Cited Layer › Measurements › GEO agencies
Measurement
GEO agencies, measured instead of ranked
21–22 September 2026 · ChatGPT (two accounts in parallel) and Perplexity · 20 recorded runs
This is not a ranking of AI-visibility agencies. It is a record of which ones the answer engines actually name, how often across repeated runs, and which sources they cite when describing them.
Nobody is scored. Nobody paid to be here. The list is whatever the engines returned, and the runs that returned nothing are included.
Read this before the tables
I sell the service these agencies sell. I am an independent GEO consultant and a direct competitor of every company named on this page. Any comparison written by a participant is marketing, this one included.
What makes it worth reading anyway is the part most such pages omit: I ran the same measurement on myself and was named zero times out of twenty. My site does not appear in a single one of the runs below. I am not on the list because the engines did not put me there.
Five of the agencies that do appear in these answers got there by publishing their own "best agencies" ranking and including themselves at or near the top. That mechanism is documented further down, with the sources. This page does not use it: there is no ranking and no winner, only counts.
Method
What was run
- Mode, held constant: Brave, temporary chat, free-tier accounts. Two ChatGPT accounts in separate browser profiles, run in parallel; Perplexity alongside.
- A control question opened every ChatGPT session. A failed control discards the session rather than reporting it. None failed.
- Queries: AI search optimization agency · who does AI visibility audits · generative engine optimization consultant · GEO consultant for ecommerce · best generative engine optimization agencies 2026 · top AI search optimization agencies · and is [agency] a good choice for AI search optimization for four named agencies.
- Runs: three per question per engine on the list questions, two on the brand questions. Twenty recorded answers in total, each saved in full with its source list.
- Disclosure: runs were made from a Ukrainian IP address. Answer engines weight location; a set reproduced elsewhere will differ. This is stated here rather than in a footnote because it materially affects who appears.
- Sample size is small and the window is one week. Nothing here should be read as a stable property of any company.
Result 1
Who gets named, and how often
Counts are appearances across six runs of the same question — two ChatGPT accounts × three runs. A count of 6 means every run named them; 1 means five runs did not.
On who does AI visibility audits
| Named | Runs out of 6 |
|---|---|
| CyberLab | 6 |
| 009 Agency | 5 |
| The Enough Agency | 4 |
| Progression Agency | 4 |
| Kambaa | 3 |
| Terrier Agency | 3 |
| APG Agency | 3 |
| Rivulet IQ | 1 |
On AI search optimization agency
| Named | Runs out of 6 |
|---|---|
| Citevora | 3 |
| Algorithmix | 3 |
| The Project SEO | 2 |
| Growth Anchors | 2 |
Two observations that matter more than the order.
The question shape decides whether a list exists at all. "Who does X" returned seven to nine names per run and 22 to 36 sources. The noun "agency" returned nought to five names and 12 to 14 sources. A third phrasing — generative engine optimization consultant — returned no vendors at all in either engine, although the same question in Perplexity had named six providers eight days earlier.
One account returned no shortlist at all. On the "agency" question, one of the two parallel accounts went advisory on two of its three runs: it explained what the service is and named nobody. Same minute, same tier, same mode. Whether a buyer sees a shortlist is not fully determined by the question they ask.
Result 2
What the engines cite when describing an agency
Each of the four most consistently named agencies was then asked about directly. The interesting column is not the verdict — it is the sources.
CyberLab
Sources cited: its own domains only. Five of them, localised for different markets. Across both engines, not one directory, review platform, press mention or roundup. The most consistently named agency in this measurement has no third-party layer at all.
009 Agency
Business directories. GoodFirms appeared four times in a single answer, alongside Clutch and Crunchbase. Perplexity summarised a verified Clutch review inside the answer text. A trust-checker page and a single-review profile were also surfaced.
The Enough Agency
Press-release distribution and media. Three wire outlets plus a national title and a Medium post. In Perplexity the agency was described entirely from third-party sources — its own website was not cited once.
Citevora
Business directories, plus a naming problem. Crunchbase and TechBehemoths. Both engines also confused it with two similarly-named companies and a WordPress plugin, and said so in the answer.
Three things follow. A third-party layer is not required — the most-named agency has none. Where one exists, it is generic business directories and press wires, not the specialist GEO roundups one might expect. And a thin third-party profile appears to be worse than an empty one: a single review invites the engine to say the sample is too small, and an ambiguous company name invites it to describe somebody else.
Result 3
Five agencies were named because they published their own ranking
On best generative engine optimization agencies 2026 and top AI search optimization agencies, the sources behind several top names were the names' own comparison pages:
| Agency named | Source cited for it |
|---|---|
| GenOptima | gen-optima.com — its own 12-best list |
| Stratabeat | stratabeat.com/best-geo-agency/ |
| GreenBanana SEO | greenbananaseo.com — its own 2026 list |
| First Page Sage | firstpagesage.com — its own top-agencies post |
| 009 Agency | 009.agency/best-geo-agencies/ |
Perplexity noticed, said so, and cited them anyway:
"Several prominent comparison pages are published by agencies that also appear on their own lists. For example, Stratabeat explicitly discloses that it ranks itself first… Treat published rankings as lead-generation material, not neutral league tables."
In one case the full chain was visible end to end: an agency published a study, pushed it through a press-release wire, and the engine then cited the wire article as its leading source and placed that agency first.
This is reported because it is what the measurement found, and because it is the reason the disclosure sits at the top of this page. The mechanism works. It is also the mechanism that makes every ranking in this category, including the ones that rank me nowhere, worth very little.
Result 4
The same question, twice, gave a different first place
Two runs of best generative engine optimization agencies 2026, same engine, same day, same mode:
Run one
Ten names, led by GenOptima, then First Page Sage, Siege Media, Directive, iPullRank.
Run two
Ten names, led by Stratabeat, then Go Fish Digital, iPullRank, First Page Sage, Omniscient Digital. GenOptima does not appear at all.
Four names of ten overlapped. The agency ranked first in one run was absent from the other. Anyone quoting a single screenshot of an AI answer as evidence of their standing — in either direction — is quoting one draw from a distribution.
Limits
What this page does not say
It says nothing about quality
Being named often means the engines reach for a company, not that its work is good. Being named rarely, or never, means the opposite of neither. No client outcome, deliverable or reference was examined here.
It is one week, one country, small numbers
Twenty runs from a Ukrainian IP address over two days. Re-run from another country, or in a month, and the names will move — that is the finding, not a caveat to it.
It is not independent, and it does not pretend to be
Written by a competitor of everyone listed. The defence is not neutrality, which is unavailable to anyone in this category. It is that the method, the queries and the raw answers are published so you can run it yourself and get your own numbers.
Run it on your own category
The protocol is four questions, two engines, repeated runs, one fixed mode, a control question per session. It takes an evening and it will tell you whether your category returns a vendor list at all — roughly half do not.
If you would rather have it done properly on your own buyer questions, that is the visibility audit. The category check that precedes it is free, and it is where I tell people the honest answer about whether there is anything to win.
alex@citedlayer.com