Rankscale recently published an evaluation of 33 agencies offering GEO and AI SEO services. The study tried to look past agency claims and measure observable AI visibility, third-party reputation, documented experience, and the risks behind that visibility. Coalition Technologies was pleased to be named the top digital marketing agency for GEO and AI SEO. The study also raised a question that deserves more attention than another agency ranking:
Should a company receive full credit when part of its AI visibility comes from pages that rank the company itself?
Rankscale answered by applying a graduated penalty for self-serving listicles. I agree with the decision. A well-researched comparison can help readers, even when one of the companies being compared publishes it, but most agency-produced top-ten pages don’t meet that standard. They exist because recommendation queries attract valuable prospects and AI systems have been willing to cite them.
Why Count a Tactic That Works Against an Agency
The strongest defense of these promotional rankings is simple: They work.
If an agency publishes a list and begins appearing more often in ChatGPT, Google AI Overviews, Gemini, or Perplexity, the page produces a marketing result.
But, it’s unlikely that result will last. Something that is only a measurement of current visibility tells us what an AI system retrieves today. That doesn’t tell us anything about whether the visibility comes from evidence the system will continue to trust once its developers recognize the pattern.
Buyers hiring an agency need both answers. They need to know that their agency not only has a good grasp on the underlying technology and a current view of what it favors, but also on how these systems are likely to respond to spam and underhanded tactics.
Rankscale handled this reasonably. It counted current AI visibility as a positive signal and treated extensive self-ranking content as a separate risk. Removing the penalty produced a 0.97 rank correlation with the original results and left the leaders unchanged, according to the full study. Ultimately, the penalty wasn’t responsible for a particular outcome; it just helped establish which agencies were independently capable of demonstrating their authority.
AI Can Mistake Publication for Corroboration
AI search is still in a toddler phase when it comes to judging motives. It can find and repeat a claim long before it can reliably assess why the claim was published. When an agency publishes a page naming itself the best agency, the system may extract the ranking without seriously testing the ranking method.
A human reader usually recognizes the conflict immediately. An AI system has to infer whether the criteria were chosen fairly, whether competitors received equal scrutiny, whether the publisher verified its own claims, and whether the article would have reached the same conclusion if another agency had performed better. Retrieval and summarization make it easy to take a shortcut: the brand recommended the brand, so the answer recommends the brand too.
This can produce a circular evidence system. An agency calls itself a leader, an AI repeats the claim, and the agency then cites its AI visibility as evidence that the market recognizes its leadership. Repetition starts to resemble consensus even when the underlying recommendation came from the interested party.
Some AI systems, including ChatGPT, already question self-serving sources in at least some responses. The behavior is inconsistent, which is unsurprising in a channel where the same prompt can produce different sources across models and dates. OpenAI says ChatGPT Search uses multiple factors intended to find relevant and reliable information, but no public standard explains exactly how it discounts a publisher’s financial interest in a recommendation.
A Useful Comparison Has to Earn Its Conclusion
Not every top-ten article should be thrown out because it could have a promotional impact in AI answers. A useful comparison should explain how companies were selected, apply the same criteria to each one, link claims to evidence, disclose conflicts, and allow for the possibility that the publisher does not finish first. Original testing or proprietary data can give the page value beyond its conclusion.
Most self-ranking agency listicles are much thinner. They summarize public service pages, offer generous descriptions of competitors, and reserve the strongest language or the first position for the publisher. Even for more rigorous evaluations, a company may have the ability to stack the deck in their favor. For instance, smaller AI SEO companies often highlight their “smallness” as an advantage.
Volume also becomes a thing worth examining. A single, well researched article may have merit, but when 50 different findings are produced off of a very thin bit of examination, toss it out. Dozens of closely related pages targeting ‘best agency’ queries show that influencing retrieval is the objective rather than being the best. At that scale, the content should be judged by how much new information it contributes, not by how many recommendation prompts it reaches.
With all of those caveats out there, it’s worth flagging that most AIs aren’t going to spend a ton of time doing that kind of analysis. It’s expensive and slows down the instant response most users want. AIs will move to take shortcuts in these evaluations, and their shortcut is likely to be carving out those entities attempting to manipulate the rankings.
Search Platforms Have Seen This Pattern Before
Traditional search has repeatedly rewarded tactics before learning to discount them. Keyword stuffing, link schemes, doorway pages, and scaled low-value content all created periods when manipulation was profitable.
Google’s current spam policies expressly cover attempts to manipulate both conventional rankings and generative AI responses in Google Search. Google also defines scaled content abuse as producing many pages primarily to manipulate rankings rather than help users. A self-ranking listicle is not automatically a violation, but a large collection of unoriginal recommendation pages created mainly to influence visibility fits the definition of the targeted manipulation.
AI retrieval systems do not begin with no defenses. A 2026 study of nine LLM-enhanced search systems tested 27,000 requests and reported a cumulative blocking rate of 99.78% against the traditional black-hat SEO attacks it examined. Retrieval performed most of the filtering. The same researchers found that newer techniques designed for LLM search could double manipulation rates compared with their baseline.
This study shows how quickly AI search can inherit established spam defenses while exposing different weaknesses of its own. A tactic built around one of those weaknesses may have a much shorter useful life than an older SEO tactic did.
Clients Carry Most of the Downside
An agency experimenting on its own website can accept the risk. The problem grows when the same approach becomes their client strategy. The agency collects fees and can point to near-term visibility, while the client pays for content whose value may disappear after a retrieval or ranking change.
The loss may also extend beyond rankings. A company can end up with dozens of pages that make poorly supported superiority claims, compare it awkwardly with competitors, or read like advertisements disguised as independent analysis. Those pages may remain associated with the brand after AI systems stop citing them. And if AI companies grow frustrated with manipulation, they may also penalize them.
A safer AI-search strategy produces material that remains useful after a ranking change. Original research, clear product information, firsthand expertise, credible case studies, and independent coverage give retrieval systems something worth citing. They also help a prospective customer who visits the source and reads it without an AI intermediary.
AI Visibility Needs a Durability Test
Marketers should measure mentions, citations, and referral traffic, but those numbers need context. Teams should ask how much visibility depends on sources the brand controls, whether those sources add original information, and whether an independent evaluator would reach the same conclusion. A citation earned through useful evidence is a different asset from one earned because a system failed to notice the publisher’s conflict.
Rankscale was right to count the risk. The study did not pretend that self-serving listicles have no effect, and it did not disqualify every agency that published one. It recognized that current performance deserves less confidence when it depends heavily on a shortcut AI systems are already learning to question.
AI search will keep changing how it selects sources and weighs brand claims. Agencies can spend client budgets chasing each temporary opening, or they can build the evidence AI systems are trying to find in the first place. A model update is less likely to erase visibility supported by original evidence and independent recognition
