On this page
On November 2, 2026 we asked ChatGPT, Perplexity and Google 20 questions that a buyer of agency services might ask, in 8 countries, and collected 1,115 answers. Google showed an AI Overview on 146 of 155 results pages, and those overviews cited YouTube, Reddit and LinkedIn more often than any agency site. Perplexity drew on a small pool of 265 domains, while ChatGPT used 678 and gave a different list of agencies almost every time: three runs of the same question shared only 17 % of names. Round-up pages of the “best agencies” kind make up a large share of what the engines cite.
AOE Marketing is an SEO and AI search agency in Almaty, Kazakhstan, a market this study did not measure. We sell in the markets we did measure, so we have an interest in the topic. We report what the engines returned, we left ourselves out of the findings, and nothing on this page is a quality ranking.
How we measured
- Questions. 20 questions in English: 6 about AI search optimization (GEO), 6 about SEO, 5 about white label work and 3 about trust. Examples: “best generative engine optimization agency”, “affordable SEO agency with transparent pricing”, “signs of a bad SEO agency”.
- Markets. United States, United Kingdom, Australia, Canada, UAE, Singapore, India and Germany. The same English wording in every country.
- Engines. All data came through DataForSEO: ChatGPT with web search and the country set by location, Perplexity sonar with a country setting, and Google results with the AI Overview loaded.
- Runs. AI answers change from one try to the next, so ChatGPT and Perplexity got each question three times in each market: 480 answers per engine. Google got each question once per market. We received 155 of 160 pages: DataForSEO did not return 5 pages (1 each for Australia, Singapore and Germany, 2 for India) after three attempts.
- Cost. The full data pull cost $5.31.
| Engine | Answers | Answers with sources | Source domains per answer | Distinct domains |
|---|---|---|---|---|
| ChatGPT | 480 | 371 | 5.5 | 678 |
| Perplexity | 480 | 480 | 9 | 265 |
| Google AI Overviews | 155 pages | 146 | 6.7 | 507 |
Findings
1. Google showed an AI Overview on almost every agency search in our sample
An overview appeared on 146 of 155 results pages, or 94 %. That includes commercial searches such as “best generative engine optimization agency”, not only “how to” questions. Two questions often had no overview. “How much does generative engine optimization cost per month” got one in 3 of 8 markets, and “affordable SEO agency with transparent pricing” in 5 of 8. Every other question got an overview in all of its markets, or in all but one.
2. Google’s overviews lean on video, forums and social sites
Of the 146 overviews, 46 cited YouTube, 36 cited Reddit, 27 LinkedIn, 15 Quora and 13 Facebook. The agency site cited most often was fatjoe.com, in 14 overviews. In our sample, Google’s overviews pointed to public platforms far more often than to any agency’s own pages.
3. Perplexity reads from a short list, ChatGPT reads widely
Every Perplexity answer had sources, from 9 domains on average, yet across 480 answers they came from only 265 domains. The same few sites came up again and again: thestacc.com in 84 answers, seo.com in 72, tripledart.com and sona.com in 70 each, webfx.com in 61. ChatGPT cited sources in 371 of 480 answers, from 5.5 domains on average, spread over 678 domains. Its most cited source was developers.google.com, in 84 answers.
4. Round-up pages are a large share of what gets cited
We took the 41 domains the engines cited most often (agencies, tool vendors, directories and publishers) and looked at which of their pages were linked. 46 % of those links went to pages whose address contains words such as “best”, “top”, “agencies” or “companies”, in other words round-ups. The rest went to service pages, pricing pages and home pages. We did not test whether the agencies named in an answer wrote the round-ups it cited.
5. ChatGPT’s agency lists change from run to run
On the 14 questions where a buyer asks for a provider, ChatGPT named 4.8 companies per answer on average and 744 different names across 336 answers, including a few generic labels and spelling variants. Three runs of the same question in the same market shared only 17 % of names (Jaccard index). One screenshot of one answer tells you very little.
6. Country differences are small next to run-to-run noise
For each question we pooled the names from all three runs in a market and compared that pool with the US pool for the same question (average Jaccard index).
| Market | Overlap with the US list |
|---|---|
| United Kingdom | 9 % |
| Australia | 17 % |
| Canada | 21 % |
| UAE | 20 % |
| Singapore | 18 % |
| India | 16 % |
| Germany | 16 % |
Repeating the same question in the same market overlaps about as little (17 %), so the country effect is hard to separate from chance. Only the UK, at 9 %, clearly stands apart.
7. ChatGPT searches with the year, the country and the price
ChatGPT turned the questions into 694 web searches of its own, 1.4 per answer. It added “2026” to 44 % of them, a country or city name to 23 %, price words to 23 %, “best” or “top” to 22 %, “case studies” to 20 % and review, rating or ranking words to 17 %. It never searched Reddit or Clutch by name.
A separate check on the same day is worth a look. In a tracking set we plan to repeat monthly (5 markets, 5 runs each) we asked “GEO agency with transparent pricing”. ChatGPT named RIFY Agency in 10 of 25 answers, 009 Agency in 6 and AEO Mint in 4. These were not the names that lead the other questions. 009.agency lists $3,000 per month and AEO Mint starts at $199 per month; RIFY’s home page has a Pricing section (we checked on November 2). We did not check the sites of the other agencies named, so we cannot say that price pages set these three apart. This is one question: treat it as a lead to test, not as proof.
Who gets named most often in our sample
The table counts answers to the 14 provider questions that named each company: 336 ChatGPT answers, 336 Perplexity answers and 104 Google AI Overviews. ChatGPT returns its own list of named companies. For Perplexity and Google we searched the answer text for the same names. We left out generic labels such as “SEO and growth marketing agency”. Frequency here is not a judgment of quality. It reflects our questions, our date and the sources each engine read.
| Company | ChatGPT (of 336) | Perplexity (of 336) | Google AI Overviews (of 104) |
|---|---|---|---|
| Omniscient Digital | 39 | 46 | 6 |
| FATJOE | 34 | 10 | 16 |
| SEOReseller | 30 | 39 | 5 |
| Loganix | 27 | 22 | 1 |
| First Page Sage | 26 | 23 | 6 |
| The HOTH | 24 | 28 | 9 |
| Semify | 23 | 20 | 2 |
| DashClicks | 21 | 26 | 10 |
| Animalz | 20 | 0 | 4 |
| Graphite | 18 | 0 | 1 |
| SEO Discovery | 18 | 23 | 3 |
| Siege Media | 17 | 24 | 2 |
No company led in all three engines. FATJOE was named most in Google overviews but in only 10 Perplexity answers. Animalz and Graphite appeared in ChatGPT and not once in Perplexity. Neither AOE Marketing nor aoe-marketing.com appeared in any of these 1,115 answers. The site was not yet in Google’s index at the time.
What this means if you want to be named
None of these steps guarantees a mention. They follow from what the engines did in our sample.
- Measure several times and in your own country
With 17 % overlap between runs, a single answer can mislead you, and the UK result shows that location can matter too.
Read more
Our AI visibility guide shows how to set up a repeatable test.
- Work on the platforms Google’s overviews cite
Video on YouTube, posts on LinkedIn and useful answers on Reddit and Quora under a real name.
Read more
In our data these sources showed up far more often than any agency site.
- Find the pages Perplexity keeps reading for your category
Its pool is small, so a few round-up authors matter a lot. Write to them with a concrete reason their readers would care about you.
- Publish an honest comparison of your own
Almost half the links to agency sites went to list articles. If you write one, say plainly that you are the author and one of the options.
- Put the year, the country, prices and case studies on your pages as text
These are the words ChatGPT adds to its searches. Change the year only when you have actually updated the page.
Limits of this study
- It is a snapshot from November 2, 2026. Answers depend on time, account and wording.
- We collected answers through an API with country settings. It is not a record of real users’ screens.
- We chose the 20 questions, and they lean toward GEO and white label work. They do not cover the whole agency market.
- ChatGPT’s list of named companies includes generic labels; three of them were in its top 25, and we did not use them. In Perplexity and Google, names were found by text search, so spelling variants could be missed.
- We marked a page as a list article by words in its address (best, top, agencies, companies), and we labelled 41 domains, not all of them.
- Three runs per question show what was cited, not why. We make no claims about the reasons an engine chose a company.
- We are an agency in this niche ourselves.
We plan to repeat the study with the same questions in early February 2027.
Questions
Is this a ranking of the best agencies?
No. It counts how often each engine named a company for our questions on one date. A company can be named often because it publishes many round-ups or because a popular article lists it, which says nothing about the quality of its work.
Why do ChatGPT and Perplexity name different agencies?
They read different sources. Perplexity kept returning to about 265 domains, while ChatGPT spread over 678 and rewrote each question into its own searches. Different sources may explain different names, but we did not test that.
Can I quote these numbers?
Yes. Please give the date of the data, November 2, 2026, and link to this page so readers can see the method and the limits.
Can you check this for my market?
Yes. Our free AI visibility check asks 10 buyer questions once each and sends you a table of who gets named. The AI visibility audit asks 30 questions in ChatGPT, Perplexity and Google AI Overviews, three runs each, and ends with a ranked list of fixes.