Does "Near Me" Work the Same on ChatGPT as on Google Maps?

In late March 2026, SEO consultant Glenn Gabe did something simple: he turned on ChatGPT's brand-new location-sharing feature and asked it for a good steakhouse nearby. Several of the recommendations it gave him were about 45 minutes away — reported by Search Engine Roundtable the same week OpenAI rolled the feature out. Not "in another city." Not "technically still driving distance if you squint." A 45-minute drive, served up as a "near me" answer, from a company that had just spent engineering effort building a feature specifically to fix this. That's not a knock on Gabe's steak preferences — it's the cleanest illustration you'll find of a question a lot of local businesses haven't actually thought through yet: when someone asks an AI assistant the same "near me" question they'd type into Google, does it work the same way? It genuinely doesn't, not yet, and the gap matters more than most local SEO advice currently accounts for.

How Google Actually Decides "Near Me"

Start with the system everyone already half-understands. Google's own Business Profile help documentation is unusually direct about this: local results run on three factors — relevance ("how well a Business Profile matches what someone is searching for"), distance ("how far each business is from the customer who's searching"), and prominence ("how well-known a business is," driven largely by reviews and how the business is represented elsewhere online). If a searcher doesn't share their location, Google still calculates proximity from whatever location signal it can get. That's the whole mechanism behind the Local Pack and Maps — a real-time distance calculation running against a profile Google already trusts.

Whitespark's 2026 Local Search Ranking Factors report, built from a survey of 47 local search experts scoring 187 individual factors, puts numbers on exactly which signals carry the most weight inside that framework. Primary Google Business Profile category tops the list at 227 points, proximity of the address to the search point comes in at 225, keywords in the GBP business title at 223, and physical address in the city being searched at 213. The newest addition for 2026 is telling on its own: "Business is Open at Time of Search" debuted at 189 points — because a Local Pack result that's currently closed is a worse answer than one that's open, even if it's slightly farther away. Reviews still carry real weight too — high numerical ratings, review quantity, and review recency all land in the top tier. None of this is a mystery. It's a well-documented, heavily gamed, thoroughly reverse-engineered system that local SEO has been optimizing against for close to two decades.

ChatGPT Didn't Actually Know Where You Were — Until Very Recently

Now compare that to what ChatGPT was doing before March 2026. It had no formal distance calculation, no equivalent of "proximity of address to search point," and for most of its life, no explicit permission system for location at all. What it had — and this is the part worth sitting with — was your IP address, and a willingness to infer a lot from it without telling you.

That exact behavior became the subject of a genuinely detailed Hacker News discussion sparked by a blog post documenting how ChatGPT seemed to know a user's city despite denying it had any location access. The thread pulled in people who actually build this infrastructure. A commenter named lpellegr, who said they work for the IP intelligence provider Ipregistry, explained the mechanics plainly: "ChatGPT receives your IP when you connect. The model itself does not know your location, but the system around it can add location context before the model answers... Accuracy depends on your network setup. VPNs, mobile networks and corporate proxies can make the location drift quite a bit." Another commenter, reincoder, identifying as an IPinfo employee, added that "geolocation, at least on a country level, can be largely inferred based on conversational context" even without an explicit lookup. A third, incolumitas, pointed to consumer-facing tools like ipapi.is as evidence this isn't exotic technology — "it's not hard to infer location data from IP addresses, albeit it's not necessarily an exact science."

What made the thread worth reading wasn't the technical explanation, though — it was what regular users reported experiencing. A commenter called allenu described asking for the "best fried chicken near me" and getting city-specific results, then separately testing whether ChatGPT would admit to using location data: "Gaslit me as well when I told it it had access to my location and its response [was] that it used Google and Google provided a 'default search location.' Obviously it can figure out location by IP, but the lying is insulting and creepy." NicuCalcea reported something similar from the opposite direction — a VPN set to Ireland caused ChatGPT to confidently tell them "since you are in Ireland, you can..." when they weren't in Ireland at all, with zero indication the assumption was even happening, unlike "a normal site" that would typically ask before switching context. A commenter named Bender ran the test cold — asked ChatGPT directly whether it had location access (it said no), then asked for nearby food options and got a list complete with the correct town name. Not everyone in the thread saw this as alarming — roscas argued the real story is that "if you use any Apple stuff, it knows your exact location... Android is the same. Windows knows the exact location also" — but the consensus that emerged wasn't really about whether ChatGPT could infer location. It clearly could, approximately, off IP alone. It was about the fact that it wasn't telling anyone when it was doing it, which is a meaningfully different situation than Google Maps, where "share your location" is a visible, deliberate permission a user grants.

What Actually Changed in March 2026

That's the backdrop against which OpenAI's location-sharing rollout matters. According to reporting from PPC Land and Search Engine Land, OpenAI shipped the feature on iOS and web for all consumer tiers in late March 2026 (Android is still pending), with two distinct layers users can toggle separately: approximate location for general regional context, and precise device-level location — described in OpenAI's own language as letting "ChatGPT use your device's specific location, such as an exact address, to provide more tailored results." It's opt-in and off by default, and OpenAI says precise coordinates are deleted after a response is generated, though the resulting recommendations stick around in your chat history. Glenn Gabe, quoted in that PPC Land coverage, framed the move in blunt competitive terms — "Google dominates" for local queries, and this is OpenAI's attempt to close that gap.

Perplexity has its own version, described in its help center under Personalization settings as the option to "opt in to precise location for location-aware answers" — but it's worth noting the location-aware infrastructure underneath is still visibly under construction. A bug report on Perplexity's own developer community forum from mid-2025 shows a developer unable to get the API's location filter working with coordinates alone — it silently required a country parameter that wasn't documented, confirmed as a known gap by a Perplexity team member (vikvang) who said the team was still working on supporting coordinate-only lookups properly. That's a small, unglamorous detail, but it's a real one: the location plumbing behind "near me" answers on these platforms is being built in public, mid-2026, in a way Google's twenty-year-old Local Pack infrastructure simply isn't.

The Adoption Curve Nobody in Local SEO Predicted

Here's what makes all of this worth taking seriously rather than filing under "interesting but early": people are already using it, at a pace that outran most forecasts. BrightLocal's research, published in March 2026 from a survey of 1,002 US consumers, found that 45% now use AI tools for local business recommendations — up from just 6% a year earlier. That's a genuinely startling jump for a twelve-month window, and it puts AI ahead of Yelp and TripAdvisor as a discovery channel, trailing only Google and Facebook. Within that group, ChatGPT is the platform of choice at 31% share, with Google's AI Mode at 23% and Gemini, Copilot, and Claude splitting the rest.

Trust is the more interesting number, though, because it isn't simple. BrightLocal found 63% of active AI users trust the recommendations they get, and 71% specifically trust AI-generated review summaries — the prose paragraph an AI system writes describing what reviewers said, rather than a star rating a person scans in half a second. But that trust comes with real skepticism baked in: 88% of those users report checking whether a review is legitimate or tracking down its original source, and 97% say they at least sometimes cross-reference an AI recommendation against the actual underlying reviews before acting on it. People are treating AI local answers less like a verdict and more like a first draft — useful, but worth double-checking, which is exactly the behavior you'd expect toward a system that just spent a year proving it doesn't always know where you actually are.

Why the "92% Invisible" Number Should Make You Suspicious, Not Scared

It's worth pausing here on the kind of statistic that circulates in this space and deserves more scrutiny than it usually gets. A Show HN post from a startup called Chatalyst — posted to Hacker News — claimed that in a small experiment testing AI models on queries like "where should I advertise my small business in Omaha," 92% of real local businesses simply didn't show up in the answers at all. That's a striking number, and it's the kind of thing that ends up quoted in a dozen agency sales decks by the following week. But the top comment on the thread, from a user named nerdsniper, cuts right to the actual problem: "I can't find a writeup of this experiment at the link posted to HN. I need to know a lot more than the few sparse sentences you wrote here." No sample size, no query list, no model versions specified in the post itself — just a headline number attached to a product pitch. That doesn't make the underlying pattern false; a real and growing gap between traditional visibility and AI visibility is well documented elsewhere in this piece. But it's a good reminder that "AI doesn't know your business exists" statistics need the same scrutiny you'd apply to any other marketing claim, and a number with no methodology behind it is worth exactly as much as the vague true answer this whole industry keeps warning against.

When AI Mixes Up Which Business It's Even Talking About

There's a sharper version of the "wrong information" risk than an AI simply not knowing a business exists, and it played out in a real courtroom rather than a Reddit thread. In June 2026, a German court in Munich issued an injunction against Google after its AI Overviews conflated two Munich-based publishers with unrelated companies, generating summaries linking them to "scams, subscription traps, and questionable business practices" that, per the court's findings, "didn't appear in the linked sources" — the AI Overview had synthesized allegations about entirely different businesses and attached them to the wrong names. The court's reasoning is worth sitting with regardless of the outcome on appeal: because AI Overviews "rewrite, combine, and evaluate information in its own words and according to its own structure," the court treated the output as Google's own speech, not a neutral reflection of what's actually on the linked pages — and held Google liable for it. That's the mechanical failure mode behind entity confusion at its most consequential: a system trying to be helpful, blending signals from businesses with similar names or overlapping categories, and getting the attribution wrong in a way traditional search — which just shows you the ten blue links and lets you sort it out — structurally can't do. It's the same underlying weakness that drives our AI Local SEO work toward entity consistency as a starting point rather than an afterthought: the less ambiguous a business's name, address, and category are across the web, the less raw material an AI system has available to accidentally merge it with something else.

So What Does Actually Transfer From Google to AI, and What Doesn't

Put the pieces together and a fairly specific picture forms, one that's more nuanced than either "it's all the same" or "it's completely different." Reviews still matter enormously on both sides — but Google reads them as a star rating driving a ranking score, while AI platforms increasingly synthesize them into a written summary, which changes what "good reviews" needs to accomplish: a five-star average with one-word reviews gives an AI system almost nothing to write a confident summary from, while detailed, specific reviews give it real sentences to draw on. Distance is the sharpest divergence point. Google has been calculating real-time proximity against verified location data for two decades; AI platforms are, as of this year, still building that infrastructure in public — sometimes inferring it silently and unreliably from an IP address, sometimes asking for explicit permission and still getting the radius wrong, as Gabe's steakhouse test showed. And entity clarity — making sure a business's name, category, and location say the same consistent thing everywhere a system might encounter them — matters on both platforms, but the cost of getting it wrong is different: Google might just rank you lower; an AI system might, per the German case, actually attach someone else's problems to your name.

None of that means the fundamentals stop mattering. A complete, accurate Google Business Profile, real reviews, and consistent NAP information are still the foundation — Whitespark's report is blunt about this, noting that "research and discovery of local businesses is still predominantly done through Google's local results" even as AI visibility becomes its own tracked category. What it does mean is that a business treating AI local visibility as identical to Google local visibility, or as a problem that solves itself once the Local Pack is handled, is working from an outdated map. The distance calculation, the location-permission model, and the way "good reviews" gets translated into an actual answer are different enough mechanisms that measuring only one of them — and assuming the other looks the same — is how a business ends up confidently ranking well on Google Maps while ChatGPT sends its customers 45 minutes in the wrong direction. If you want a real, evidenced read on where your own business currently stands across both systems rather than a guess, the AEO Score tool is a free way to check before deciding what, if anything, needs fixing.