September 2026 · Clusters A–F · 179 cells
GEO Visibility Index — September 2026
We put thirty fixed questions to ChatGPT, Gemini and Perplexity, twice each, and counted which agencies were mentioned, which sources the answers were built from, and where our own site was cited. 179 of the 180 planned cells are valid.
The short answer
Between 16 and 18 September 2026 we put thirty fixed questions to ChatGPT, Gemini and Perplexity, twice each: choosing an agency (cluster A, 8 questions), price and scope (B, 4), problem awareness (C, 6), sector-specific choice (D, 6), brand and entity (E, 4) and English-language queries (F, 2). We counted 179 of the 180 planned cells. On the agency-choice questions the most frequently mentioned agency differed by engine: on Perplexity, Lein Digital appeared in 10 of 16 cells; on ChatGPT, OFG Medya in 7 of 16; on Gemini, Crabs Media, Lein Digital and Zeo in 6 of 16 each.
The sources differed by engine and by question type. In cluster A, of the 130 sources Perplexity showed, 65 were list, guide or blog posts on agency and consultant sites and 31 were lists on news and magazine sites; of ChatGPT's 76 citations, 54 went to agencies' service, pricing or company pages; of the 51 site names shown on Gemini's chips, 32 were agency or consultant sites. On the pricing questions of cluster B, news and magazine lists became rare: 3 of Perplexity's 90 sources were of that type, all three in the same cell, and ChatGPT's 38 citations included none.
Next GEO Agency was mentioned by name in 12 of the 179 cells, but 8 of those 12 belong to the two questions that carried our brand name in the question itself; in questions that did not name us, the count stays at 4 cells. In a further 6 cells the answer cited nextgeoagency.com without writing our name: in 4 of 8 cells on ChatGPT and 2 of 8 on Gemini, all six on pricing questions. In the remaining 161 cells neither our name nor our site appeared. These figures were taken with a single account per engine and two runs each; they are not a ranking.
179
valid cells: 3 engines × 30 questions × 2 runs (180 planned)
12/179
cells where Next GEO Agency was mentioned by name
4/179
cells where we were mentioned although the question did not name us
6/179
cells where our site was cited without our name
Our own result: 12 mentions, 6 uncredited citations, absent in 161 cells
The agency running this index is part of the market it measures. That is why we give our own result first, counted with the same rule as every other agency. We keep three measures apart: a mention by name means "Next GEO Agency" appears in the answer text; an uncredited citation means the answer cites nextgeoagency.com without our name appearing in it; and both measures are split by whether our brand name was inside the question — being mentioned in a question that asks about us by name is not a discovery.
| Engine | Cells where we were mentioned by name | Cells where we got an uncredited citation | Where |
|---|---|---|---|
| Perplexity | A 0/16 · B 0/8 · C 0/12 · D 0/12 · E 5/8 · F 0/3 | A 0/16 · B 0/8 · C 0/12 · D 0/12 · E 0/8 · F 0/3 | By name: questions 25 and 26 in both runs, and question 28 in the second run — in each case the cited page is our About page. No uncredited citation at all: in none of the 56 cells with a recorded source panel was our site cited without our name. |
| ChatGPT | A 2/16 · B 0/8 · C 0/12 · D 0/12 · E 2/8 · F 1/4 | A 0/16 · B 4/8 · C 0/12 · D 0/12 · E 0/8 · F 0/4 | By name: question 7 in both runs (source: our GEO pricing page), question 26 in both runs, and the English question 29 in the second run (source: our English About page). Uncredited citations: questions 9, 10 and 11. |
| Gemini | A 0/16 · B 0/8 · C 0/12 · D 0/12 · E 2/8 · F 0/4 | A 0/16 · B 2/8 · C 0/12 · D 0/12 · E 0/8 · F 0/4 | By name: only on question 25, in both runs. On question 26 the whole answer was about a different company with a similar name. Uncredited citation: question 9, in both runs. |
Cells where we were mentioned in the two questions that named us (25 and 26): 8. A mention in those cells is not the engine finding us but confirming what was asked; in the same eight cells we also checked our company details — founding year, location, team size and service list came back correct.
Cells where we were mentioned in questions that did not name us: 4. Both runs of the affordable-agency question on ChatGPT, the second run of the Kahramanmaraş question on Perplexity, and the second run of the English agency question on ChatGPT. The last two did not repeat in the other run of the same question, so this is an unstable result rather than a position we have won.
Two of the four mentions in questions that did not name us came on the seventh question: "Can you recommend an affordable GEO agency for a small business?" In the first run ChatGPT listed us in one row of a price table at 20.000 TL a month; in the second run it named us as an example that publishes its prices, with a one-off audit at 12.000 TL and monthly work from 20.000 TL. In both runs the cited page was our GEO pricing page. The other two mentions came from two separate questions: in the second run of the question about agencies providing GEO or SEO services in Kahramanmaraş, Perplexity put us first in a list of eight providers and cited our About page; in the second run of the English agency question, ChatGPT named us in the fifth row of a ten-agency table and cited our English About page. Neither repeated in the other run of the same question. Across the whole index, 8 unique pages of ours were cited as sources; by contrast, in none of the 72 cells of clusters C and D did either our name or our site appear.
Uncredited citations: all six on pricing questions
Every uncredited citation sits in cluster B: in 6 of the 24 cells of the pricing questions the answer cited our site without writing our name. In the 155 cells of the other five clusters there is none. On ChatGPT the source address is visible; Gemini's source chip shows only a label, not the page address.
| Question | Engine and run | Our page cited as a source |
|---|---|---|
| 9. GEO hizmeti ne kadara mal olur? (How much does a GEO service cost?) | ChatGPT, run 2 | nextgeoagency.com/hizmetler/geo-fiyatlari (GEO pricing page, Turkish)nextgeoagency.com/blog/geo-fiyatlari-2026-maliyet-ve-paket-rehberi (article on GEO cost line items, Turkish) |
| 10. Bir GEO denetimi kaç para? (How much is a GEO audit?) | ChatGPT, run 2 | nextgeoagency.com/en/services/geo-pricing (GEO pricing page, English) |
| 11. GEO ajansları aylık ne kadar ücret alıyor? (How much do GEO agencies charge per month?) | ChatGPT, run 1 | nextgeoagency.com/hizmetler/geo-fiyatlari (GEO pricing page, Turkish) |
| 11. GEO ajansları aylık ne kadar ücret alıyor? (How much do GEO agencies charge per month?) | ChatGPT, run 2 | nextgeoagency.com/hizmetler/geo-fiyatlari (GEO pricing page, Turkish) |
| 9. GEO hizmeti ne kadara mal olur? (How much does a GEO service cost?) | Gemini, run 1 | Chip label "nextgeoagency.com"; page address not visible |
| 9. GEO hizmeti ne kadara mal olur? (How much does a GEO service cost?) | Gemini, run 2 | Chip label "Next GEO Agency"; page address not visible |
On ChatGPT, question 11 cited our Turkish GEO pricing page in both runs. On questions 9 and 10 our site was a source only in the second run; in the second run of question 10 the cited page was our English pricing page. On Gemini, question 9 tied a source chip to our site in both runs; the chip label was the domain name in the first run and the brand name in the second, but our name did not appear in the answer text.
This table is a record taken with a single account per engine and two runs each; it is not a ranking. An uncredited citation shows that part of the answer relied on our site; it does not show that the reader came to know us as an agency.
Which agencies did the engines mention?
The figure is the number of that engine's valid cells in which the agency's name appears in the answer text. How many times it appears within a cell does not change the figure; a name that appears only in the source list does not count as a mention. Tables are drawn only for the two clusters that produce names: choosing an agency (A, 16 cells per engine) and sector-specific choice (D, 12 cells per engine). In cluster A at least one agency name was given in 12 cells on Perplexity, 12 on ChatGPT and 11 on Gemini; in cluster D those figures were 11, 8 and 12 respectively.
Cluster A — agency-choice questions (16 cells per engine)
| Agency | Cells mentioned |
|---|---|
| Lein Digital | 10 / 16 |
| Sheltron | 9 / 16 |
| Crabs Media | 8 / 16 |
| Seobaz | 5 / 16 |
| Webtures | 5 / 16 |
| Zeo | 5 / 16 |
| Agency | Cells mentioned |
|---|---|
| OFG Medya | 7 / 16 |
| Magna Dijital | 6 / 16 |
| Adverpeak | 5 / 16 |
| ARGEO | 5 / 16 |
| Lein Digital | 5 / 16 |
| Agency | Cells mentioned |
|---|---|
| Crabs Media | 6 / 16 |
| Lein Digital | 6 / 16 |
| Zeo | 6 / 16 |
| Webtures | 5 / 16 |
| Clickzone | 4 / 16 |
Cluster D — sector-specific questions (12 cells per engine)
| Agency | Cells mentioned |
|---|---|
| Lein Digital | 5 / 12 |
| Vayes | 4 / 12 |
| Webtures | 4 / 12 |
| Zeo | 4 / 12 |
| Sherazi Marketing | 3 / 12 |
| Agency | Cells mentioned |
|---|---|
| Marmara Dijital Medya | 4 / 12 |
| Vayes | 3 / 12 |
| Kriko | 2 / 12 |
| Prosis Digital | 2 / 12 |
We draw no table for Gemini in cluster D: the highest name, PromptFront, was mentioned in 4 of 12 cells, no name after it exceeded 2 cells, and twelve agencies are tied at 2 cells. Cutting a top five would be an arbitrary choice.
Agencies with equal figures are listed alphabetically. The Perplexity table for cluster A has six rows because three agencies are tied at 5 cells; the ChatGPT table for cluster D has four rows because of a tie at 2 cells. The tables are a count; they contain no assessment of the agencies' service quality.
On the pricing questions of cluster B, answers mostly listed price bands without naming agencies. On Perplexity 2 of 8 cells named an agency, both on question 10; on ChatGPT and Gemini none of the 8 cells named one.
None of the 36 cells of cluster C named an agency; there is nothing to count there.
In cluster E the questions turn on a brand, a person and a city; most of the names come from a single Kahramanmaraş question, and a denominator of 24 cells is not enough for a ranking.
In cluster F the denominator is 3-4 cells per engine; a table built on that would mislead. The findings from the English questions are described in prose below.
Which sources were the answers built from?
Each engine displays sources differently, so the unit differs by engine: on Perplexity, the addresses in the source panel; on ChatGPT, the unique addresses of in-answer citations; on Gemini, the site names shown on source chips. In clusters A and B we split the sources into seven classes; classification is based on the address and page title, and whether each page is editorial or paid was not checked one by one. In clusters C, D, E and F only the number of cells showing sources and the total number of sources are given; the reason is stated below the tables.
| Source type | Perplexity | ChatGPT | Gemini |
|---|---|---|---|
| List, guide or blog post on an agency or consultant site | 65 | 14 | — |
| List on a news, magazine or blog publication site | 31 | 0 | 8 |
| Agency listing or directory platform | 9 | 3 | 0 |
| Service, pricing or company page of an agency or consultant | 19 | 54 | — |
| Agency or consultant site (page type not visible) | — | — | 32 |
| Foreign platform, agency or source | 6 | 5 | 4 |
| Unclear (the chip shows only a generic name) | 0 | 0 | 7 |
| Total | 130 | 76 | 51 |
| Cells showing sources | 13 / 16 | 12 / 16 | 12 / 16 |
| Source type | Perplexity | ChatGPT | Gemini |
|---|---|---|---|
| List, guide or blog post on an agency or consultant site | 45 | 3 | — |
| List on a news, magazine or blog publication site | 3 | 0 | 0 |
| Agency listing or directory platform | 0 | 0 | 0 |
| Service, pricing or company page of an agency or consultant | 19 | 17 | — |
| Agency or consultant site (page type not visible) | — | — | 4 |
| Foreign platform, agency or source | 23 | 18 | 2 |
| Unclear (the chip shows only a generic name) | 0 | 0 | 4 |
| Total | 90 | 38 | 10 |
| Cells showing sources | 8 / 8 | 8 / 8 | 3 / 8 |
| Source type | Perplexity | ChatGPT | Gemini |
|---|---|---|---|
| Total | 180 | 7 | 4 |
| Cells showing sources | 12 / 12 | 2 / 12 | 2 / 12 |
| Source type | Perplexity | ChatGPT | Gemini |
|---|---|---|---|
| Total | 169 | 44 | 54 |
| Cells showing sources | 12 / 12 | 10 / 12 | 12 / 12 |
| Source type | Perplexity | ChatGPT | Gemini |
|---|---|---|---|
| Total | 100 | 22 | 25 |
| Cells showing sources | 8 / 8 | 6 / 8 | 6 / 8 |
| Source type | Perplexity | ChatGPT | Gemini |
|---|---|---|---|
| Total | 46 | 24 | 18 |
| Cells showing sources | 3 / 3 | 4 / 4 | 4 / 4 |
We do not split sources into seven classes for clusters C, D, E and F. The address-based classifier written for A and B is measurably wrong on these records: in cluster C, 140 of Perplexity's 180 sources are marked as foreign in the raw records, while the classifier counts most of the same sources as agency content because their address contains "/blog", and reports only 4 as foreign.
The raw records' own class column is not uniform either: 84 of 184 files have one, and in cluster A only 1 does. Neither method gives a consistent result across a whole cluster, so we do not publish the breakdown. The number of cells showing sources and the total number of sources are a direct count and are given for all six clusters.
3 Perplexity cells in cluster A (the first run of questions 1, 2 and 4) are not in the tables because their source panel was not recorded in full. ChatGPT ran no web search on questions 3 and 6 in either run. On Perplexity, the free preview of advanced search was switched on in the interface during the second run of cluster B, and 2 answers showed 15 sources (the others 10); whether the difference comes from that change cannot be separated. In cluster F the Perplexity denominator is 3 cells; the second run of question 30 could not be taken. Gemini's chips show no address, so service pages and blog posts could not be separated; "—" marks where that split was not possible. During the measurement we found that Gemini adds source chips to the page only once they scroll into the visible part of the screen. In cluster B, the "0 chips" reading for the second run of question 11 was taken before that reading method was corrected and cannot be verified; it should not be read as an answer without sources. The same risk applies to Gemini's other cells read as 0 chips; treat the Gemini columns as a lower bound.
A list labelled as advertising was cited
On the fourth question ("Which agency should I work with for AI search optimization?"), Perplexity cited haber32.com.tr in both runs. The address recorded in the second run is a list titled "Türkiye'nin En İyi 5 GEO Ajansı (2026)" (Türkiye's top five GEO agencies); the site publishes it on its own page under the category "TANITIM YAZILARI (REKLAM)", which translates as "promotional articles (advertising)". The address was not recorded in the first run; both GEO agency lists we found on the site carry the same category label. We read the label by opening the page on 16 September 2026.
We did not verify at whose request the list was published. Whether the habergazetesi.com.tr list cited in the same answer is advertising could not be verified. Across the whole index the same address appeared only in Perplexity cells: besides the fourth question, on question 20 in cluster D and question 29 in cluster F. haber32.com.tr was never a source in the valid cells of ChatGPT or Gemini.
On pricing questions, sources turned to agency pages
On the four pricing questions of cluster B, list and news sources became rare. Of the 90 sources Perplexity showed across 8 cells, 45 were price guides, blog posts or lists on agency and consultant sites, 19 were agencies' service or pricing pages, and 23 were foreign sites. Lists on news and magazine sites made up 3 of the 90 sources, and all three were in the second run of question 11.
Of ChatGPT's 38 citations across 8 cells, 17 went to agencies' service or pricing pages, 3 to agency blog posts or guides and 18 to foreign sites; there was no citation of a news site or listing platform. On the pricing questions Gemini showed source chips in 3 of 8 cells, with 10 site names on those chips in total. In 2 cells it declined to answer: the second run of questions 10 and 12.
On the consulting question, sources shifted to service pages
The eighth question of cluster A ("Who are the people I can get GEO consulting from?") changed the source mix. On Perplexity, 16 of the 20 sources across its two runs were service pages of agencies or consultants; on the same engine, none of the 20 sources in the fully recorded runs of the first and second questions was a service page.
On ChatGPT, 13 of the eighth question's 18 citations went to service pages; across the 17 citations of the first and second questions that figure was 3. Gemini's 7 chips on this question included no news or list site. Magna Dijital was mentioned in 6 of the eighth question's 6 cells.
Findings cluster by cluster
Cluster C: problem questions do not produce agency recommendations
None of the 36 cells of the problem-awareness questions named an agency; all three engines gave a to-do list instead of recommending a provider. Visibility here can only be measured as a citation, and in none of the 36 cells were we cited. We have articles on the site that match these questions; none of them became a source.
The engines behaved differently too: ChatGPT ran no web search in five of the six questions in either run, searching only on the question about being cited on Perplexity, where every citation went to that platform's own documentation. Perplexity searched on every question, but its source pool changed almost entirely from run to run; in all six questions the two runs shared only 0-1 addresses. Of Gemini's eleven verified readings in this cluster, nine had no sources at all.
Cluster D: sector questions produce names, but not ours
In 31 of the 36 cells of the sector-choice questions at least one agency name appeared, and in none of them were we there. Five of the questions have a matching sector page on our site — dental and aesthetic clinics, e-commerce, real estate offices, health tourism, accountants — and none of those pages was a source in either run.
The only shared feature measured in the records is having opened a separate page for the sector: on the law-firm question every one of ChatGPT's citations was a page dedicated to the legal sector, and all four agencies that overlapped across engines had a separate page for that sector. The stability of the name pool differed by engine; on some questions Gemini returned the same names in both runs, while in the "recommend an agency" cells of ChatGPT and Perplexity the two runs could share 0 names.
The location signal was measured too: Perplexity and ChatGPT referred to Kahramanmaraş in some answers, and in one cell Perplexity explicitly suggested starting with an agency in another city; Gemini did not use the location in either run.
Cluster E: our details come back correct, but the trust question finds no third-party source
The counting rule is different in this cluster. Our brand name is inside questions 25 and 26, so finding the name in the answer body proves nothing — the name appears even when the engine says it does not know the brand. In cluster E a mention was therefore taken from the record's own verdict; in the other five clusters the published rule was applied, namely the name appearing in the answer text.
On "What is Next GEO Agency?" Perplexity and Gemini pulled our pages and returned our details correctly; ChatGPT asked for context in both runs without running a web search. On "Is it a trustworthy agency?" the engines looked for third-party sources and the number of third-party sources written about us came out as 0; ChatGPT and Perplexity said so explicitly, while Gemini described a different company with a similar name in both runs.
Our founder's name resolved to other people on all three engines, and the result was the same in all four cells. On the question about our own city, 15 providers were named across the four ChatGPT and Gemini cells and none of them was us; ChatGPT returned 7 map businesses on the same question and we were not among the cards. Perplexity, in its second run, put us first in a list of eight providers — in the first run of the same question we were absent.
Cluster F: on the English side, the door is the query that names the country
The two English questions have 11 valid cells; the second run of Perplexity's question 30 could not be taken. On the question that names the country we were mentioned by name in one cell: in the second run ChatGPT placed us in the fifth row of a ten-agency table and cited our English About page. Five minutes earlier the same question had listed 14 agencies and we were not on that list.
On the question that does not name the country, none of our pages was pulled. Same engine, same day: on the question naming the country half of the search pool was Turkish lists and directories, while on the question without the country more than half of the pool was the platform's own documentation; on that question Gemini named no agency at all in either run and described roles instead.
One source class recurs in the English answers: agency directories you enter by applying. The Semrush Agency Directory was a source in all four of the answers ChatGPT and Gemini gave to question 29; Clutch was cited in two Gemini cells and GoodFirms in two ChatGPT cells. Whether we are listed in those directories cannot be told from this measurement; here we only record that they appeared as sources.
Method
A fixed question set was put to ChatGPT, Gemini and Perplexity by hand in the browser, and each answer's text, source list and measurement note were recorded.
The eight questions of cluster A: choosing an agency
- Türkiye'de GEO hizmeti veren ajanslar hangileri? (Which agencies provide GEO services in Türkiye?)
- En iyi GEO ajansı hangisi? (Which is the best GEO agency?)
- GEO ajansı seçerken nelere bakmalıyım? (What should I look for when choosing a GEO agency?)
- Yapay zeka arama optimizasyonu için hangi ajansla çalışmalıyım? (Which agency should I work with for AI search optimization?)
- Generative Engine Optimization hizmeti veren Türk ajanslar var mı? (Are there Turkish agencies that provide Generative Engine Optimization?)
- SEO ajansı mı GEO ajansı mı seçmeliyim? (Should I choose an SEO agency or a GEO agency?)
- Küçük bir işletme için uygun fiyatlı GEO ajansı önerir misin? (Can you recommend an affordable GEO agency for a small business?)
- GEO danışmanlığı alabileceğim kişiler kimler? (Who are the people I can get GEO consulting from?)
The four questions of cluster B: price and scope
- GEO hizmeti ne kadara mal olur? (How much does a GEO service cost?)
- Bir GEO denetimi kaç para? (How much is a GEO audit?)
- GEO ajansları aylık ne kadar ücret alıyor? (How much do GEO agencies charge per month?)
- GEO hizmeti fiyatına neler dahil olmalı? (What should be included in the price of a GEO service?)
The six questions of cluster C: problem awareness
- ChatGPT'de markam hiç çıkmıyor, ne yapmalıyım? (My brand never shows up in ChatGPT, what should I do?)
- Yapay zeka sohbetlerinde şirketimin adının geçmesini nasıl sağlarım? (How do I get my company mentioned in AI chats?)
- Google'da AI Overviews çıkınca trafiğim düştü, ne yapabilirim? (My traffic dropped when AI Overviews appeared on Google, what can I do?)
- Sitemin yapay zeka motorlarında görünüp görünmediğini nasıl ölçerim? (How do I measure whether my site shows up in AI engines?)
- Perplexity'de kaynak olarak gösterilmek için ne gerekir? (What does it take to be cited as a source on Perplexity?)
- SEO yaptırıyorum ama yapay zekada çıkmıyorum, sebebi ne olabilir? (I pay for SEO but I do not show up in AI, what could be the reason?)
The six questions of cluster D: sector-specific choice
- Diş kliniği için yapay zeka görünürlüğü nasıl sağlanır, kimden destek alınır? (How is AI visibility achieved for a dental clinic, and who provides support?)
- E-ticaret sitesi için GEO hizmeti veren ajans önerir misin? (Can you recommend an agency that provides GEO services for an e-commerce site?)
- Hukuk bürosu için dijital görünürlük danışmanlığı kim veriyor? (Who provides digital visibility consulting for a law firm?)
- Emlak ofisi için yapay zeka arama optimizasyonu yapan var mı? (Is there anyone doing AI search optimization for a real estate office?)
- Sağlık turizminde yapay zekada görünürlük için kiminle çalışılır? (Who do you work with for AI visibility in health tourism?)
- Mali müşavirler için dijital görünürlük hizmeti veren ajanslar hangileri? (Which agencies provide digital visibility services for accountants?)
The four questions of cluster E: brand and entity
- Next GEO Agency nedir? (What is Next GEO Agency?)
- Next GEO Agency güvenilir bir ajans mı? (Is Next GEO Agency a trustworthy agency?)
- Osman Kavak kimdir? (Who is Osman Kavak?)
- Kahramanmaraş'ta GEO veya SEO hizmeti veren ajans var mı? (Are there agencies providing GEO or SEO services in Kahramanmaraş?)
The two questions of cluster F: English-language queries (asked in English, verbatim)
- Which agencies provide GEO (generative engine optimization) services in Turkey?
- Who should I hire to make my brand visible in ChatGPT answers?
Of the 180 planned cells, 179 were taken. The one missing cell is the second run of Perplexity's question 30. It first hit the free search limit; when we asked again after the limit reset, the answer said "based on your earlier research" and drew three agency names from the history of questions we had asked in the same account — none of the three appeared in its own source list. Personalised answers are not counted in this index, so the cell stays in the record but not in the figures. Every ratio on this page is out of 179 cells, and no figure depends on that cell.
Engines, modes and date
- Perplexity: with the agency's own logged-in account (free plan), in the default search mode, each question in a new conversation.
- ChatGPT: with the agency's own logged-in account, in a temporary chat with "Kişiselleştirilmemiş" (not personalised) selected in the header menu; the selection was checked before sending each question. The model picker and the thinking setting were left untouched.
- Gemini: with the agency's own logged-in account, in a temporary chat. The model picker was left untouched; it showed "Flash" in the first run of cluster A and "Pro" in every other measurement.
- Date: both runs of cluster A and the first run of cluster B were taken on 16 September 2026; the second run of cluster B and clusters C and D on 17 September 2026; clusters E and F on 17 and 18 September 2026. Clusters A to E were asked in Turkish and the questions of cluster F in English; apart from questions 25 and 26, none contains a brand name, and all are worded the way a buyer would ask.
What we counted and what we did not
- A cell is one combination of engine, question and run. The question set consists of 30 questions in six clusters: A 8 questions, B 4, C 6, D 6, E 4, F 2. Of the 180 planned cells, 179 were taken; per engine, Perplexity 59 cells, ChatGPT 60 and Gemini 60.
- An agency counts as mentioned in a cell if its name appears in the answer text (including tables, lists and map cards). A name that appears only on a source chip or in the source list does not count as a mention. The "cells naming an agency" figure in the tables is the number of cells in which at least one agency or provider name appears in the answer; our own name is included in that figure.
- Uncredited citations were counted separately: if the name Next GEO Agency does not appear in the answer text but one of the answer's sources is nextgeoagency.com, the cell counts as an uncredited citation. On ChatGPT we checked the addresses of in-answer citations, on Gemini the label of the source chip, on Perplexity the addresses in the source panel. Addresses that appeared in ChatGPT's search results without being cited in the answer were not counted.
- Cells in which the engine declined to answer were counted as valid; a refusal is also the result of that run. In cluster B, Gemini declined in 2 cells; in the second run of question 12 the answer was withdrawn while it was being written and replaced with a refusal.
- The mention rule cannot be applied in cluster E: our brand name is inside questions 25 and 26, so finding the name in the answer body is not evidence of a mention. In that cluster a mention was taken from the record's own verdict, and the 8 cells belonging to those two questions were counted separately. The "cells naming an agency" figure must also be read differently in cluster E, because the subject of the question is already a brand.
- In total, 9 personalised measurements were not counted: 8 ChatGPT measurements taken in a normal chat and 1 Gemini measurement taken in a normal chat, all in cluster A. In a normal chat, ChatGPT answered knowing that the account holder runs a GEO agency. In a normal chat, Gemini put Next GEO Agency at the top of its list on the first question without citing a source; when the same question was asked in a temporary chat, it did not mention us. That is why the valid measurement on both engines is the temporary chat. The uncounted measurements remain on record.
- For 3 Perplexity cells (the first run of questions 1, 2 and 4 in cluster A), a measurement note was kept rather than the full answer text; agency names in those cells were counted from the note. In the other 176 cells the count was made from the raw answer text.
Variables measured
- Number of cells in which an agency name appears in the answer text
- Number of sources by source type (clusters A and B) together with the number of cells showing sources and the total number of sources (all six clusters)
- Number of cells in which Next GEO Agency was mentioned by name, split by whether the brand name was inside the question
- Number of cells in which nextgeoagency.com was cited without the name Next GEO Agency
Limits
- The measurements were taken with logged-in accounts located in Kahramanmaraş. Temporary chat and "not personalised" mode leave out memory and custom instructions but not the location signal: some answers recommended businesses in Kahramanmaraş or referred to the location.
- Each engine was measured with a single account. A different account, a different city or a logged-out user may see different answers.
- Answers change from run to run; the agencies mentioned and the sources cited in two runs of the same question may not overlap. That is why each cell was measured twice and the figures are given as counts of cells rather than a single answer. Half of the mentions in questions that did not carry our brand name failed to repeat in the other run of the same question; two runs are not enough to prove a trend.
- Perplexity was used on the free plan. The free search limit appeared several times during the measurement; in cluster A the second run of the seventh question was taken after the warning, and the second run of question 30 could never be taken in a countable form. The ratios are given out of 179 valid cells.
- The measurement affected the instrument it was measuring. Perplexity's 59 cells were asked one after another in the same account, and on the last question of the set the engine said "based on your earlier research" and produced three agency names from that history. That cell was therefore not counted, but it shows that the next measurement round needs an account kept solely for measurement. Note: even with that memory in play, our name did not appear in the answer.
- On Gemini the two runs of cluster A were taken under different model labels (Flash and Pro); whether the differences between the runs come from the model or from chance cannot be separated with this data.
- Gemini's source columns are a lower bound: source chips are added to the page only once they scroll into the visible part of the screen, so they may have been read as fewer than they are.
- The eight cells in which our brand name was inside the question were counted separately; a mention in those cells is the engine confirming what was asked, not finding us. Without that split the figure 12 reads stronger than it is.
- No source-class breakdown is published for clusters C, D, E and F; for those clusters only the number of cells showing sources and the total number of sources are given.
- Google AI Overviews is not included in this measurement.
- This index does not rank the service quality of agencies. It only counts which names and sources appeared in the answers to specific questions on specific days.
Raw data
The raw answer text, source list and measurement note of every cell are kept. We do not publish the raw files on this page; readers who want to check where a figure comes from can have the raw text of the relevant cells on request. To ask, use our contact page.
Next
All six clusters of the question set have now been measured. 179 of the 180 planned cells are valid; the page will be updated again once the missing cell is taken. The figures are recalculated from the raw records only, and every number on this page is checked against those records by an automated test.
- A Perplexity account kept solely for measurement, with no history; only then can the missing cell be taken in a countable form
- Adding Google AI Overviews to the measurement
- Repeating the same question set in October 2026 and publishing the difference between the two months
- Re-measuring the two unstable cells: the Kahramanmaraş question and the English agency question