The short answerYes. In 392 answers from ChatGPT, Perplexity and Google AI Overviews on 11 November 2026, naming the country in an agency question cut the share of named firms in common with the US list to 0% to 7%. Changing only the search country kept the US leaders on top and added local firms.
Each cell below is the share of named firms two lists have in common, averaged across 12 agency questions. The first column is the baseline: the same question asked twice from the same country.
| Engine | Same question, same country, two runs | Silent location: UK / Canada / Australia vs US | Named country: UK / Canada / Australia vs US |
|---|---|---|---|
| ChatGPT | 32% | 25% / 29% / 26% | 3% / 7% / 2% |
| Perplexity | 79% | 49% / 57% / 59% | 2% / 2% / 0% |
| Google AI Overviews | 35% | 13% / 16% / 14% | 2% / 0% / 1% |
And the share of located firms headquartered in the asking country:
| Engine | US searcher (baseline): UK / Canadian / Australian firms | Silent location: UK / Canada / Australia | Named country: UK / Canada / Australia |
|---|---|---|---|
| ChatGPT | 8.3% / 2.4% / 1.8% | 39% / 20% / 28% | 97% / 81% / 91% |
| Perplexity | 8.5% / 2.3% / 2.8% | 26% / 4% / 13% | 95% / 80% / 95% |
| Google AI Overviews | 21.2% / 0% / 0% | 37% / 17% / 25% | 95% / 93% / 97% |
When a buyer types the country into the question, all three engines swap the US agency list for a local one; when only the search location changes, the US leaders stay on top and local firms join them.
That second case is the one most people never see. If you track AI visibility from a US tool, or shortlist agencies from a US laptop, you are reading one country's answer. Your buyer in Manchester or Melbourne may be reading another.
How we ran the study
I froze the method before collecting any data, then ran every call on 11 November 2026.
The questions. Twelve agency questions LoudFace tracks, each one where LoudFace was named in under 10% of Peec answers: "Best GEO agencies", "Best AI search agencies", "Best AI visibility agencies", "Best LLM SEO agencies", "Best agency for getting cited in ChatGPT", "Best agency for getting mentioned in Perplexity and ChatGPT", "Which agencies will help us show up in ChatGPT answers?", "Top AEO consultants 2026", "Best SEO agency for B2B SaaS companies", "Top B2B SaaS SEO agencies in 2026", "Boutique SEO agency for B2B SaaS brands" and "Best agency for growing SaaS organic traffic".
The engines. All three ran through DataForSEO. ChatGPT used model gpt-5.5 (version gpt-5.5-2026-04-23) with web search on. Perplexity used sonar-pro with web search on. Google AI Overviews came from Google organic results, desktop, English. Each run set its own search country. These are API runs, a stand-in for the consumer apps, which matters for the limits at the end.
The countries. United States, United Kingdom, Canada and Australia.
Two arms.
- Silent location (arm A). The same question text, with only the search country changed. Two runs per question per country on each engine: 96 answers each on ChatGPT and Perplexity, and 96 Google searches. Google's second run was added after a first look at the data, so it would have a same-country baseline like the other two.
- Named country (arm B). The question with "in the UK", "in Canada" or "in Australia" added, run from that country. One run per question per country per engine, 36 each.
That is 396 runs: 132 ChatGPT answers, 132 Perplexity answers and 132 Google searches. Google showed an AI Overview on 128 of its 132 searches, which gives 392 answers.
What counts as a named firm. We built a dictionary of 458 agencies and 29 named consultants from every bold name, list head and table first cell in the pooled answers (2,256 strings), reviewed it and froze it before any comparison. A firm counts when one of its names appears in the answer. Firms whose name is a common word, such as Found, Impression or Directive, count only when bolded or at the head of a list item or table row. Tools like Profound, Peec AI and AirOps don't count. Neither do directories like Clutch.
How we compare two lists. For each pair of answers we take the firms both lists name and divide by the firms either list names. That is the share of named firms the two lists have in common (statisticians call it Jaccard overlap). We average it across questions and skip a pair when either answer names no firm.
Where each firm is based. The 201 firms named in 3 or more answers were located from their own sites: head office if stated, otherwise the first-listed office. That came out as US 49, UK 46, Australia 45, Canada 29, elsewhere 15, and 17 with no stated location. Local share counts only the located firms, so firms named in one or two answers sit outside it.
One quirk. ChatGPT read "Boutique SEO agency for B2B SaaS brands" as a request for positioning copy in 7 of its 8 silent-location runs and named no agency at all. Those runs stay in the counts and drop out of the overlap pairs.
Same question, same country: how much answers vary anyway
Before you can say a country changed an answer, you need to know how much the answer changes on its own. So every question ran twice from every country.
| Engine | All four countries | US | UK | Canada | Australia |
|---|---|---|---|---|---|
| ChatGPT (44 pairs) | 32% | 31% | 26% | 28% | 43% |
| Perplexity (48 pairs) | 79% | 75% | 77% | 88% | 77% |
| Google AI Overviews (42 pairs) | 35% | 41% | 25% | 25% | 51% |
Google's pairs count only questions where an AI Overview appeared and named at least one firm both times: 9 in the US and 11 in each other country. In the US, two questions lost their pair because a search showed no AI Overview and one because the Overview named no firm. In the UK, Canada and Australia every silent-location search showed an Overview, and one question lost its pair because an Overview named no firm.
Perplexity is the steady one. Ask it the same agency question twice from the same place and about four in five named firms repeat. ChatGPT and Google AI Overviews repeat about a third.
None of this is new. Every top-ranking page on this question agrees ChatGPT doesn't give everyone the same answer, and that wording changes it. SparkToro and Gumshoe found ChatGPT or Google's AI has under a 1 in 100 chance of giving the same list of brands in any two responses to the same prompt. Our own 14-day study of ChatGPT answers held location fixed at the US, and our phrasing study covers wording.
What this table adds is a floor per engine. Any country effect has to show up below it.
Silent location: what changes when only the search country moves
Same words, different search country. Here is how much each country's list shares with the US list, run matched to run:
| Engine | UK vs US | Canada vs US | Australia vs US | Same-country floor |
|---|---|---|---|---|
| ChatGPT (22 pairs per country) | 25% | 29% | 26% | 32% |
| Perplexity (24 pairs per country) | 49% | 57% | 59% | 79% |
| Google AI Overviews (20, 19, 20 pairs) | 13% | 16% | 14% | 35% |
Read each row against its own floor. Perplexity's lists move 20 to 30 points more between countries than between two runs in one country. Google's move about 20 points more. Those are real location effects, and the only thing we changed was the search country.
ChatGPT is harder. Its cross-country overlap sits 3 to 7 points under its own run-to-run overlap. On overlap alone, a silent country change is hard to tell apart from ChatGPT's ordinary variation.
Overlap isn't the only lens, though. Look at who gets added, and ChatGPT moves more than either of the others. For a US searcher, 8.3% of the located firms ChatGPT named were UK firms. Run the same questions from the UK and that share is 39%. Canadian firms go from 2.4% to 20%, Australian firms from 1.8% to 28%.
So both readings are true at once. ChatGPT's lists churn so much run to run that the overlap measure can't separate a country shift from the churn. The local-share measure can, and it shows the shift.
US firms still hold most of the room in silent-location answers: 55%, 66% and 59% of located firms on ChatGPT (UK, Canada, Australia), 60%, 74% and 68% on Perplexity, and 52%, 55% and 47% on Google AI Overviews.
How many silent-location answers named at least one local firm:
| Engine | UK | Canada | Australia |
|---|---|---|---|
| ChatGPT | 21 of 24 | 13 of 24 | 16 of 24 |
| Perplexity | 24 of 24 | 4 of 24 | 12 of 24 |
| Google AI Overviews | 21 of 24 | 10 of 24 | 14 of 24 |
A UK searcher who never mentions the UK will usually see a UK firm somewhere in the list. A Canadian searcher on Perplexity almost never will.
Naming the country: a different list
Add "in the UK" to the question and the overlap with the US list falls close to zero:
| Engine | UK vs US | Canada vs US | Australia vs US | Named vs silent list, same country (UK / Canada / Australia) |
|---|---|---|---|---|
| ChatGPT | 3% | 7% | 2% | 13% / 11% / 14% |
| Perplexity | 2% | 2% | 0% | 10% / 2% / 5% |
| Google AI Overviews | 2% | 0% | 1% | 12% / 5% / 1% |
Each cell has 10 to 12 pairs. The last column compares the named-country list with the silent list from the same country, and it is low too. Typing the country moves the list further than any silent location change: in the UK, Canada and Australia, the named list has little in common with the silent list from the same place.
The lists don't get shorter. Firms named per answer stay in the same range across arms: ChatGPT 8.4 to 9.5 in silent answers and 7.6 to 8.6 in named ones, Perplexity 8.3 to 8.6 and 7.7 to 9.1, Google AI Overviews 5.4 to 6.5 and 5.3 to 6.3. The engines replace the list at full length.
And the replacement is local. Of the located firms in named-country answers, 97% of ChatGPT's picks for the UK were UK firms, and 81% of its picks for Canada and 91% for Australia were local. Perplexity and Google sit at 80% to 97%. Every named-country answer on every engine included at least one local firm (12 of 12; Google Canada 11 of 11).
One caveat on those local shares. Named-country answers lean further into firms named only once or twice across the study. On ChatGPT, those long-tail mentions are 21 of 97, 30 of 91 and 32 of 103 (UK, Canada, Australia) in named answers, against 25 of 215, 15 of 223 and 13 of 228 in silent ones. Long-tail firms aren't located, so the 81% to 97% figures describe the firms that recur.
Which agencies each country gets
The five firms named most often in US silent-location answers (all engines pooled, 69 answers), and what happens to them:
| Firm | US, silent (69 answers) | UK / Canada / Australia, silent (72 each) | UK / Canada / Australia, named (36 / 35 / 36) |
|---|---|---|---|
| Omniscient Digital | 45 | 35 / 50 / 46 | 0 / 1 / 1 |
| Siege Media | 37 | 31 / 34 / 32 | 0 / 0 / 0 |
| First Page Sage | 31 | 24 / 29 / 28 | 1 / 3 / 0 |
| iPullRank | 29 | 29 / 30 / 31 | 1 / 1 / 0 |
| Directive Consulting | 20 | 14 / 14 / 15 | 1 / 1 / 1 |
Silently, the US leaders travel. The top four in UK, Canadian and Australian silent answers are the same four firms: Omniscient Digital, Siege Media, iPullRank and First Page Sage. SimpleTiger and the UK firm Skale come next. Siege Media, named in 37 US answers, appears in no named-country answer at all.
Who leads when the buyer names the country (answer counts):
| UK (36 answers) | Canada (35 answers) | Australia (36 answers) |
|---|---|---|
| The SEO Works, 16 | Search Engine People, 19 | StudioHawk, 19 |
| Found, 14 | Powered by Search, 11 | Prosperity Media, 18 |
| Rank4AI, 14 | Konstruct Digital, 10 | Impressive Digital, 13 |
| ClickSlice, 11 | Azuro Digital, 9 | Rocket Agency, 13 |
| Impression, 11 | BlueHat Marketing, 8 | Online Marketing Gurus, 7 |
| Skale, 11 | DoodleWeb Canada, 7 | Latent Analytics, 6 |
Skale is the one name that sits on both sides: it follows the US leaders in silent answers and makes the UK named-country list.
Each engine treats location its own way
Every page ranking for this question treats it as a ChatGPT question. The three engines behave differently enough that the answer depends on which one you mean.
Perplexity is the most repeatable engine (79% same-country overlap), and its silent cross-country overlap drops to 49% to 59%. That gap is the clearest silent location effect in the study. Yet its silent answers add the fewest local firms. Canadian firms make up 4% of its silent Canada answers, and 4 of 24 named one at all. Perplexity reshuffles the international names by location more than it adds local ones.
Google AI Overviews repeats about a third of its firms run to run (35%), a little more than ChatGPT's 32%, and drops to 13% to 16% across countries. The US answers on Google were also the most international to begin with: UK firms made up 21.2% of located firms for a US searcher, against 8.3% and 8.5% on the other two.
ChatGPT varies most run to run (32%), and its silent location shift shows in local share (UK firms 8.3% to 39%) more than in overlap. Sometimes its own text says it is answering for the location. In two silent-location runs from the UK, where the question never mentioned the UK, it offered agencies "to shortlist, especially from a UK buyer perspective" (second run of "Best GEO agencies") and included a section headed "My suggested shortlist for a UK buyer" (first run of "Top AEO consultants 2026").
So ChatGPT does not ignore your location. It changes who it adds. Its lists are just too noisy for a simple overlap number to show it.
Which sources each country cites
Citations follow the country too. Cited URLs on the asking country's domain (.uk, .ca or .au):
| Engine | UK, silent | Canada, silent | Australia, silent | UK, named | Canada, named | Australia, named |
|---|---|---|---|---|---|---|
| ChatGPT | 42 of 193 | 21 of 182 | 42 of 188 | 46 of 95 | 36 of 97 | 59 of 99 |
| Perplexity | 4 of 240 | 0 of 240 | 0 of 240 | 40 of 120 | 15 of 120 | 49 of 120 |
Google AI Overviews is left out of this table. 101 of its 396 references came back as google.com redirect links that hide the cited site, so a country-domain count would miss about a quarter of its sources.
ChatGPT pulls local-domain sources even when nobody names the country. Perplexity barely does: 4 of 240 citations in silent UK answers, none in Canada or Australia. Name the country and both engines shift. That fits the Perplexity pattern above, where location reshuffles the international names but rarely adds local firms.
The most-cited domains in named-country answers show what the engines read. On ChatGPT for the UK: seocompare.co.uk (6 answers), castle.co.uk (4) and saas-hackers.com (4). On ChatGPT for Australia: latentanalytics.com.au (6) and digitaldigest.au (5). On Perplexity for Canada: citehawk.com (10), agencies.semrush.com (6) and doodleweb.ca (5). Latent Analytics and DoodleWeb Canada are on the leader table above, and their own sites sit among the most-cited sources.
LoudFace shows the same pattern. It is named in 9 of 392 answers, all with a silent location (ChatGPT 7, Google AI Overviews 2, Perplexity 0), and in none when the country is named. 7 of those 9 answers cite a loudface.co page. loudface.co is cited in 54 of 285 silent-location answers and 1 of 107 named-country answers. The sources moved with the country, and the names moved with the sources.
What ChatGPT, Perplexity and Google say they do with location
All three companies say location feeds their search. None says how much it weighs.
ChatGPT. OpenAI's help center says ChatGPT search sends rewritten queries to partner search providers, and that "ChatGPT also collects general location information based on your IP address and may share that general location with third-party search providers to improve the accuracy of your results." Its example is a "near me" restaurant prompt from the San Francisco area, which ChatGPT may rewrite as "top restaurants San Francisco." OpenAI says it does not pass the IP address itself or account information to those providers.
The same page says "ChatGPT may use your IP address to estimate your general location, such as your country, state, or city." Precise device location is a separate opt-in that is off by default. Two other things can shift the inferred location: "A VPN or network location may affect the approximate location inferred from your IP address," and "If memory is enabled, saved location information may also influence search results." On ranking, OpenAI says only that ChatGPT uses "multiple factors" and that "Placement is not guaranteed."
For developers, the OpenAI API web search tool takes an optional approximate user location by country, city, region and timezone, "To refine search results based on geography."
Perplexity. Its privacy notice says "Our services may collect your general location from the IP address from which you connect. This lets us determine your general location for location-specific queries." Its help center lists Location as an optional profile setting that "can help tailor responses to relevant local information." Its API web search tool takes a user location by country, region, city, latitude and longitude, and says "city and region significantly improve location accuracy."
Google. At I/O in May 2025 Google said "AI Overviews are now available in more than 200 countries and territories and in more than 40 languages." Its AI Overviews help page lists the US, Canada, the UK and Australia among them, and says AI Overviews show only when its systems decide generative AI can be especially helpful. Google's How Search works page says it uses "information such as your location, past Search history, and Search settings to determine what is most relevant for you in the moment." Its example: "football" in Chicago likely returns American football and the Bears, while in London it may return soccer and the Premier League. And per Google Search Help, "When you search on Google, Google will always estimate the general area that you're searching from."
On the publisher side, Google Search Central says targeting one country "can improve your page rankings in the target country, but at the expense of results in other locales or languages." It reads country-code domains, local addresses, local language and currency, and links from other local sites, among other signals.
Read that list next to the source table above. The ccTLD pages that named-country answers cite are the kind of page Google says it treats as local.
What earlier tests found
Luxeo's test. The closest earlier work is a test by Luxeo, an SEO agency, published in August 2025. It ran ChatGPT through 2 logged-in accounts and 2 incognito sessions behind proxies in the USA, Britain, Canada, Australia and New Zealand, with 30 queries split into geo-dependent, not obviously geo-dependent and geo-independent groups. It concluded that "ChatGPT's output is geo-dependent for queries that are geo-dependent or not obviously geo-dependent" and that "For geo-independent queries, the results are independent of the user's location." It published no rates or significance tests.
Our agency questions name no place, and silent location still moved the lists on Perplexity and Google, and moved local share on ChatGPT. Our result differs from Luxeo's without refuting it: Luxeo tested ChatGPT alone, with different queries, through a consumer app, and judged the results by reading them. We tested three engines through the API and measured overlap. An agency question may simply count as "not obviously geo-dependent" in Luxeo's terms.
The MIT study. Cybernews reported an MIT study as showing chatbots give worse answers to non-US users. The paper, "LLM Targeted Underperformance Disproportionately Impacts Vulnerable Users" by Elinor Poole-Dayan, Deb Roy and Jad Kabbara, tested GPT-4, Claude 3 Opus and Llama 3 with a short user biography added before each question. Country of origin was a trait stated in that biography, and the paper reports that "Claude significantly underperforms for Iran on both datasets." The study did not test IP-based location, web search or brand recommendations. It shows a model can treat users differently by stated nationality. It says nothing about the location mechanism in this study.
SparkToro and Gumshoe. Their 600 volunteers ran 12 prompts 2,961 times through ChatGPT, Claude and Google's AI, with default settings and no requirement to be in the same country. Their verdict on rank: "any tool that gives a 'ranking position in AI' is full of baloney." They found how often a brand appears across many runs a steadier measure. Google's AI named the agency Smartsites in 85 of 95 responses to one consultant prompt.
That last point shapes how to read our tables. One answer from one country is a sample. Frequency across runs and countries is the signal.
What this means if you are choosing an agency, or tracking AI visibility
This is our read of the data. The study measured what the engines named. It did not test any of the advice below.
- Track from the country your buyers search from. A US tracker shows you a list where 72.9% to 84.5% of located firms were American. A UK buyer on the same engine sees UK firms in most answers without typing "UK". If your pipeline is in London or Sydney, a US-only reading is the wrong sample.
- Track in the buyer's own words, with and without the country. A buyer who writes "best GEO agencies in Australia" gets a list with almost nothing in common with the US one. If your buyers name their country, a silent-location tracker misses the list they actually see.
- For ChatGPT, watch who gets added as well as the overlap. Its lists churn so much that a country effect hides inside the noise. Local share showed what overlap could not.
- Measure frequency over many runs. With a same-country floor of 32% on ChatGPT and 35% on Google AI Overviews, a single run tells you little. Count how often a name appears across runs and countries.
- If you are shortlisting, ask in your own country, and then ask once more naming it. The silent list gives you the international names. The named list gives you local firms, and many recur across engines: StudioHawk and Prosperity Media in Australia, Search Engine People in Canada, The SEO Works in the UK.
- Check the sources behind a local list. Named-country answers lean on country-domain sites and a few local directories. In some cases a firm's own site is among the most-cited sources for the list that names it.
If you want to be named by buyers in a market, the source table suggests where to start: local-domain pages and local lists the engines already cite. Our GEO agency service tracks AI citations per engine, every week. For other cuts of the same question, see our studies on industry, funding stage, SEO versus AI search agencies and whether the three engines name the same agencies.
Limitations
- API runs, not the consumer apps. No logged-in account and no memory. OpenAI says memory and device location can change the inferred location, and our runs had neither.
- English-speaking countries only. The US, UK, Canada and Australia. The study says nothing about other languages or markets.
- One day. Every call ran on 11 November 2026.
- One run per named-country cell. The named-country arm has 10 to 12 pairs per cell, and no same-country repeat to measure its own noise.
- Headquarters for recurring firms only. Local share covers the 201 firms named in 3 or more answers. Named-country answers lean further into the long tail, so their local shares describe the firms that recur.
- One question broke on ChatGPT. It read "Boutique SEO agency for B2B SaaS brands" as a copy request in 7 of 8 silent runs. Those runs drop out of the overlap pairs.
- Google citations. 101 of the 396 Google AI Overviews references came back as google.com redirect links that hide the cited site, so the source table covers ChatGPT and Perplexity only. Google also showed no AI Overview on 4 of its 132 searches.
- Overlap and local share can disagree. On ChatGPT they did: silent-location overlap stayed close to its run-to-run floor while UK firms went from 8.3% to 39% of located firms. Either measure alone would have told half the story.
