Most Publications AI Engines Cite Answer Exactly One Buyer Question

If you run earned media against AI answers, your pitch list is probably one list. It should be six.

I read the Machine Relations Index release of September 22, 2026 whole and asked one question of it: how many of the buyer questions does a cited publication actually hold? The release classifies 1,238 domains as editorial publications. Forty-seven of them hold a top-ten position in any published buyer-question segment. Thirty-five of those forty-seven hold exactly one.

Four publications hold a top-ten position across all six buyer question shapes: Medium, Forbes, TechRadar and NerdWallet. Everybody else is a specialist, whether their media kit says so or not.

That is the whole finding, and it changes one thing on Monday: a placement's value is bounded by the single question the outlet holds, so the target list has to be built per question, and the brief has to name which question the placement is supposed to win.

What was measured

The Index runs real buying questions against six answer engines every day and records which source domains get cited. Release mri_score_v2.0+2026-09-22+4aa6d302f7ed covers May 10 to September 22, 2026 — 129 days, 16,307 monitored answer runs, 22,828 cited domains, 127,980 source events.

It splits a subject category into question shapes. Six of them are what a buyer asks while deciding: best tools, how buyers choose, is it worth it, problem-first research, top lists, and head-to-head comparisons. A category paired with a shape is a segment, and a segment publishes once it clears the evidence floor of ten observations across seven distinct run dates. Ninety-three segments are published in this release.

I used top-ten position, not top-100 presence. In a segment with 1,375 ranked sources, a top-100 slot is the long tail with a certificate. Top ten is where the citations concentrate and where a placement can plausibly change an answer.

Forty-seven publications, and most of them do one thing

Buyer question shapes held in the top ten	Publications
All six	4
Four	4
Three	1
Two	3
Exactly one	35
Any at all	47

Read the bottom row first. Of the 1,238 publications this index has observed being cited at all, 1,191 hold no top-ten position on any published buying question. They are cited. They are not deciding anything.

The four that cover the whole board look less like a media plan than you would hope:

Publication	Top-ten segments	Where it holds all six shapes
Forbes	14	Consumer finance and consumer products, plus single-shape positions across five more categories
Medium	12	AI infrastructure, where it is rank 1 of 258 on how buyers choose
TechRadar	10	Family software and prosumer hardware
NerdWallet	7	Consumer finance, where it holds six of the six shapes by itself

Two of the four are not pitch targets in any normal sense. Medium is a self-publishing platform, and NerdWallet is a review and affiliate property. That leaves Forbes and TechRadar as the only two outlets in this dataset that a media plan can call on and that carry across the whole buyer journey — and each of them does it inside two consumer categories, not everywhere.

One more caution about the class itself. The release files Amazon as the seventh-ranked editorial publication and Apple as the sixteenth. Neither is a publication. Para Labs published the Amazon case in detail on September 20; Apple is the same defect, and it is why Apple appears as the top-cited "publication" for how families choose software. When you benchmark against an editorial leaderboard, check what is actually in the column.

The question that decides the deal is the one publications lose

Split every citation in the published segments by the kind of source that earned it, and the shape where editorial publications do worst is the one where a buyer is choosing.

Buyer question	Share of citations going to editorial publications
Best tools	11.6%
Is it worth it	10.1%
Head-to-head comparisons	8.3%
Top lists	8.9%
Problem-first research	7.3%
How buyers choose	7.0%

Fifteen categories publish a how-buyers-choose segment. In nine of them at least one editorial publication holds a top-ten position. In six — AI security and privacy, consumer health, cybersecurity, education and training, fintech, and HR and talent — not one does.

Here is who actually holds it where an editorial publication is in the top ten:

Category	Editorial publications in the top ten for how buyers choose
AI infrastructure	Medium — rank 1 of 258, cited in 33 of 106 runs
Enterprise software	TechTarget — rank 3 of 172, cited in 18 of 99 runs
Consumer finance	NerdWallet rank 3 of 191; Forbes rank 9 of 191
Healthcare services	OmniMD — rank 1 of 214, cited in 16 of 107 runs
Family software	Apple rank 1 of 153; TechRadar rank 5; PCMag rank 10
Prosumer hardware	TechRadar rank 4 of 192; Amazon rank 7
Deep tech hardware	EE Times rank 4 of 195; EE Journal rank 6
Consumer products	Allure — rank 5 of 427
AI visibility and GEO	Shadow — rank 10 of 448, at 9.7%

In the six categories with none, the sources answering the deciding question are vendor documentation, Reddit, YouTube, NIH and a long tail of operator-run sites the release does not classify at all. In cybersecurity the top three are Microsoft, Reddit and JumpCloud. In HR and talent they are a vendor, LinkedIn and Reddit. Coverage is not the lever in those categories this quarter. Saying so is more useful to a CMO than another list.

One nuance worth keeping, because it protects you from over-reading the table: a publication can hold a rank inside a segment while its overall grade is still collecting. EE Times is rank 4 for how deep-tech buyers choose and still carries a collecting grade across the whole index. The segment position is real; the domain-level score is not settled. Quote the segment, with its rank and its total, not the domain.

The Monday version

Rebuild the target list as six lists. One per buyer question shape, in your category, from the segment leaderboards rather than from a media database. Most outlets will appear on one of the six. That is the correct result, not a gap to fill.

Put the question in the brief. Every pitch, contributed piece and data drop should name the buyer question it is meant to win, and the segment rank the outlet currently holds. "TechTarget, rank 3 of 172 on how enterprise buyers choose" is a target. "Tier one trade" is not.

Change the intake question for any agency or platform. Not "what placements can you get" but "which buyer question does that outlet hold, at what rank, of how many ranked sources, in which category". If nobody can answer that, the placement is being sold on its masthead.

Give sales one line per placement. The handoff sentence is: this outlet is cited when a buyer asks [question shape] about [category], at rank N of M sources. It is checkable, it expires, and it stops a rep claiming an article proves anything about a different question.

Stop paying a premium for breadth. Outside those four publications, breadth does not exist in this dataset. An outlet that holds one shape is worth what one shape is worth. Price it there.

I would run this list again at the next quarterly release rather than treating it as settled. Segments clear the evidence floor and enter the ranking as dates accumulate, and a category with no editorial publication in its deciding answer today can acquire one.

FAQ

Does a top-ten position mean my placement will be cited?

No. The Index measures which source domains engines cite when buyers ask a category's questions. It does not measure your article, and no measurement here shows a placement causing a citation. The position tells you which outlets an engine already reaches for, which is the right place to spend, not a guarantee about one piece.

Why top ten instead of the top 100 the leaderboards publish?

Because the tail is long. A segment can rank over 1,300 sources, so a top-100 slot can mean a single citation across four months. Top ten is where citation actually concentrates, and it is the threshold at which a target list stays short enough to act on.

My category has no editorial publication in the deciding answer. What do I do?

Fund the sources that are in it. In those six categories the answers are being built from vendor documentation, community threads, video and operator-run sites. That usually means your own documentation and comparison pages, plus the community surfaces your buyers actually read, before it means another trade placement.

Does this mean trade publications do not matter?

It means they matter for a specific question, usually news, and that the question is rarely the one asked before a purchase. A trade placement can still be the right buy for recruiting, credibility or analyst attention. Just do not book it against a buying-answer objective it does not serve.

How do I check any of this myself?

Every figure has a public page. Category-and-shape leaderboards live at machinerelations.ai/index/categories/<category>/<shape> — for example enterprise software, how buyers choose — and each domain's full set of observed segments is at machinerelations.ai/index/domains/<domain>, such as TechRadar's. Read the rank with its total beside it, and note the release id; these move daily.