I Asked AI for the Best GEO Agency

AI cites agency lists where the publisher ranked itself No. 1, then recommends someone else. What 30 AI answers and 156 citations show.

Part of the GEO Roadmap › Stage 5: Believe

When you ask AI for the best GEO agency, about a third of its citations go to “best agencies” lists written by agencies that ranked themselves No. 1. But AI doesn’t recommend the agencies that wrote them. It recommends the brand that keeps showing up in everyone else’s lists.

Below: what 30 AI answers and 156 citations show, the “I ranked myself No. 1” club, how consensus gets bought in Turkey, and the one thing to do this week.

On Oct. 12, 2026, a press release of mine went out on the wires in the U.S. It carries the findings of a study in which I put more than 100 websites through our GEO Score Card. Three numbers stand out: about three in four sites give AI crawlers no explicit access in robots.txt, roughly 87% have no Wikidata entry, and about half have no independent reputation signal. By that I mean a third-party review, award or industry mention the company didn’t publish itself.

While preparing the release, that last number kept bugging me. If a brand has no independent signal, what does AI believe instead? Then I turned the question on myself: in my own industry, GEO agencies, what does AI read as an “independent source”? The answer made me laugh. Then it made me think.

Grab a coffee.

The thesis: Ranking yourself No. 1 earns a citation, not a recommendation

When AI answers “best GEO agency” questions in the U.S., it gives about a third of its citations to “best X agencies” lists. I opened the lists published by agencies: 12 of 12 included their publisher, and nine put the publisher straight at No. 1.

Here’s the good part: AI uses these lists as sources, but it doesn’t recommend the agencies that wrote them. It recommends the brand that keeps showing up in other people’s lists. The Believe door doesn’t open with your own voice. It opens when other people talk about you, consistently.

1. What I measured

The method in one paragraph: I set up a GeoGenie workspace to track Stradiji’s U.S. visibility: six topics, 18 prompts. No brand names appear in the prompts. They’re written the way a neutral buyer would ask. The prompts ran 90 times between Oct. 5 and Oct. 11, 2026, across ChatGPT, Perplexity, Gemini, Google AI Overviews and AI Mode. The output: 30 answers, 110 domains, 137 pages and 156 citations. I classified every page by its title, by hand, and opened the 12 agency-published lists one by one. I also pulled Authority Score for 20 cited domains from Semrush on Oct. 11, 2026.

2. The “I ranked myself No. 1” club

I took notes while reading these lists, and I smiled a fair bit. A few examples, all straight from the pages:

• Thrive put itself at No. 1 in its GEO agencies list and gave itself 98/100. Like a student writing their own report card. Whether the missing two points are modesty or a credibility buffer, we’ll never know.
• Minuttia’s list says “October 2026” in the heading. The page was last updated on April 3. The calendar flips every month; the content doesn’t. AI likes adding the current year to its fan-out queries, and agencies know it.
• Grizzle says its list is “unranked.” It still put itself first. No ranking, but there’s an order.
• Omniscient Digital is at least honest: it says it included itself because it offers GEO services. The most likable member of the club.
• The tool side isn’t exempt. In Profound’s list of top GEO experts, someone from Profound’s founding team sits at No. 3. On the same person’s own site, he’s No. 2. Credit where it’s due: he discloses it.
• NoGood’s “top GEO and AEO experts to follow” is the masterpiece of the genre. At No. 1: John Mueller, Google’s search advocate. The spokesperson for a company that says you don’t need any special optimization for its AI features sits at the top of a GEO experts list. At No. 2: NoGood’s own founder. At No. 3: Rand Fishkin. I have endless respect for Rand; he’s one of the people who built this industry. But he left Moz in 2018 and today runs SparkToro, an audience research tool. His spot on a 2026 “GEO experts” list says less about him and more about the list maker: they picked fame, not expertise.

Every one of these lists follows the same recipe. I call it the credibility sandwich: the list maker on top, two or three famous names in the middle (borrowed for credibility), unchecked filler underneath. The famous names are the bread. What’s actually being sold is the top slice.

And the example that matters most to me: GenOptima, a Shanghai-based agency, distributed a list titled “The 10 Best Generative Engine Optimization Agencies in 2026” as a press release via Medianet, Australia’s press release distribution network, on Aug. 18, 2026. GenOptima is No. 1. AI cited the page.

So press release networks are among the sources AI reads. And that channel can be used to distribute a self-ranked list dressed up as news. I’ll come back to this.

3. A citation is not a recommendation

The fun part ends here. The real finding starts.

GeoGenie tracks not only citations but also which brands get recommended. I put the two side by side:

• Cited but not recommended: The lists of small agencies that ranked themselves No. 1 get citations. None of those agencies appear among the top 16 recommended brands.
• Recommended: The most recommended agency is iPullRank (visibility score 28.9). There’s no self-written “best agencies” list from iPullRank in the sample, but it appears in seven of the 12 competitors’ lists. Siege Media (26.7) appears in eight of 12, and Omniscient Digital (27.8) in seven.

AI reads the list as a source. When it decides whom to recommend, it looks at the common ground across those sources. Being No. 1 on your own list is one vote. Showing up again and again on other people’s lists is consensus. It fits what we saw when ChatGPT decides before it searches: a recommendation comes from what has piled up about you, not from a single page.

Authority or format?

I pulled Authority Score from Semrush for 20 cited domains:

Two sites with an Authority Score of 4 and 6 got cited. Stradiji’s is 23. Six of the 20 cited domains score lower than ours. So the reason AI doesn’t cite us isn’t a backlink gap. It’s page type: for “best agency” questions, AI looks for a list format, and our site has no such page.

And it won’t. The data is clear: that format earns citations, not recommendations.

Limitations. Eighteen prompts, one workspace, U.S. only. Citations come mostly from two days (Oct. 7-8). Lists were classified by title, by hand. I didn’t run a correlation between citations and authority: most pages were cited only once, and any correlation from this data would be noise. I’m not saying “the whole GEO industry.” I’m saying “in these 30 answers.”

I turned these findings into a step-by-step deck in NotebookLM; you can open it here.

4. What do the big studies say? Does my data refute them?

I’m not the first to measure this. I compared my numbers with three studies:

• Lily Ray, in a study published on Search Engine Land on June 18, 2026, looked at 100 B2B “best software” queries in Google AI Overviews on three dates. When Google cited a self-promotional list, it left the list’s publisher out of its recommendations 69% of the time (224 of 323 citations). My data confirms this and extends it from software to GEO agencies.
• Peec AI data (The AI Marketing Newsletter, Aug. 31, 2026) shows listicles fell to 7.8% of ChatGPT citations after GPT-5.6, roughly half their earlier share. In my data, the share is 30.8%. That’s not a refutation. My sample comes from several engines and from one of the most list-heavy niches there is: agency recommendations. We’re likely seeing a category effect. And I’m looking at 30 answers; they’re looking at far more.
• Promptwatch’s July 2026 data puts listicles at 18% of AI Overviews citations (down from 26% in the first quarter) and 9.7% in ChatGPT. My category sits above both.

So do I refute anything? Yes, but not a study. A sales pitch. In recent months, dozens of guides have been published on how to write a listicle AI will cite. Their message: write a list, put yourself at No. 1, and AI will recommend you. My data shows half of that works: the citation comes. The other half doesn’t: the recommendation doesn’t. And your buyer sees the recommendation, not the citation.

From the field: In Turkey, consensus is for sale

I work from Istanbul, so I ran the same question in Turkish. On the evening of Oct. 11, in an incognito window, I searched Google for “en iyi geo ajansı türkiye” (best GEO agency Turkey). The AI Overview listed agencies and cited three pages. The first was a “Best GEO Agencies: Updated List (2026)” article on Webtekno, one of Turkey’s biggest tech news sites. Right next to the date, in small print: “Sponsored Content.”

I searched for the same headline on other Turkish news sites:

Four major news sites, four paid placements, the same agency at No. 1 in every one. Underneath, the credibility sandwich again: well-known U.S. agencies like Siege Media, Directive, First Page Sage and Go Fish Digital sprinkled in. The DonanımHaber list is titled “Turkey’s 10 Best GEO Agencies.” At least nine of the 10 aren’t based in Turkey, and one isn’t even an agency; it’s a link outreach tool. And at No. 8 on the Medyascope list: GenOptima. Yes, the same Shanghai agency that ranked itself No. 1 in the Australian press release. The template travels across continents.

The AI Overview drew on this network. It cited the “Sponsored Content” page and summarized it as agencies frequently mentioned in industry lists.

Now go back to Section 3. In the U.S. data, consensus decided who got recommended. In Turkey, that consensus has a price. Run the same list on four news sites, and AI sees you as the agency that’s “frequently mentioned.”

A question for Google: How is this different from SEO spam?

To be fair, Webtekno played by the rules. The content is labeled “Sponsored,” and the agency link I checked is rel="nofollow". Google’s spam policy doesn’t treat labeled sponsored content written for readers as site reputation abuse. On paper, everything is in order.

But Google’s link rules were written for the PageRank era. nofollow stops a link from passing value. It doesn’t stop the text from entering the answer. AI Overviews read the text, not the rel attribute. And the “Sponsored” label doesn’t travel into the answer. The reader doesn’t see an ad. They see an answer in Google’s own voice.

This game isn’t new. In 2013, Google made Interflora, the U.K. florist that ran advertorials in British newspapers, disappear even for searches on its own brand name. In 2026, the same kind of advertorial is safe because its link is nofollow, and it sits as a source in Google’s own AI answer.

My question: Why does Google reward in its AI answers a tactic it has penalized in the blue links for more than a decade?

Note. I’m saying nothing about the quality of this agency’s work. It may be a good agency. What I’m saying is that AI’s “best” answer can be built from labeled ads. The problem isn’t the agency. It’s the answer engine.

One more confession, because it fits. I’m on a Turkish “Best SEO Experts (2026 Updated List)” myself: No. 11, with the list’s author at No. 1. My education field reads “–,” and so does my profile link. Five of the 12 people on that list have the same blank. My first reaction was a laugh. My second was GEO: whoever compiled it, a person or an AI assistant, checked machine-readable sources, and my education isn’t in any of them. I have no Wikidata entry, and our schema has no alumniOf field. My press release says roughly 87% of sites have no Wikidata entry. I’m in that 87%. So I’m fixing my own Recognize door first.

5. And my own press release?

Back to my own desk. My release went out through the same channel GenOptima used: press release distribution. Both land in the same pool. The difference is the content:

Let me be honest: I don’t know whether that difference works in AI. I can’t measure the effect of a release that went out a few hours ago. And over the same weeks, our homepage, schema and Wikidata setup are also changing, so even if I see movement later, I can’t credit all of it to the release.

So I’m doing the most honest thing I can: I’m calling my shot in advance.

• I added a new topic to GeoGenie, “Press Releases & Earned Media for AI Visibility,” with seven prompts. It’s measured every Monday. The first measurement ran before the release went out, so I have a clean baseline.
• My hypothesis: The release may earn citations in AI, but it won’t bring recommendations on its own. Recommendations come when the data in the release gets repeated in other publications and lists. In other words, I’m testing this finding on our own brand.
• I’ll share the interim result in week six and the full result in week 12, in my newsletter and on this blog. If my hypothesis turns out wrong, I’ll write that too.

I’m open to being proven wrong on our own brand. I don’t know a better way to be more credible than a list that ranks itself No. 1.

The one thing to do this week

If you’ve been pitched a “press release for AI visibility” or “get placed on best-of lists” package, ask three questions before you sign:

1. Does this content carry data, or a ranking? Getting yourself placed at No. 1 on a list may earn a citation; it won’t earn a recommendation. Content that carries data, method and limitations becomes a source others can cite.
2. Why would anyone else cite this? If the answer is “because we say so,” AI will reach the same conclusion.
3. Will you measure before and after? Write 10 buyer prompts for your category with no brand name. Ask them in ChatGPT, Perplexity and Google AI Mode before publication and again four weeks later. Look at who gets recommended, not who gets cited.

Want to see where your brand stands at the Believe door? The GEO Score Card is free. If you’d like help setting up the measurement, see our GEO consulting page.

Closing

In the AI era, “Who tells your story?” has become the question that matters most. Some agencies have answered it in the shortest way possible: “We do. At No. 1.” AI took note, used it as a source and gave its recommendation to someone else.

My view: The Believe door doesn’t open with your own voice. It opens when others talk about you, consistently. Being No. 1 on your own list is easy. Being No. 7 on someone else’s is worth more.

So, a question: In your industry, how many of the “best of” lists AI cites were written by the company sitting at No. 1? Start counting. You’ll be surprised.