The best GEO agencies in the US, ranked by how often AI actually recommends them

The overall ranking rewards agencies the engines name on the generic questions. Most buyers do not ask generic questions. If you already know your situation, start here: the top three agencies on each buyer question, with how many of that question's runs named them. Only iPullRank and Siege Media appear in both a segment top three and the overall top five. Nine of the eighteen agencies on this page are named almost exclusively on one segment question and nowhere else, which is the single most useful thing here if you are hiring.

Recommended in = the share of all 198 discovery runs in which an engine named the agency. Best engine = the one that named it more often; “Both” means the two were within one mention of each other. Seer Interactive and WebFX tied on 35 mentions each; Seer ranks higher because the engines place it earlier in the answer, average position 1.0 against 5.5. Digital Elevator was added to tracking on 4 September, so it has been measured over 144 runs rather than 198. Its 6.6% is calculated on the full 198 for comparability; on its own sample it is 9%.

We asked six questions, written the way a buyer types them, covering the three labels people use for the same discipline (AI SEO, GEO, AEO) and three buyer situations (B2B SaaS, a limited budget, Reddit and community): 1. What are the best AI SEO agencies in America? 2. What are the best GEO agencies in America? 3. Which US agencies specialise in generative engine optimization for B2B SaaS companies? 4. Best answer engine optimization (AEO) agencies for getting cited in ChatGPT and Perplexity? 5. Which GEO agencies are a good fit for a startup or small B2B company on a limited budget? 6. Which agencies can improve a B2B brand's visibility in AI search through Reddit and community marketing? We asked ChatGPT (gpt-4o-search-preview) and Claude (Sonnet 4.6, both with search), the two assistants that reliably return named shortlists with sources. Gemini rarely names agencies on these questions and Google AI Overviews trigger inconsistently, so we left both out rather than report thin numbers. Each prompt ran on each engine at least twice a day from 3 to 8 September, and four times on 7 September: 198 runs across the six questions. The prompts did not all go live at the same hour on the first day, so run counts vary from 32 to 34 by prompt. Every percentage on this page is an agency's mentions divided by all 198 runs; every “x of y” figure inside an entry is divided by that prompt's own run count, which is always stated. Answers vary run to run, which is exactly why a single query tells you nothing. What counts is a named recommendation in the answer body. A citation to an agency's website without naming it does not count, and being named as an example of what not to do does not count. Ties break on average position, then on sentiment. What this does not tell you is whether an agency is good. It tells you which agencies the engines have enough public evidence about to recommend, which is a proxy for public track record and nothing more. Small specialists with strong private client work will rank low here. Treat it as a shortlist, not a verdict. Agencies do not apply and cannot pay. Nineteen are tracked, and they are tracked because the engines named them. A few more were named or cited in only one or two runs and are not tracked yet: Searchbloom, Redd, Discovered Labs, RevenueZen, Karma Alien, Nerativ, Sculpt, 009 Agency, Mainsail, The CEO of GEO, Thrive and LoudFace. Searchbloom is the only one with enough mentions to place — four runs, which would put it seventeenth. We will fold them in at the next refresh. We have a link exchange with OutreachBloom: this page links to them, their AI-visibility roundup links to us. No money changed hands and no position was promised. They were tracked on the same prompts, over the same window, under the same rules as everyone else, and they finished thirteenth, behind twelve agencies we have no relationship with. If the exchange had bought anything, it would have bought a better number than that. All of it was measured with CrunchJunkie, the AI-visibility platform we build. That is a conflict of interest of a sort, so the prompts, the run counts and the window are all above, and the underlying runs are available on request.

Rank hides a split worth knowing about before you shortlist anyone. Seer Interactive is named in 32 ChatGPT runs and 3 Claude runs. Omniscient Digital is close to the reverse: Claude names it nearly twice as often as ChatGPT. NoGood is a ChatGPT pick almost four times over. Bay Leaf Digital is effectively a Claude-only recommendation. So “the AI recommends us” is not a claim any agency can make without naming the engine. If your buyers live in ChatGPT, Seer and NoGood are on your shortlist and Bay Leaf is not. If they live in Claude, it is the other way round. The agencies that show up in both — iPullRank, Siege Media, Directive, WebFX — are the ones with public evidence broad enough that two differently-trained systems reach for them independently. The movement inside six days is also worth reading. Directive climbed the most of anyone, and Omniscient, Obility, Siege and NoGood all moved several points, because the engines re-read the web between runs. That is the argument for measuring continuously rather than once.

Rank tells you how often an agency is named. This tells you what the engines say when they name it. The phrase column is the description they keep reaching for; the caveat is the reservation they attach. Sentiment is CrunchJunkie's 0–100 score pooled over the mentions in the window, and a dash means there were not enough scored mentions to compute one — not that sentiment was poor.

The same data seen a different way. Each figure is the share of an agency's own descriptions that reach for a given theme, so agencies with different mention counts stay comparable. iPullRank's shape is the technical one: relevance engineering in 42% of its descriptions, embeddings and technical SEO in a third each, information retrieval in a quarter. Go Fish Digital is the digital-PR shape, 57%. OutreachBloom's is narrow and sits on digital PR and AI visibility, which is what you would expect of an agency the engines only name on the Reddit question. Siege Media barely registers on these technical themes at all, because the engines describe it in terms of content and PR outcomes rather than techniques — which is not an absence of reputation, it is a different one. If you are an agency, this is the more actionable chart of the two. Rank tells you whether you are known. This tells you what you are known for, and whether that matches what you sell.

If you want to understand why the list looks like this, look at the sources. Across the 198 runs, nineteen of the twenty most-cited domains are agency websites. The single exception is Google's own developer documentation. That is the finding, and it cuts both ways. The engines are largely reading agencies writing about agencies, which is why a roundup its own author appears in can move a ranking. But on three of the six questions the most-cited page is an agency's own services page rather than anyone's list — Directive on B2B SaaS, iPullRank on AEO, Obility on Reddit — with a fourth, AI SEO, tied between iPullRank's own pages and a competitor's roundup. The engines also read the platforms' own documentation before they answer: Google's AI features pages picked up 14 citations on the AI SEO question, OpenAI's help centre 9 on AEO (where ChatGPT quotes OpenAI's own statement that there is no way to guarantee placement in ChatGPT search), and Reddit's business pages 11 on the Reddit question. The lesson for anyone trying to get recommended is in that middle paragraph: publish the thing the engines want to cite. A services page that answers the question beats a listicle that mentions you.

“Best AI SEO agency”, “best GEO agency” and “best AEO agency” mean the same thing to a buyer. They do not to the engines. Ask for AI SEO and you get iPullRank (30 of 32), Siege Media (24), Omniscient Digital (18), WebFX (14), then First Page Sage and Seer Interactive on 10 each. Ask for GEO and Siege Media appears in 31 of 32 answers, Go Fish Digital in 29, iPullRank in 26, Directive in 19 — and Seer drops to 9. Ask for AEO and iPullRank is named in 24 of 34 runs, NoGood second on 17, Seer third on 15, Siege fourth on 12, and nobody else above 9. Ask about your segment and the list changes again: Directive, Obility and Omniscient for SaaS; Obility, Foundation and OutreachBloom for Reddit; Digital Elevator, Yes Optimist and Citant.ai for a limited budget. If you are a buyer, ask the engines your own question rather than the generic one. If you are an agency, the space between those three lists is where the opening is.

The engines are consistent about what they value when they hedge on an agency, so these questions are theirs as much as ours. On the AEO question ChatGPT says outright that no agency can reliably guarantee a ChatGPT or Perplexity citation, cites OpenAI's own help page to prove it, and says a credible partner sells a measurable program instead. On the budget question it puts a credible engagement at around 2,000 to 3,000 dollars a month, and says that below that you should buy a focused audit rather than a vague retainer. 1. Which prompts will you measure, and how often? If the answer is “keywords”, they are selling SEO with a new label. 2. What does the share-of-voice report look like, and against whom? Ask to see a real one with the client name redacted. 3. Do you guarantee citations or placements? The right answer is no. 4. Can we start with a scoped pilot? ChatGPT recommends a 90-day pilot for any agency it cannot verify from public proof, which on this list is most of them. 5. Which of your own pages do the engines cite? An agency that cannot answer is not measuring itself. 6. What is the case study for a company like ours? Not a logo wall — a before-and-after on named prompts. The engines reward the agencies that publish one. If you want to run this measurement on your own brand rather than take our word for it, that is what AI visibility tracking does, and there is a free AI visibility check if you just want to see where you stand.