When I Would Call a Vendor an AI Shortlist Leader

I would not call a vendor an AI-shortlist leader because it holds rank one on one newly published leaderboard. The first HR software buyer-choice segment in the Machine Relations Index has a real leader and a thin margin. Pin is cited in 14 of 101 observed runs. The honest use of that fact is an experiment with a written stop rule, not a category claim.

The record of that list I would keep attached to any sentence about it is One HR AI-Citation List Is Live, and It Isn't Finished. That page is the deeper dataset. This essay is how I would use it as a founder: which tests a rank has to pass before I change budget or positioning, and which failure would make me withdraw the sentence.

What the first HR list actually ranks

The Index separates question shapes because a buyer asking how to choose software is not asking for industry news, a head-to-head, or a verdict on price. This leaderboard is one shape: how buyers choose HR and talent software. It cleared the floor at 101 observed answer runs across seven dates. The release is mri_score_v2.0+2026-09-21+d73e239e7c5b. The segment covers 235 cited domains. Five other buyer shapes in the same category — best tools, is it worth it, problem-first research, top lists, and comparisons — were still collecting.

Rank	Source	Cited runs	Role	Whole-window engine reach
1	Pin	14 of 101	Vendor-owned	5 of 6
2	LinkedIn	13 of 101	Community and social	6 of 6
3	Reddit	9 of 101	Community and social	4 of 6
4	Selenios	9 of 101	Other observed source	2 of 6
5	Umantis	9 of 101	Other observed source	2 of 6
6	US Tech Automations	9 of 101	Other observed source	6 of 6
7	ClearCompany	8 of 101	Other observed source	6 of 6
8	CVViZ	8 of 101	Other observed source	2 of 6
9	Gem	8 of 101	Vendor-owned	5 of 6
10	Lever	8 of 101	Other observed source	6 of 6

I would not read this as the ten best HR software companies. LinkedIn and Reddit are not HR vendors. Several other rows are operator or company sites whose source class is still broad. The list measures which domains answer engines cited inside one monitored question, in one window. It does not measure product quality, satisfaction, revenue, market share, or conversion.

I would also refuse to treat rank as distance. Pin leads Lever by six cited runs, and leads ranks three through six by five. Ranks seven through thirteen are tied at eight cited runs. The line between "top ten" and "outside the top ten" is a sort key inside a seven-way tie. A founder who turns that cutoff into a market claim is promoting the display order.

The words for a question shape and a cited run are in the glossary. The habit of separating one shape from a category claim is part of the work on Machine Relations.

Four tests before the sentence

Is the shape published? This one is, at 101 runs across seven dates. The rate is citable under the instrument's rules. The HR category is not thereby mature. One published shape can support a decision about that shape. It cannot support "engines recommend us" as a category-wide line.

How wide is the margin? Rank one is 14 cited runs. Rank ten is eight. The absolute gap is six runs on a denominator of 101. That is enough to call Pin the measured leader on this release. It is not enough to assume the order survives a handful of new observations. One more citation moves positions inside the tied block. The smaller the margin, the more the next run matters, and six runs is a small margin.

Does the source travel across engines? The public release exposes whole-window engine presence by domain, not the per-engine rate inside this HR segment. I would keep that boundary in the footnote every time.

The collection itself had gaps, and an earlier reading of this list got the gaps backward. Across the three engines collected every day of the window — ChatGPT, Claude, and Perplexity — seven of the ten are cited by all three, and six are cited by those three plus Gemini. That six is a floor: a recorded citation happened, so those six were cited by all four, and Gemini's 22 blind days could be hiding a seventh. Selenios, Umantis, and CVViZ were observed cited only by Claude and Perplexity. ChatGPT is absent from four of the ten domains, Claude from one, and Perplexity from none. Those three absence counts are exact, because those three engines were collected every day. Gemini was not observed citing three of the ten, which is an upper bound on a real absence, not a finished count.

Google AI Overviews recorded zero citations from August 15 to September 24, 2026, which is 41 of the window's 132 days. Google AI Mode recorded zero from September 14 to September 24, 11 days. Gemini recorded zero on 22 intermittent days: May 11 to 15, then scattered single days from August 11 to September 16. Whole-window presence only accumulates. A missing collection day can knock a domain out of an engine's set and can never add one. Presence on those lanes is real. Absence from them is not measurable in this release. On September 26, 2026, Gemini joined the set of engines whose absences I would not treat as exact.

An earlier version of this count said only four of the ten appeared in all six engines, that Google AI Overviews was absent from five, and that the narrowest engine was a market fact. That reading is withdrawn. The engine that looked narrowest was the one the instrument had measured least. A September 25 recomputation that treated Gemini's absences as exact was wrong on the same point and is withdrawn too. I would quote the three-engine floor and the upper bounds. I would not quote the withdrawn six-engine absence ranking.

A rank built on sources with uneven reach is a narrower asset than the same order built on sources every measured engine uses. This list is the narrower kind, and the narrowness is only partly knowable until the gaps close.

Does it repeat in another buyer shape? Not yet. The other HR shapes are still collecting. If the next published shape has a different top, this list is shape-specific. If the same sources repeat, the category has started to show a durable layer. Until then a category sentence is ahead of the measurement. That unanswered test is the most important line in the file.

The stop rule I would use

I would treat the first published rank as an experiment when any two of these are true:

1. Only one commercially relevant question shape is published.
2. Rank one leads rank ten by fewer than ten cited runs.
3. The cutoff contains a tie that extends past the displayed top ten.
4. Fewer than half of the top ten have a clean six-engine reach, on a release where every engine was actually collected.
5. No source in the top three repeats in another published buyer shape.

On this leaderboard, the first three fire: one shape, a lead of six cited runs, and a tie from rank seven through rank thirteen. The fifth cannot be tested. The fourth I would mark untested for a different reason. The six-engine reach clause was doing work in a reading the same measurement later withdrew, because three engines had collection gaps. I would not let a withdrawn absence count carry a leadership claim, and I would not let it carry a stop all by itself either. Three fired tests are already more than two. The rule says experiment.

The decision is to act narrowly, keep the evidence, and refuse the category claim. For the rank-one vendor, I would document the pages and independent sources around the result, preserve the release id, and watch the next two releases. For a vendor outside the top ten, I would compare the source roles already present before I funded more generic pages. For everyone, I would separate buyer-choice work from news visibility.

The HR news leaderboard is the contrast that keeps the point honest. It has 608 runs across 53 dates, an older and thicker segment. News and buyer-choice are different source markets. I would not copy a tactic from one into the other because both say "HR."

What I would do with each kind of row

The top ten is a source list, and I would brief it as roles before I briefed it as a horse race.

Pin, 14 of 101, vendor-owned, whole-window reach 5 of 6, is the measured leader on this shape and this release. The lead over Lever is six cited runs. The lead over the block at nine citations is five. I would preserve Pin's pages and the independent sources around them, and I would schedule the next two releases before anyone used the word leader in a market sentence. Six runs on a denominator of 101 is a lead I can say out loud and a lead a few new observations can narrow.

LinkedIn at 13, community and social, 6 of 6 engines across the whole window, is the awkward second place for anyone who wanted this to be a software ranking. Reddit at 9, reach 4 of 6, is the same lesson one row down. A buyer-choice answer that cites a network and a forum is telling me where the engines went for this question. It is not telling me those companies sell HR software. I would study what on those domains is being cited, the thread or the company page, before I funded a page of our own that imitates a product sheet.

Selenios, Umantis, and CVViZ are the rows that keep the engine story honest. Nine, nine, and eight cited runs, and observed only on Claude and Perplexity. Their whole-window reach is 2 of 6. A rank that depends on two engines in the panel is a narrower asset than US Tech Automations or ClearCompany or Lever, each of which shows 6 of 6 on the whole-window field. I would not copy a tactic from a two-engine row and expect it to show up in ChatGPT. ChatGPT is absent from four of the ten, an exact count because ChatGPT was collected every day. Claude is absent from one. Perplexity is absent from none. Gemini's three missing observations are an upper bound, because Gemini had 22 intermittent days with no collection.

Gem, at eight cited runs, is the other vendor-owned row, reach 5 of 6. Lever, also eight, is classified as an other observed source, reach 6 of 6. Ranks seven through thirteen sit on that same eight. Alphabetical order inside a seven-way tie is not a performance cliff. I would not spend to "break into the top ten" from nine, and I would not spend to defend ten against eleven, on this file.

The news segment is the control I would keep on the wall. Six hundred and eight runs across 53 dates, against 101 runs across seven dates for buyer choice. Eight of the news top ten show up across all six engines somewhere in the full release. Buyer choice is younger, thinner, and shape-specific. Five other HR shapes were still collecting on this release: best tools, is it worth it, problem-first research, top lists, and comparisons. Until one of those publishes, I cannot know whether Pin, LinkedIn, or the two-engine operator sites are a property of "how buyers choose" or a property of HR. The reversal I would accept needs two additional shapes and three releases. A thicker news list does not count as one of those shapes. It counts as a warning against mixing the tactics.

The September 21 release around this segment is 16,146 answer runs, 126,924 citation events, and 22,377 domains, from May 10 through September 21. I would cite the release id mri_score_v2.0+2026-09-21+d73e239e7c5b in any sentence that uses the 14 or the eight. A sentence that drops the release id is how a tied cutoff becomes a slogan.

The decision

Field	What I would write down
Decision	Treat the first HR buyer-choice leaderboard as an experimental shortlist
Date	2026-09-21
Evidence	MRI release mri_score_v2.0+2026-09-21+d73e239e7c5b
Scope	HR and talent, how buyers choose; 101 observed runs, seven dates, 235 cited domains
Action	Keep the rank and the source-role baseline; make only shape-specific decisions
Stop	Do not publish a category-leadership claim from this leaderboard
Reversal	The same vendor or source layer holds a top-three position in at least two additional published HR buyer shapes, and remains top three across three successive releases
Review	When the next HR buyer shape publishes, then at the next two releases

Scope stays inside how buyers choose HR software, on this release, until another shape publishes. The September 21 release as a whole covers 16,146 observed answer runs, 126,924 citation events, and 22,377 cited domains from May 10 through September 21, across ChatGPT, Claude, Gemini, Google AI Mode, Google AI Overviews, and Perplexity. I would point a skeptic at the machine-readable release and the HR buyer-choice leaderboard, not at a cropped chart.

The instrument watches a fixed basket of questions. It does not measure product quality, sales, or causation. Whole-window engine reach does not say which engine produced each HR citation. A rank is an observation.

What would change my mind is the reversal line, on purpose harder than "still number one tomorrow." Pin, or whichever source I was tempted to crown, would need a top-three position in at least two additional published HR buyer shapes and would need to hold top three across three successive releases. A wider margin on this same shape, still alone, would make the experiment more interesting. It would not clear the category sentence. If a later release closed the collection gaps and showed six-engine reach for fewer than half the top ten, I would add that clause back as a live test. If the gaps stay, I would keep quoting the three engines that were collected every day and the upper bounds on the rest.

FAQ

How do engines shortlist vendors?

For a defined buyer question, they cite a set of domains, and the set changes with the question shape and with engine reach. This HR file does not support one universal shortlist mechanism. It supports 101 runs, seven dates, and a leader at 14 citations.

Is the number-one source the best vendor?

Pin is the most-cited domain in this shape on this release, at 14 of 101, with vendor-owned as its role and whole-window reach of 5 of 6. LinkedIn is second at 13 and is not an HR product. The list ranks cited sources. It does not score the software.

When would I use the word leader?

After the result survives more than one question shape and more than one release, with a margin that is not a tied cutoff. The reversal I would accept is top three in at least two additional published HR shapes, held across three releases. That evidence is not in the September 21 file. Ranks seven through thirteen tied at eight cited runs are the reason a top-ten badge is especially weak here.

Why did the engine story change in late September?

Because the instrument's own collection was uneven, and an absence count treated the unevenness as a market fact. Google AI Overviews was silent for 41 of 132 days, Google AI Mode for 11 days at the end of the window, and Gemini for 22 intermittent days. The September 25 fix, which still treated Gemini's misses as exact, was corrected again on September 26. I would rather publish a smaller exact count — ChatGPT absent from four of the ten, Claude from one, Perplexity from none — than a precise-looking ranking of engines the crawler did not run.

What should an HR company do with the list?

Inspect the source roles and the pages that are winning the buyer-choice question. Preserve the release id. Do not generalize it into a category strategy while the other five shapes are still collecting. A news segment at 608 runs across 53 dates is a different market, and I would plan it as one.