of the practice is providing language models with extractable claims. A website that describes the business in generic terms — "passionate about your wellbeing," "a team of experienced professionals" — is providing marketing copy that does not survive the extraction process. The shift in web writing that AEO demands is a shift from adjective-led prose toward noun-and-verb specificity: not "exceptional care" but "root canal treatment, composite bonding, and Invisalign, available in evenings from Tuesday to Thursday."

V. How to think about AEO investment today

It would be irresponsible to prescribe a precise AEO playbook at this stage of the technology's development. The platforms are moving fast, their selection logic is only partially documented, and the data on what works at a local business level is still thin. What can be said with confidence is the following.

The actions that improve AI citability — building a strong, recent, consistent review profile; maintaining accurate business information across all online surfaces; implementing schema markup; and writing website content that is specific and extractable — are the same actions that improve performance in traditional local search. There is no meaningful tradeoff between optimising for Google's Local Pack and optimising for AI Overviews. A business that does neither is doubly disadvantaged; a business that does both is protected against the channel shift rather than exposed to it.

The businesses that will be most harmed by the shift to AI-mediated local discovery are not those that never ranked well — they are those that ranked adequately under the old model through a combination of proximity and light optimisation, and who will find that "adequate" is no longer sufficient for inclusion in an answer that names only two businesses. The bar has moved upward, and the businesses that did the minimum necessary to appear in a ten-result list may not clear the bar required to appear in a two-business recommendation.

For a dental practice in Toronto, a spa in Singapore, or a personal trainer in London, the practical implication is not abstract. These are competitive, high-intent local markets where AI Overviews are already appearing on discovery queries. The window to build the review profile, the schema markup, and the content specificity that makes a business citable is open now. The businesses that build it will benefit as the channel grows; the businesses that wait will be optimising against a gap that has already widened.

A note on what we do not yet know

This piece has tried to be careful about the distinction between established fact and directional inference, but it is worth being explicit about the limits of the current evidence base.

We do not know the precise weighting that Google's AI Overviews system gives to review signals versus schema versus website content versus other factors. Google has not published this specification, and reverse-engineering it from observed outcomes is difficult because the system is multivariate and continuously updated.

We do not know how the local AI discovery landscape will look in two years. The platforms competing for this surface are numerous (Google, Perplexity, ChatGPT, Apple Intelligence, Bing Copilot), their market shares are in flux, and their approaches to local business citation differ in ways that may matter for optimisation strategy.

What we do know is that the directional shift is real, that multiple major AI platforms are investing heavily in local business data, and that the selection logic — wherever it is set — rewards the same structured, consistent, high-evidence presence that good reputation management has always produced. That convergence is not coincidental. It reflects a deeper truth: the signals that help humans trust a local business are, at their core, the same signals that help machines extract and cite one.