--- title: "Do Not Hire an AI Visibility Vendor on Rank Alone" description: "I would not hire an AI visibility vendor on a rank screenshot. In the buyer-choice segment, 399 of 699 cited domains show up in a single run." canonical: "https://jaxonparrott.com/blog/do-not-hire-ai-visibility-vendor-rank-alone" date: "2026-09-28" last-updated: "2026-11-04" --- # Do Not Hire an AI Visibility Vendor on Rank Alone I would not hire an AI visibility vendor because an engine ranked them first. I check the grade behind the rank. On the question a buyer actually asks when choosing, the 2026-11-04 release of the Machine Relations Index shows a long thin tail: 399 of the 699 domains cited in that segment appear in exactly one run, and the most-cited source appears in 12.06% of runs. The live leaderboard for that question is [AI Visibility & GEO: How buyers choose MRI leaderboard](https://machinerelations.ai/index/categories/ai-visibility-geo/how_choose). It moves with each release; the figures below are from release `mri_score_v2.1+2026-11-04+94f24cb38da0`. That page is the deeper dataset. This essay is how I would use it as a founder: what I ask in the room, the scope of the refusal, and the release that would let a rank carry the hire. ## What the hiring question is made of The Machine Relations Index publishes a segment for this decision: the category "AI Visibility & GEO" crossed with the question shape "how buyers choose." That is the segment an engine draws on when someone asks which AI visibility vendor or tool to pick. In the public view released 2026-11-04, release `mri_score_v2.1+2026-11-04+94f24cb38da0`, window 2026-05-10 to 2026-11-04, the segment cleared the evidence floor at 282 observed runs across 13 distinct dates. Across the segment, 699 distinct source domains were cited at least once, and 1,856 is the sum of every domain's cited runs. Of those 699 domains, 399 were cited in exactly one run. Only 33 were cited in 10 or more of the 282. | Source | Source role | Rank | Citation rate | Cited runs of 282 | | --- | --- | --- | --- | --- | | YouTube | Search or media platform | #1 | 12.06% | 34 | | Stackmatix.com | Other observed source | #2 | 10.99% | 31 | | Llmpulse.ai | Other observed source | #3 | 8.87% | 25 | | Hubspot.com | Vendor-owned source | #4 | 8.51% | 24 | | Tryprofound.com | Vendor-owned source | #5 | 8.16% | 23 | | Semrush.com | Vendor-owned source | #11 | 5.67% | 16 | The table is rows from the public leaderboard; I use the rate with its run counts, not the rank alone. By source role, summing each domain's cited runs, 75.1% of the 1,856 sit on what the Index labels "other observed source." Vendor-owned pages hold 12.6%. Editorial publications hold 3.1%. Analyst and consulting research holds three of the 1,856, about 0.2%. The two sources an engine reaches for most often on this question are YouTube, at a 12.06% citation rate (34 of 282 runs), and Stackmatix.com, at 10.99% (31 of 282 runs). HubSpot sits at #4 with 8.51% and Semrush at #11 with 5.67%. Tryprofound.com, labelled vendor-owned, is #5 at 8.16%. No single source is cited in more than about one run in eight, and the leaderboard lists 699 of them. That spread is consistent with [Google's own documentation](https://developers.google.com/search/docs/appearance/ai-features), which says AI Overviews and AI Mode may use "query fan-out" to show "a wider and more diverse set of helpful links" than classic search. A young question shape stays volatile until enough runs accumulate. Volatility is a property of the evidence, not a reason to pretend the top row is durable. ## What I do with a rank in a pitch If a vendor says they are the top-cited AI visibility platform, I ask which segment, which release, and which confidence grade. For the question I am trying to answer when I hire, the leader is cited in 12.06% of runs and 399 of 699 cited domains appear in a single run. I would say it that way in the room, and I would show the table. This is not a reason to wait to hire. It is a reason to hire on something other than a ranking screenshot: a documented methodology, a named release I can re-pull, and a willingness to show the collecting-grade rows next to the graded ones. [The public Index](https://machinerelations.ai/index) publishes the release id, observed runs and run dates for each segment. A number without its run count is a number I would not build a company decision on. The SEO-era mistake was buying a rank instead of a capability. The AI-visibility hire repeats it faster because the category is younger and the instrument measuring it is months old, not a decades-old discipline. [NIST's AI Risk Management Framework](https://airc.nist.gov/airmf-resources/airmf/5-sec-core/) asks that the test sets, metrics and tool details used in evaluation be documented (Measure 2.1) and that performance be shown for conditions similar to deployment (Measure 2.3). A grade of "collecting" is the dataset telling the truth before the market's story catches up. ## The decision I would not award the work, the renewal, or the case study on a rank that cannot name its release, its segment, and its grade. I would hire on a method I can re-run: the prompt set, the engines, the date window, the evidence floor, and the rows that have not cleared it. The scope is the how-buyers-choose segment of AI Visibility and GEO on the 2026-11-04 release, and any pitch that cites a rank inside that question. It does not say vendor rankings are fake. A domain can be genuinely cited and still be graded collecting because the window has not run long enough. The grade is a statement about evidence volume. It is also not a judgment of the other five question shapes the Index tracks beside this one: best tools, is it worth it, problem-first research, top lists, and comparisons. Engines cite different domains depending on the question. A rank on a different shape is a different conversation, and it still has to survive the same grade check. I would change my mind when a later release shows an AI-visibility-specific vendor with a well-evidenced row on this segment, the long tail of single-run domains thinner than it is today, and the vendor willing to show the thin rows from the same cut. One strong row in a segment where 399 of 699 domains appear in a single run would not do it. The tail would have to thin. Until that release, the screenshot is a lead, and the grade is the decision. The order I want the work done in, once someone is hired, is the order in [Machine Relations](/blog/machine-relations). The names for confidence, citation rate, and source class are in the [glossary](https://machinerelations.ai/glossary). I coined Machine Relations in 2024 because engines were already deciding who a market trusts, and a rank without a grade was already a bad way to watch them do it. ## FAQ ### What does "how buyers choose" mean in this dataset? It is one of six question shapes the Index tracks per category, alongside best tools, is it worth it, problem-first research, top lists, and comparisons. Each shape is measured separately because engines cite different domains depending on the question the buyer asked. The hire I am describing is the how-buyers-choose shape inside AI Visibility and GEO. ### How thin is the evidence under a rank? A domain needs at least 10 observed runs across at least 7 distinct dates before the Index grades it A, B, or C. The segment itself cleared the floor at the category level in the 2026-11-04 release, at 282 runs across 13 dates. Under it, 399 of the 699 cited domains appear in exactly one run, and only 33 appear in 10 or more. ### Does that mean the rankings vendors quote are fake? No. It means a rank is early evidence. A citation can be real and still be graded collecting because the observation window has not run long enough. I refuse the rank as a hiring reason. I do not refuse the underlying observation. The vendor who will show me the collecting rows is the vendor I can keep talking to. ### What do I ask before the second meeting? The release id, the segment, the confidence grade behind the specific number, and whether I can open the same cut on the public leaderboard and re-check it. On this release the check is `mri_score_v2.1+2026-11-04+94f24cb38da0`. If the answer is a screenshot with none of those, I already have the decision. --- *Published September 28, 2026* By [Jaxon Parrott](https://jaxonparrott.com) — Founder & CEO, [AuthorityTech](https://authoritytech.io) Updated 2026-11-04 Part of [Machine Relations](https://jaxonparrott.com/machine-relations) · [Glossary](https://jaxonparrott.com/glossary) · [Founder judgment](https://jaxonparrott.com/topics/founder-judgment) Tags: AI Visibility, Founder Judgment, Measurement ## Continue reading - [The Category Ownership Trap](https://jaxonparrott.com/blog/category-ownership-trap): Owning a category definition does not mean owning the buyer journey. How I connect Machine Relations to the questions people ask before they know its name. - [I Read the Domains Before I Size the Budget](https://jaxonparrott.com/blog/founder-decision-ledger-unclassified-top-ten-slots-2026): I size an AI visibility budget from the domains in the top ten, because in every measured category the largest block of those slots has no source class. - [How I Check the Sentence Behind a Visibility Score](https://jaxonparrott.com/blog/how-i-check-what-an-engine-says-about-a-company): I would read the sentence behind a visibility score, separate a cited answer from a co-occurrence hit, and refuse a number that blends the two. --- *Machine-readable version of [https://jaxonparrott.com/blog/do-not-hire-ai-visibility-vendor-rank-alone](https://jaxonparrott.com/blog/do-not-hire-ai-visibility-vendor-rank-alone)*