The Convergence

The view from above — and how it's written

Most research on AI search reports one study at a time. This report does the opposite. It gathers every bucket of evidence we could find on how local businesses get recommended by AI, lays them on one table, and steps back until the whole picture resolves. From that altitude, the disconnected facts collapse into a single conclusion and four currents that run beneath all of them.

Because the goal is to be useful to a business owner and defensible to a skeptic, every section is written twice. In Plain Terms states the point in language anyone can act on. The Proof then establishes it rigorously — naming the specific researcher, study, court record, or dataset behind each claim, with the chart that shows it.

How every claim is labelled

• CITED A figure from a named, linkable source. Real or omitted — never invented.
• ANALYSIS Our own conclusion from reading the sources together.
• PROJECTION A forward inference from a measured trend. Flagged, never disguised as data.

In Plain Terms

We pulled together every kind of evidence we could find about how AI decides which local businesses to recommend — what shoppers do, what the data shows, what Google's own people and documents say, what scientists have proven about how these systems retrieve information, and what the law now requires. They come from totally different places and were never meant to be read together. But when you put them side by side, they all say the same thing: AI doesn't recommend a website, a profile, or whoever is closest. It recommends a business it can recognize as real — one that many trustworthy sources independently describe the same way. Being that recognizable, verifiable business is the whole game.

The Proof

The diagram below places eleven independent research streams around a single center. Each was produced by different people, with different methods, for different reasons. Read individually they are a pile of disconnected facts. Read together they converge — every one of them describes, from its own angle, the same underlying object: the verifiable entity.

Figure I · The convergence. Eleven independent streams; each source named below and in References.

Here is who and what each stream is, and what it independently establishes CITED — the named evidence behind the picture:

BrightLocal. Local Consumer Review Survey 2026. 45% of consumers now use AI for local discovery (up from 6%), and 88% fact-check the result — so people seek, then verify, a recommendation.¹

Glenn Allsopp · Ahrefs. 75,000-brand visibility study. The top correlate of AI visibility is YouTube and branded web mentions (0.737), not links or pages — the signal is off-site, distributed conversation.²

Ahrefs · 146M SERPs. AI Overview trigger study. Specific, conversational, entity-rich questions trigger AI answers far more than bare keywords (57.9% vs 9.5%).³

Local Falcon. ~60,000 simulations. Distance from the searcher correlates with AI rank at r = 0.001 — proximity, the old master variable, is gone; what remains is authority.⁵

Seer Interactive. Citation analysis. The AI assembles answers by quoting other sources about a business; ChatGPT matches Bing's results 87% of the time.⁶

Sensor Tower. State of AI 2026. The assistant market is fragmenting (ChatGPT 46%, Gemini 28%), yet every assistant grounds answers in retrieved, attested sources the same way.⁷

Darren Shaw · Whitespark. Local Search Ranking Factors 2026. Even the classic local pack is shifting weight toward reviews — third-party attestation — now 16% and rising.⁸

Google + Judge Mehta. Search Central docs; U.S. v. Google. Robby Stein describes "query fan-out"; court filings disclose FastSearch/RankEmbed — semantic retrieval of entities, not link-counting.⁹,¹²,¹⁷

Retrieval science. Peer-reviewed literature. Aggarwal (Princeton), Dai, Fayyaz, Yamada (LUKE): retrievers reward prominent, well-grounded, verifiable entities.¹⁹,²⁵,²⁶,³⁰

FTC + European Commission. Consumer Review Rule; EU AI Act. Fabricated attestation is now illegal ($53,088/violation) — only genuine signals are durable.³⁵,³⁸

Field reports. Owners & local SEOs. Businesses surface in AI without buying placement, and mentions lift then decay (Andrew Shotland) — exactly as a click-trained retriever predicts.⁴¹

Eleven instruments, eleven methods, one reading: AI local visibility is won as a verifiable entity, attested consistently across independent sources.

That single substrate is the bird's-eye finding. Everything that follows — the four currents — is simply what this convergence does to a local business, and the one thing the whole landscape is still missing.

It's no longer about what you publish

In Plain Terms

For twenty years, "doing your SEO" meant working on things you owned: your website, your pages, your Google profile. That era is ending. The things that now make AI recommend you mostly live off your own property — what other people say about you on YouTube, in reviews, in community threads, in local round-ups. You've gone from building your storefront to earning a reputation you don't fully control. The job is no longer to publish more about yourself; it's to be talked about, consistently, by others.

The Proof

Glenn Allsopp's Ahrefs analysis of 75,000 brands ranked every candidate signal by how strongly it correlates with AI visibility. The signals describing what others say about you — YouTube mentions (0.737), branded web mentions (0.66–0.71) — sit at the top. The signals describing what you publish about yourself — your domain rating (0.266), the backlinks you build (0.218), the pages you publish (0.194) — sit at the bottom.² The gap is not subtle.

Figure II · Off-property signals win. Ahrefs, 75,000 brands (Glenn Allsopp), 2026.

The mechanism, disclosed in Judge Amit Mehta's November 2025 ruling in U.S. v. Google, explains why. Google grounds its Gemini answers through a retriever called FastSearch, built on a deep-learning model, RankEmbed, that matches queries to documents semantically rather than by counting links.¹⁷ Google's own VP of Search, Robby Stein, describes the front end as "query fan-out" — the model issues dozens of background searches and synthesizes what it retrieves.¹² Neither process rewards a business for publishing more pages about itself; both reward a business that the wider web independently corroborates. ANALYSIS The practical relocation is total: the battlefield moved from your domain to the ecosystem's consensus about you.

Generic loses; specific wins — everywhere

In Plain Terms

The single most repeated pattern across all the evidence is this: the more specific you are, the more AI rewards you — and it shows up at every level. Specific questions get AI answers more often than vague searches. First-hand, specific content beats generic listicles. A review that names exactly what you did ("great with anxious dogs") is worth more than a bare five stars. And a business known for one clear thing beats one that claims to do everything. Being known for the right thing isn't a tagline — it's the rule the whole system runs on.

The Proof

The same generic-to-specific gradient appears independently at four layers. At the query layer, Ahrefs' study of 146 million SERPs shows single-word searches trigger an AI Overview 9.5% of the time, while full question-style queries trigger 57.9%.³ At the content layer, Google's own Search Central guidance explicitly rewards "non-commodity content" and first-hand experience over generic aggregation.¹⁰ At the review layer, AI answers quote the specific language of reviews verbatim — the named service, not the star count. At the entity layer, the retrieval-science consensus (LUKE-style entity grounding) favors precisely-described, unambiguous entities.³⁰

Figure III · The specificity ladder. Query: Ahrefs. Content: Google. Review/entity: platform mechanics & LAIR analysis.

This is the academic and mechanical basis for Local AI Registry's founding principle — alignment, not coverage. ANALYSIS A business fails in AI not by being invisible but by being known for the wrong, generic thing. The corrective is specificity, applied deliberately at all four layers at once.

Four unrelated timers strike at once

In Plain Terms

Four completely separate things are all changing at the same moment — roughly 2026 into 2027. How many people use AI for local search. How open AI is to recommending smaller businesses. How people phrase their searches. And when the new AI transparency laws kick in. None of these has anything to do with the others — which is exactly why their lining up matters. When four unrelated clocks all strike the same hour, that's not coincidence; it's a transition. The businesses that move during the window get an advantage the latecomers won't.

The Proof

Four independently-measured trends converge on the same window. Adoption: BrightLocal records consumer use of AI for local discovery rising from 6% to 45% across 2025–26.¹ The citation window: Ahrefs finds the share of AI-Overview citations drawn from the organic top ten fell from ~76% to ~38%, widening the pool to the long tail — a window the retrieval-bias literature predicts will re-narrow.⁴,²⁵ The exposure runway: as phrasing turns conversational, local AI-trigger rates climb off their 7.9% floor toward the 57.9% question rate.³ Regulation: the EU AI Act's Article 50 transparency duties take effect 2 August 2026.³⁸

Figure IV · Four clocks, one transition. BrightLocal; Ahrefs; European Commission. Convergence framing is analysis. PROJECTION

The convergence itself is the argument. ANALYSIS Any one trend could be dismissed as noise; four unrelated research streams agreeing on a transition date is a structural signal that the ground is shifting now, and that early, consistent action dominates a wait-and-see posture.

The Keystone · No one can tell you what the AI says

In Plain Terms

Here's the strangest part. Everything above says your AI reputation now matters enormously — but nobody can actually tell you what it is. Ask the same question twice and the AI gives different answers. The platforms don't report which businesses they recommend or why. There's no record that follows you across ChatGPT, Gemini, and Google. And the law polices AI that lies to customers, not AI that's wrong about you. So the one thing that now decides whether customers find you is the one thing no one is keeping score of. That gap is the whole reason a registry needs to exist.

The Proof

Four properties you would expect of any system of record are absent across every major AI surface. Reproducibility: practitioners document that AI Overview results "aren't repeatable" run-to-run, and Google's own Search Console does not separate AI-driven clicks. Accuracy: Nelson Liu and Percy Liang (Stanford) found only 51.5% of generated sentences are fully supported by their citations, and the Tow Center at Columbia found leading AI search tools wrong on more than 60% of queries.³²,³⁴ Cross-platform record: none exists. Regulatory scope: the FTC and EU AI Act target deception of consumers, not accuracy about third-party businesses.³⁵,³⁸

Figure V · The accountability vacuum. LAIR analysis of platform reporting, reproducibility & regulatory scope, 2026.

This is the keystone the other three currents rest on. ANALYSIS The entire body of evidence converges on "be a verifiable entity," moving fast, on a closing clock — while no neutral party measures whether you are one, reproducibly, across platforms, over time. That negative space, visible only when all eleven streams are held in one frame, is not a gap in the research. It is a description of an institution that does not yet exist.

If the evidence converges on a substrate no one is measuring, the implied object is not a tactic or a tool — it is an institution: an independent, reproducible record of what AI says about local businesses, and whether they are known for the right things.

That is the function Local AI Registry performs. It scores independent local businesses across ChatGPT, Gemini, Perplexity, and Google AI Overviews; it measures not just whether a business is named but whether it is named for the right service — the alignment this report shows the whole system rewards; and it does so repeatedly, across platforms, over time, which is exactly the reproducibility the landscape lacks. The registry is not a product appended to the research. It is the instrument the convergence implies.

From reading to record

Seven findings underneath the view

The bird's-eye view is built on a deeper layer of cross-source findings — each one a conclusion drawn by overlaying two or three datasets that none of them states alone. They are summarized here as the supporting evidence beneath the four currents; each is documented at length in companion analysis.

The seven triangulated findings beneath the convergence

• 01 · The Broken Scoreboard. The canonical local-pack model and AI visibility are nearly orthogonal. Built from: Ahrefs × Whitespark × antitrust.
• 02 · The Exposure Runway. The 7.9% local AI rate is a countdown toward the 57.9% question rate. Built from: Ahrefs × BrightLocal.
• 03 · Re-Centralization. Proximity's death plus prominence bias tilts AI toward incumbents. Built from: Local Falcon × retrieval-bias research.
• 04 · The Closing Window. A temporary long-tail citation opening that bias predicts will re-narrow. Built from: Ahrefs × field experiment × academia.
• 05 · Reviews Are the Keystone. The only lever that is ranking signal, citation source, safeguard & legal moat at once. Built from: Whitespark × mechanism × Liu × FTC.
• 06 · The Verification Dividend. The fact-check, not the recommendation, is where the funnel breaks. Built from: BrightLocal × Tow Center.
• 07 · Two Disjoint Games. ChatGPT-via-Bing visibility is the underpriced channel. Built from: Seer × Ahrefs × Sensor Tower.

Each finding is graded ANALYSIS — a strong hypothesis from real, cited inputs, not a measured fact. Several become measurable original statistics when cross-checked against the live index; those measurements will be published when complete.

See where you actually stand

This report describes the landscape from above. The next step is the instrument it implies — specific to you, whether you run a business that needs to know what AI says about it, or you're a customer who wants the truth about who's actually good.

For business owners

Register your business — free. Find out, in under a minute, what ChatGPT, Gemini, Perplexity, and Google AI Overviews currently say about you — what you're named for, what you're invisible for, and your AI Score across every service you offer. Registering and claiming your listing is free.

Get your AI Score →

For consumers

Check what a business is known for. Before you book, see what the AI really says — what a local business is genuinely known for, where it's strong, and where the reputation doesn't match the marketing. An honest, editorial read, never pay-to-rank.

Look up a business →

Local AI Registry is the independent, editorial registry of how AI sees local business. We never sell rankings.

How this report was built

This is a work of synthesis at altitude: eleven independent evidence streams, read together to surface the conclusions that only their overlap reveals. Every input is a real, cited figure from a named source; the convergence and the four currents are our own ANALYSIS; forward inferences are marked PROJECTION. We hold a strict "real or omitted" rule — no quote, statistic, study, court finding, or researcher is invented, and unverifiable figures (such as a widely-shared "797,444 impressions, 7 clicks" anecdote) are excluded rather than repeated.

What this method can and cannot claim

Convergence is strong evidence, not proof of mechanism. Many independent sources agreeing on a substrate is a powerful signal, but several inputs are correlational and inherit that limit. Projections are reasoned, not certain — the transition window and the runway are extrapolations from measured trends, flagged as such. The local-AI research gap is real: little peer-reviewed work studies local-business inclusion in AI answers directly, so local conclusions are careful inferences from adjacent retrieval, bias, and entity-grounding research. Vendor incentive exists in some sources; figures were cross-checked where possible. The accountability vacuum is, by definition, where measurement is thinnest — which is precisely the gap the registry is built to close, and why proprietary figures are shown as forthcoming rather than asserted.

Local AI Registry scores independent local businesses on whether AI assistants recommend them — and, just as importantly, on whether they're recommended for the right things. We're editorial, transparent, and never pay-to-rank. Yellow Pages told you who existed. We tell you what the AI actually says.

Free to register. Free to be found correctly.

Local AI Registry Research. The Convergence: What Eleven Streams of Evidence Reveal About How Local Businesses Win AI Recommendations. White Paper No. 03 (Definitive Edition), June 2026. localairegistry.com/research.