How Gemini Cites Local Businesses Through the Vertex Grounding Wrapper

Two Burlington heating companies. Same week, same question typed into Gemini: "who are the best HVAC companies in Burlington, Ontario." The first company, Laird & Son Heating & Air Conditioning, is named in the answer, three sentences of praise, family-owned for seventy years, non-pushy technicians. The second company, two kilometres up the road, is nowhere in the response and has no idea why. Its owner opens Gemini, sees a tidy little Sources panel under the answer, clicks it expecting to learn which competitor pages Gemini read, and lands on a string of links that all begin with the same baffling address: vertexaisearch.cloud.google.com/grounding-api-redirect/. Not Laird's site. Not a directory. A Google redirect that resolves to who-knows-where. The first owner got a recommendation they cannot fully explain. The second got an absence they cannot diagnose. Both are staring at the same wall, and the wall has a name.

That wall is the subject of this analysis. This page is a data study of the grounding wrapper itself, what it is and what it does to measurement, not a how-to for optimising your way into Gemini; the practical playbook lives separately in our guide to earning visibility inside Gemini's grounded answers. Most guides on getting cited in Gemini treat it as a transparent engine you optimise the way you optimise Perplexity: find the URLs it pulls, get listed there, watch the citations roll in. Our first-party data says that frame is structurally wrong for Gemini specifically. Gemini does not hand you its sources the way the other engines do. It hands you a wrapper. Understanding that wrapper, and what it does to measurement, is the difference between tracking Gemini honestly and fooling yourself with a metric that cannot mean what you think it means.

The vertexaisearch wrapper is the single most-cited domain in our entire dataset

Gemini returns nearly everything through one grounding layer, and that layer is the most-cited domain we recorded anywhere. Across nine Ontario cities and five service verticals, vertexaisearch.cloud.google.com was cited 384 times by Gemini. To put that in proportion: no individual publisher, directory, or business website in the whole 1,732-citation scrape came close. The wrapper out-cited every real source on the open web because it is the conduit every real source passes through before it reaches you.

This is the key behaviour that separates Gemini from the other three engines we study. ChatGPT leans on google.com and hands you the underlying Maps and Knowledge Graph data. Perplexity spreads across review aggregators like homestars.com and opencare.com and footnotes each one by name. Anthropic's Claude reaches for curated directories and tells you which directory. Gemini, alone among them, interposes a redirect between the answer and the source, so the thing you can see and count is not the source at all. It is the pipe.

Read that fourth line again, because it is the one that matters. The wrapper is not unusual in volume alone. It is unusual in that it is the only thing Gemini shows you. Every named source sits behind it. That single design choice is what reshapes how a local business has to think about Gemini visibility, and it is why this page exists separately from our wider study of why each AI engine names a different set of local businesses for the same Ontario query.

What the grounding wrapper actually is, and why Gemini routes everything through it

The wrapper is Vertex AI Search doing its job, and the job is grounding. When Gemini answers a question that needs current, real-world facts, it does not generate the answer purely from its training weights. It runs a Google Search grounding step: it issues search queries, retrieves passages, and writes the answer from what it found. Those retrieved sources come back attached to the response as structured objects Google calls groundingChunks, and each chunk's web URI is not the publisher's URL. It is a tokenised redirect on vertexaisearch.cloud.google.com, the grounding-api-redirect path you saw in the opener.

This is not a scrape artifact or a quirk of how we collected the data. It is Google's documented, intended output. The Gemini API grounding documentation shows the groundingChunks returning source URIs in exactly the https://vertexaisearch.cloud.google.com/... form, and it shows the accompanying title field carrying only a bare domain name, its own example is aljazeera.com, rather than the full original article URL. So by design, the richest source identifier Gemini volunteers is a domain string sitting next to a redirect, not a link you can click straight through to the business.

Why build it this way? Google has practical reasons that are easy to infer even where it does not spell them out: the redirect lets Google meter and attribute grounding traffic, apply its own safety and freshness checks at click time, and keep the resolved URL under its control rather than exposing a raw third-party link inside an API contract. None of that is sinister. But the consequence for anyone trying to measure visibility is the same regardless of intent. The source of record, the thing Gemini will admit to citing without extra work, is a Google domain. The business underneath is withheld at the field level. That is the opacity, stated precisely.

The opacity quietly breaks the way most tools track AI visibility

Standard AI-visibility tracking counts source URLs, and Gemini does not give you usable source URLs, so the standard method misreports Gemini by default. Most monitoring tools, and most agencies, built their AI tracking around the Perplexity and ChatGPT model: capture the answer, extract the cited links, check whether the client's domain is among them, count the hits. That pipeline runs fine until it meets Gemini. Point it at a Gemini answer and every citation resolves to the same domain, vertexaisearch.cloud.google.com. A naive tracker concludes one of two equally wrong things: either "Gemini cites only Google" or "the client's domain never appears." Both are false. The client may well be named, in the prose, three sentences deep, while the link field shows nothing but the wrapper.

We watched this happen in our own raw data before we corrected for it. The 384 figure is precisely the symptom: a single domain swamping the per-domain leaderboard, not because Google is the source, but because Google is the wrapper. If we had stopped at counting domains, the Gemini column of our study would have read "vertexaisearch.cloud.google.com, 384" and nothing else, and we would have learned nothing about which Ontario businesses Gemini actually recommends. The real information was one layer down, in the named entities inside the answer text, and it only surfaced because we read the prose rather than trusting the link.

There is a second, compounding problem, and Google states it openly. Not every Gemini response carries sources at all. Google's Gemini Apps help page is blunt about it: "Not all responses include related links or sources. If you don't have the Sources button below a response, Gemini Apps didn't provide any links for that particular response." So absence of a Sources panel is normal, by design, not a signal that your business was excluded. A tracker that treats "no Sources button" as "client not cited" double-counts the opacity: it misses the businesses hidden behind the wrapper, and it misreads the responses that were never going to show a wrapper in the first place.

If you are auditing your own Gemini presence right now and seeing a flat line, this is the first thing to rule out before concluding anything. The flat line may be a measurement failure, not a visibility failure. Before you spend a dollar trying to fix Gemini, it is worth a free AI visibility audit that reads the named entities, not just the links, so you know which of the two you are actually looking at.

What the named businesses behind the wrapper tell us about Ontario

Once you read past the wrapper to the businesses Gemini actually named, a clear and useful pattern appears. The redirect hides the URL, but it does not hide the prose, and the prose is full of real Ontario firms. Resolving what Gemini surfaced across our nine cities, three names recur often enough to be worth calling out, and they tell you what kind of business Gemini tends to reach for.

The lesson is not that these three companies bought their way in. It is that Gemini's grounding step favours businesses with a wide, consistent, well-corroborated presence across the web, the kind of footprint that shows up no matter which passages Google's retrieval happens to pull. A single great page rarely does it. A firm that appears, consistently named and consistently described, across directories, review sites, news mentions, and its own well-structured pages, is the firm the grounding step keeps finding. That is a different optimisation target from "rank a page," and it is the same entity-consistency principle we apply when we strengthen the local signals that AI near-me answers depend on.

Notice also what the wrapper does to the also-rans. Every Burlington HVAC company that was not Laird & Son, every Hamilton roofer that was not D'Angelo & Sons, received the same treatment in the data: nothing. Not a lower position. Not a footnote. They simply were not named, and because the only visible citation is the Google wrapper, their owners cannot even see which competitor pages won the slot. The opacity that hides Laird's source also hides everyone else's exclusion. You are left with presence or absence, named or not named, and very little explanation in between.

Why Gemini tells you the least about its sources of any engine we tested

On the identical Ontario prompts, the three engines differ most in one place: what they let you see about their own sources. We ran the same "best {vertical} in {city}, Ontario" prompts through each engine, and the contrast in transparency was sharper than the contrast in the businesses they named. Gemini is the engine that tells you the least about where its answer came from.

The practical takeaway is that you cannot port a Perplexity or ChatGPT tracking method onto Gemini and trust the output. With Perplexity, the cited URL is the measurement, you match your domain against it and you are done. With Gemini, the cited URL is a dead end, and the measurement lives in the answer prose. This is also why blanket "AI visibility scores" that average all engines into one number quietly mislead: they fold a link-based signal (Perplexity, ChatGPT) together with a wrapper that has no usable link signal (Gemini), and the blend hides exactly the engine that is hardest to read. For the deeper version of this argument, our look at the four-engine consensus gap shows how little the engines overlap once you account for each one's grounding layer, and our notes on how Google's AI Overviews surface local answers trace the same grounding layer into the search results page itself.

There is a real-world reason this engine matters in Canada specifically, beyond its market share. AI Overviews, Google's clipped answer at the top of a results page, fire far less often on local and transactional queries than on informational ones, and SeoProfy's 2026 data puts Canadian AI Overview adoption at around 19.2 percent. That means the place Ontario local recommendations increasingly surface is not the AI Overview, it is the grounded Gemini conversation, the exact surface this wrapper governs. And with DataReportal reporting that more than 94 percent of Canadians were online at the start of 2024, the audience asking Gemini for a roofer or a dentist is effectively the whole local market these businesses serve. The opacity is not an edge case. It sits on the main road.

How to actually measure whether Gemini is recommending your business

You measure Gemini by reading the named entity in the answer, repeatedly, rather than by counting links once. Because the wrapper hides URLs and the Sources panel is inconsistent, the only durable signal Gemini gives you is whether your business is mentioned by name in the prose. That is harder to automate than URL matching, but it is reliable, and it is the method behind the 384 figure in this very study. Here is the approach we use, framed as something a careful owner could replicate.

Read Gemini honestly and the method is five disciplined habits. Track the name, not the link: search the answer text for your exact business name and close variants, since the vertexaisearch redirect is not your visibility signal and the named mention is. Sample the same prompt many times, because Gemini's output varies run to run, so report how often you are named rather than whether you appeared once. Resolve the redirect only when you need the source, following the grounding-api-redirect token to its destination as forensic work, not routine tracking. Expect responses with no Sources panel, since Google's own help docs confirm many answers carry no links at all, so a missing panel is not an exclusion. And watch your wider web footprint as the leading indicator, because grounding favours consistently corroborated businesses, so growth in your named presence across directories and reviews tends to precede growth in Gemini mentions.

This is Vector 11, Measure, applied to the one engine that resists conventional measurement hardest. Most of Formative Digital's 12 Vectors describe how to earn visibility; Measure is the discipline of confirming it honestly, and Gemini is where that discipline earns its keep. We run this reading through the Formative Forces, our orchestrated multi-agent system, so a single business can be checked across hundreds of prompt-and-city combinations and scored on named-mention frequency rather than a brittle link count. A human team reading Gemini prose at that volume, repeatedly, to average out the run-to-run wobble, is not economical. Orchestration is what makes honest Gemini measurement affordable, and it is the reason we trust the 384 rather than guessing at it. The fuller method sits alongside our work on diagnosing AI visibility when there is no single ranking to point at.

One honest qualifier belongs here, because GEO advice is something owners spend real money on. What you can do about Gemini depends on your industry, your competition, and the web footprint you start with; named-mention frequency is a measurable target, not a guaranteed one. The peer-reviewed GEO study by Pranjal Aggarwal and co-authors (arXiv:2311.09735, presented at KDD '24) found that targeted optimisation, adding citations, quotations, and statistics, can raise a source's visibility in generative answers by up to 40 percent, and that the effective tactics vary by domain. So citability is engineerable even inside an opaque engine. It is just engineered against the named entity and the footprint, not against a link you can see.

Matt Griffin, who founded Formative Digital, puts the discipline in plain terms: "Everyone wants the link Gemini used. Gemini will not give you the link, and chasing it is a waste of a measurement budget. We stopped tracking the wrapper and started tracking the name. Across nine Ontario cities, one Google domain accounted for 384 of Gemini's citations, which tells you the source is hidden, not that Google is the source. The only thing you can see, and therefore the only thing worth measuring, is whether your business gets named in the words on the screen. Engineering Principles, not magic ranking dust."

What an Ontario local business should do to get named inside a Gemini answer

You earn a Gemini mention by building the kind of broad, consistent, well-corroborated web presence its grounding step keeps rediscovering, then you measure the mention rather than the link. Because you cannot see Gemini's sources, you optimise for the trait the named businesses shared, footprint breadth and consistency, rather than for any single placement. The businesses that won in our data, Laird & Son, Preszler Law, D'Angelo & Sons, were not the ones with one brilliant page. They were the ones whose name and description were the same everywhere Google's retrieval looked.

In practice that means a few concrete moves, drawn from the 12 Vectors. Anchor your entity so Gemini's retrieval resolves you unambiguously: one consistent name, address, and phone number, a clean Organization and LocalBusiness schema, a tight set of facts repeated identically across your own pages and your listings. Distribute that entity widely, the directories, review sites, and local news mentions that grounding tends to pull from, because breadth of corroboration is what the recurring names had in common. Structure your own pages with clear Schema.org markup so that when Google's grounding pipeline does read you directly, it can parse exactly who you are and what you do. And refresh, since grounded answers favour current, dated material over stale pages.

What you should not do is treat Gemini like Perplexity and pour effort into "getting your URL cited," because the URL is the one thing Gemini will never show. The point of effort is the entity and the footprint, the parts the wrapper cannot hide, not the link the wrapper always hides. This is the same entity-and-distribution work we describe for being found as a local service business across search and AI surfaces at once, applied to the engine that happens to mask its sources hardest.

Does the work pay off? It depends on the starting point, and we say so plainly rather than promising a number. Our Brantford retail client Mattress Miracle grew from roughly 1,000 to over 82,400 monthly organic visits (SEMrush, April 2026) through sustained entity and structured-content work, and as owner Brad put it, "In 40 years of advertising I've never seen anything like this. It's a completely new business." That is one industry and one starting position; a Burlington HVAC firm or an Oakville law practice will see different curves. The principle that carries across is the one this whole page rests on: with Gemini you build for the named mention and you measure the named mention, because the wrapper has taken the link off the table.