The short answerYes, by a wide margin. Across 1,677 answers from ChatGPT, Perplexity and Google AI Overviews (20 September to 1 November 2026), best-of questions named at least one of 51 agencies tracked in Peec AI in 81.9% of answers. Questions that described a problem, such as 'We publish a lot but nothing gets cited by AI', did so in 49.6%.
| Engine | Best-of questions: answers naming a tracked agency | Problem questions: answers naming a tracked agency | Tracked agencies per answer (best-of vs problem) |
|---|---|---|---|
| ChatGPT | 95.1% | 75.8% | 4.6 vs 2.2 |
| Perplexity | 69.2% | 30.2% | 2.6 vs 0.6 |
| Google AI Overviews | 81.4% | 42.8% | 2.6 vs 0.9 |
| All three | 81.9% (845 of 1,032) | 49.6% (320 of 645) | 3.3 vs 1.2 |
Ask an engine the way you would describe the pain to a colleague ("our brand is invisible in ChatGPT, who can fix that?") and you get a different answer from "best GEO agencies". The agencies that dominate the best-of answers lose most of their lead, the answers cite different pages, and Perplexity often names no firm at all.
How we ran the study
We track buyer questions for LoudFace in Peec AI, which runs the same question on ChatGPT, Perplexity and Google AI Overviews each day and records each answer, the sources it cites, the searches the engine ran and the brands it names.
For this study we took 13 tracked agency questions and split them by how the buyer asks.
8 best-of questions. "Best GEO agencies", "Best AI search agencies", "Best AI visibility agencies", "Best agency for getting cited in ChatGPT", "Best agency for ChatGPT and Perplexity citations", "Best agency for getting mentioned in Perplexity and ChatGPT", "Which agencies will help us show up in ChatGPT answers?" and "I'm looking for an SEO agency that can help us get more recommendations from AI. What are some of the best ones?"
5 problem questions. "We publish a lot but nothing gets cited by AI. Which agency can tell us why and fix it?", "Our brand is invisible in Perplexity and ChatGPT. Who can help a B2B SaaS fix that?", "I need an agency to audit why our SaaS never appears in AI search answers.", "We get AI referral traffic but it doesn't convert. Which agency can fix that?" and "We want to be cited in AI answers for 'best software' in our category. What kind of agency does that, and who is good?"
The window runs 43 days, 20 September to 1 November 2026, on all three engines. 20 September is the first day all 13 questions have answers. Peec ran some questions twice that day, so we kept one answer per question, engine and day and dropped 18 repeats. That leaves 43 answers per question per engine, 129 per question and 1,677 in all: 559 per engine, 1,032 best-of answers and 645 problem answers.
What counts as named. We use the same unit as Peec's visibility metric, which "measures the percentage of AI responses that mention your brand": an agency is named when Peec's brand detection lists it in the answer. Peec tracks 53 brands in our project, mostly agencies plus a few Webflow and content firms and one individual consultant. We left out two: LoudFace, because we publish this, and GEO Agency, because its name is the phrase the questions themselves use, so the detector cannot tell the firm from the words. That leaves 51, which we call tracked agencies.
The tracked set is not the whole market. So we also ran a wider count: an answer names an agency if Peec detects a tracked agency, or if a bold name or link label in the answer matches a list of 147 firm-name forms (the tracked agencies' names with common spellings, plus firms the answers name that Peec does not track). By that count, 13.3% of best-of answers and 40.9% of problem answers named no agency we could find. Treat those as upper bounds, since a firm missing from the list is not counted.
Best-of questions name a tracked agency 81.9% of the time; problem questions, 49.6%
The gap shows up on every engine, and answer length does not explain it. Problem answers were as long as best-of answers or longer on every engine: the median ChatGPT answer ran 567 words on a problem question against 523.5 on a best-of question. They named fewer agencies all the same. The median problem answer named zero tracked agencies, against three for a best-of answer.
The tracked firms stayed in the problem answers but spread out. Problem answers named 38 different tracked agencies over the 43 days, best-of answers 37. No single agency owned the problem answers the way a few own the best-of answers.
Perplexity named no tracked agency in 69.8% of problem answers
Perplexity is where the phrasing matters most. On best-of questions it named a tracked agency in 69.2% of answers, even though its median best-of answer ran 53 words, the shortest of any engine and phrasing. On problem questions that fell to 30.2%. With the wider count, 65.1% of its problem answers named no agency we could find.
ChatGPT held up best. It named a tracked agency in 75.8% of problem answers, and the wider count left 19.1% with no agency at all. Google AI Overviews sat between the two: 42.8% named a tracked agency, and 38.6% named no agency by the wider count.
The best-of leaders lose most of their lead
Share of answers naming each agency, top ten in each group:
| # | Best-of questions (1,032 answers) | Share | Problem questions (645 answers) | Share |
|---|---|---|---|---|
| 1 | iPullRank | 57.9% | Omniscient | 18.3% |
| 2 | Omniscient | 42.5% | iPullRank | 15.2% |
| 3 | Siege Media | 40.1% | Derivatex | 10.5% |
| 4 | First Page Sage | 33.9% | Siege Media | 8.7% |
| 5 | NoGood | 23.4% | First Page Sage | 6.2% |
| 6 | Directive Consulting | 21.4% | Animalz | 4.5% |
| 7 | Animalz | 14.4% | Skale | 4.3% |
| 8 | Graphite | 12.8% | Veza Digital | 4.3% |
| 9 | Optimist | 12.7% | Optimist | 4.2% |
| 10 | Onely | 12.6% | Directive Consulting | 4.0% |
| 10 (tie) | Omnius | 4.0% |
Two pairs tie in the problem column: Skale and Veza Digital at 28 answers each, and Directive Consulting and Omnius at 26 each. By text match, Thrive would be named in 12.1% of best-of answers, just outside the top ten (Peec's detector records it far less often; see the limitations).
Seven agencies sit in both top tens: iPullRank, Omniscient, Siege Media, First Page Sage, Directive Consulting, Animalz and Optimist. The problem answers pick from the same pool, far less often. iPullRank goes from 57.9% to 15.2%. Siege Media goes from 40.1% to 8.7%.
Of the 320 problem answers that named a tracked agency, 50.3% put first an agency that also appeared in that engine's best-of answers on the same day.
Results question by question
Each question has 129 answers, 43 per engine. "None" is the share of answers that named no tracked agency.
| Question | Group | None | Agencies per answer | Most named (share of answers) |
|---|---|---|---|---|
| Best GEO agencies | Best-of | 4.7% | 4.8 | Siege Media 83.7%, iPullRank 73.6%, Omniscient 68.2% |
| Best AI visibility agencies | Best-of | 5.4% | 4.5 | iPullRank 72.1%, Omniscient 65.1%, NoGood 58.1% |
| I'm looking for an SEO agency that can help us get more recommendations from AI... | Best-of | 10.9% | 4.5 | iPullRank 78.3%, Omniscient 68.2%, Siege Media 59.7% |
| Best AI search agencies | Best-of | 5.4% | 4.1 | iPullRank 79.8%, NoGood 56.6%, Siege Media 52.7% |
| Best agency for getting mentioned in Perplexity and ChatGPT | Best-of | 10.9% | 2.7 | iPullRank 42.6%, Omniscient 34.1%, First Page Sage and Siege Media 26.4% each |
| Best agency for getting cited in ChatGPT | Best-of | 17.8% | 2.2 | iPullRank 61.2%, First Page Sage 33.3%, Omniscient 28.7% |
| Which agencies will help us show up in ChatGPT answers? | Best-of | 52.7% | 1.8 | iPullRank 36.4%, Siege Media 25.6%, Omniscient 24.8% |
| Best agency for ChatGPT and Perplexity citations | Best-of | 37.2% | 1.6 | iPullRank 19.4%, Omniscient 17.1%, Siege Media 16.3% |
| We want to be cited in AI answers for 'best software' in our category... | Problem | 14.7% | 2.7 | Omniscient 54.3%, iPullRank 45.0%, Siege Media 27.9% |
| Our brand is invisible in Perplexity and ChatGPT... | Problem | 50.4% | 1.3 | Derivatex 32.6%, Omniscient and Breaking B2B 13.2% each |
| I need an agency to audit why our SaaS never appears in AI search answers. | Problem | 44.2% | 1.0 | Derivatex 12.4%, Omnius and iPullRank 10.9% each |
| We get AI referral traffic but it doesn't convert... | Problem | 66.7% | 0.6 | Optimist 10.1%, Omniscient 9.3%, NoGood 7.0% |
| We publish a lot but nothing gets cited by AI... | Problem | 76.0% | 0.5 | iPullRank 11.6%, then Siege Media, Omniscient, Animalz and Percepture 4.7% each |
The "best software" question describes a problem but ends with "who is good?". It behaves like a best-of question: 14.7% of its answers named no tracked agency, and Omniscient was named in 54.3%. All five problem questions ask for a firm. This one is the only problem question that asks who is good, and the only one with the word "best" in it. Its ChatGPT searches contained "best" or "top" 32.1% of the time, inside the best-of range of 16.7% to 55.4%, against 4.6% to 10.9% for the other problem questions. Our read is that those features, or another difference between the questions, pulled its answers toward best-of answers. With a single question, the data cannot tell what did it.
The plainest best-of question, "Best GEO agencies", is the most predictable answer in the set. Only 4.7% of its answers named no tracked agency, and Siege Media appeared in 83.7%, iPullRank in 73.6% and Omniscient in 68.2%.
What do the engines search for and cite under each phrasing?
Google says that AI Overviews and AI Mode may use a "query fan-out" technique, which it describes as "issuing multiple related searches across subtopics and data sources". Our read is that the words of a question shape which searches run. We wrote more on this in fan-out queries.
Peec records the searches ChatGPT ran for each answer, so for ChatGPT we can see what it looked for.
- Best-of questions: 626 searches across 344 ChatGPT answers, of which 237 ran at least one search. 38.8% contained "best" or "top". None contained fix, audit, why, not cited, invisible or diagnose. One of them: "best Generative Engine Optimization GEO agencies 2026".
- Problem questions: 341 searches across 215 ChatGPT answers, of which 156 ran at least one search. 13.8% contained "best" or "top", and 24.3% contained one of those problem words. One of them: "agency diagnose why brand content is not cited by AI search engines generative engine optimization GEO agency 2026".
Both kinds of search usually include the word agency (81.0% and 90.3%). What changes is whether ChatGPT goes looking for a ranking or for someone who fixes the problem. Perplexity recorded 61 searches in all (42 best-of, 19 problem) and AI Overviews none, so this view covers ChatGPT only.
Across all three engines, problem answers also cited far fewer ranked lists. We counted every cited URL whose address reads like a ranked list (it contains best, top, -agencies, ranked or alternatives):
| Engine | Best-of questions: cited URLs that look like a ranked list | Problem questions |
|---|---|---|
| ChatGPT | 54.7% | 18.1% |
| Perplexity | 71.4% | 27.9% |
| Google AI Overviews | 51.6% | 18.3% |
| All three | 57.0% of 4,994 | 20.3% of 3,158 |
On best-of questions the most-cited page was a ranked GEO agency list on thedigitalelevator.com, cited in 130 of 1,032 answers (12.6%). On problem questions the most-cited page was an agency's GEO service page on verticality.co, cited in 46 of 645 (7.1%). Six of the ten most-cited problem pages were agencies' own service, use-case or audit pages (verticality.co, growtika.com, percepture.com, blennd.com, search.agency and derivatex.agency). The other four were a semrush.com blog post, an AI search audit tools page on growpredictably.com, a Reddit thread and a LinkedIn post. None was a ranked agency list. By domain, LinkedIn, YouTube and Reddit were cited in the most problem answers.
Our read of this, which the data does not prove: the searches, the cited pages and the named agencies move together, and that fits a simple story. A best-of answer is mostly a summary of other people's lists, and the agencies on those lists get named. A problem answer reads pages that talk about the problem, spread across many sites, so no agency dominates it.
What the answers say when they name no tracked agency
Reading the answers shows two behaviours.
Some name smaller firms Peec does not track. A ChatGPT answer of 5 October to "We publish a lot but nothing gets cited by AI" shortlisted six: Searchtides, Passionfruit, Ai Media Group, Fifty One Degrees, Verticality and NYSEO/CO. A Perplexity answer to the same question on 20 October named one firm, Cited By AI, and stopped there.
Others decline to name anyone. A Perplexity answer of 5 October to the same question opened with "There isn't a single agency that fixes AI citation across all engines" and pointed the buyer to regulators, standards bodies, a content or SEO agency or consultant, and an internal audit, without naming a firm.
If you are choosing an agency
This is our read of the data, and the study did not test it.
- Ask both ways. A best-of question gets you the agencies with the most list placements. A problem question gets you the firms whose pages talk about your problem. Those are different shortlists, and you want both.
- Ask each engine. The same problem questions named a tracked agency in 75.8% of ChatGPT answers and 30.2% of Perplexity answers. One engine's silence is no verdict on the market.
- Describe the problem, then ask who is good at fixing it. This is our read, and it rests on one question: the only problem question that asked who is good, using the word "best", named a tracked agency in more than eight answers out of ten. We cannot tell whether "who is good" or "best" did the work.
- Check the page behind the name. On problem questions the most-cited pages were agencies' own service, use-case and audit pages. Open the page and see whether it answers your problem with specifics or just repeats it.
If you want to see how both kinds of question treat your own brand, start with what an AI visibility audit checks.
If you want the engines to name you
The two phrasings reward two different kinds of work. Best-of answers lean on ranked lists, so placement on the lists engines already cite covers that phrasing. We covered how to get onto them in getting into the best-of lists AI answers pull from.
Problem answers read pages that address the problem itself. In our data six of the ten most-cited were agencies' service, use-case and audit pages. The other four were a semrush.com blog post, an AI search audit tools page on growpredictably.com, a Reddit thread and a LinkedIn post. Each was cited far less often than the top best-of list. A page that answers "why does nothing we publish get cited?" in plain words, with a method, is the kind of page those answers read.
For other cuts of agency answers, see our studies on funding stage and on whether the three engines name the same agencies.
Limitations
- The tracked set. The main count only sees 51 agencies. The wider count catches more firms but is a name-matching rule, so treat its no-agency figures as upper bounds.
- The detector. Peec's brand detection misses in both directions. It flagged GEO Agency from the generic phrase, which is why we left it out. It also under-records two tracked agencies: Thrive Agency is detected in 5 best-of answers while Thrive is named in the text of 125, and The Digital Elevator is detected in 20 against 95 text mentions. Counting those text mentions moves the best-of share from 81.9% to 83.8% and the problem share from 49.6% to 49.8%, so the gap holds.
- Phrasing or problem. Our five problem questions each name a specific symptom (no citations, invisibility, conversion). The study cannot separate "describing a problem" from "this particular problem".
- The list rule. "Looks like a ranked list" is judged from the URL alone; we did not open each page.
- The window. 43 days, one answer per question per engine per day.
