I Would Not Pitch Only the Vertical Press

I would build an AI citation target list from domains with a measured observation count, and I would use the vertical as a filter I apply after that count. At 50 or more observations in the September 23, 2026 Machine Relations Index release, only 14.0% of cited domains are single-category, and the median domain spans 5 of 20 categories. A media list organized as "our trade press, then everyone else" matches how I would still reach a human editor. It does not match how this release behaves once a domain has been seen enough times to trust.

The industry cut of the same question is Which Publications Do AI Engines Cite by Industry?. That page is a research synthesis. This essay is how I would use it as a founder: when the vertical instinct is still right, when the headline percentages are incomparable, and what I would need to see before I went back to pitching by category alone.

What I would decide

The templates in circulation say to tier outlets, and they treat trade titles as a group of their own. Everything-PR's template makes Tier 3 "trade and niche outlets," and AMW's guide puts niche publications in Tier 3 (Everything-PR, AMW). They say to segment by outlet type, beat, and industry (PRLab), and to build a universe of national, trade, regional, and other outlets (ContentGrip, AMW). A placement in a high-quality industry vertical is said to create more qualified pipeline than a brief mention in a general business daily (Presskid).

I would keep that advice for a human pitch. I would not let it be the sort order of a citation list. The two lists can share outlets. They should not share a method.

Why the vertical story looks strong, and why the famous percentages cannot share a slide

Webserv, a behavioral-health marketing agency, says Profound and SEMrush data on 2026 behavioral-health AI citations, which it does not link, show DR 50–70 vertical publications carrying roughly 64% of LLM citations on treatment-center queries while DR 70+ generalist sites carry under 3%, and argues that mid-DR verticals are the productive targets (Webserv). The same firm's playbook says to cluster cited URLs by DR tier and pitch mid-DR verticals first (Webserv). We have published a reading of those third-party studies on the research side (Machine Relations Research). I would keep that reading in the file. I would not let it outrank a denominator we can state.

Put the widely cited concentration figures next to each other and they cannot describe one market:

Even inside one source the denominator moves: the Webserv piece that gives Wikipedia 47.9% of ChatGPT's top ten cited sources also puts it at 7.8% of all ChatGPT citations. Each number uses a different population: one engine or five, answers or links, domains or URLs, a curated head or everything observed. Almost none of them say which. I would not build a pitch list out of numbers that are not comparable. I would ask for the denominator before I let a percentage onto a slide.

What our release does once the one-off domains are set aside

Release methodology v2.0, contract machine_relations_index_public_view_v2.0, generated September 23, 2026, window May 10 to September 23, 2026: 130 observed days, 16,475 monitored answer runs, 129,264 source events, 23,280 cited domains, six engines. The evidence floor is 10 observations across 7 distinct run dates. Confidence in this release: 14 A, 59 B, 441 C, and 22,766 still collecting.

The first reading is the one that would have made an easy post. Of cited domains, 86.6% appear in exactly one category. Pitch the vertical. The tell is in the same output: 46.9% of all cited domains were observed exactly once in 130 days. A domain seen once can only sit in one category. The 86.6% is mostly a census of things barely seen.

Condition the share on how often the domain was observed and it falls in one direction:

Across all 23,280 domains the median span is one category. Among the 228 domains at 50 or more observations it is 5 of the 20 categories that carry citation data, and those 228 hold 25.0% of all 115,066 citation observations in the window. All 14 grade-A domains appear in more than one category. That is 14 rows. I would not dress fourteen domains up as a percentage. On this release, the sources that carry weight are the ones that travel. Category span is what citation volume looks like from the side.

Where the vertical instinct is still the right instinct

Both shapes are real. Thirty-two domains cleared 50 observations inside a single category: omnimd.com at 105 in Healthcare Services, huntress.com at 94 in Cybersecurity, allaboutcookies.org at 91 in Family Software. Forty-one cleared 50 observations across ten or more categories. Deep category-local sources exist, they are reachable, and they are exactly why a trade-press instinct feels right.

There are 32 of them. They are nameable. I would find them by reading an observation count, not by pulling every publication that serves the industry and calling the pile a tier. A vertical list that starts from the 32 is a list I would fund. A vertical list that starts from a media directory is the list I would cut.

A ClearSight analysis of 600 product queries across 12 Amazon-brand categories in Google AI Overviews says the top 3 domains in a category captured a median of 47% of weighted citations and the top 10 captured 78% (ClearSight). Read as one study's result, that is useful. Read as a portfolio constant, it tells a founder to stop looking in the middle of the market, which is where the reachable sources in this release sit. In this release, across the 20 categories with data, the ten most-cited domains in a category hold between 7.0% and 24.2% of that category's citations, and reaching half of a category's citations takes between 49 and 304 distinct domains (the 304 is the legacy news-topic bucket; the largest measured category needs 247). AuthorityTech's category market-size study reports the same cut on the October 14 release: 7.6% to 23.2% and 57 to 250 domains. I would keep the 78% figure labeled as someone else's segment. I would plan against the Index range, because I can see the denominator.

The question I would require on any proposed domain is simple. How many times has it been observed being cited, over what window, and in how many categories? "Once" is a row in the long tail, and 46.9% of the domains in this release are that row.

Per-category run denominators in the release sum to 16,475, which is the release's own published run count. Category-span figures are computed over the 20 categories that carry data, out of 25 taxonomy nodes. I would want those sentences next to any chart a vendor builds from this release, so a later chart cannot quietly change the population.

The discipline that makes me want a count I can recompute, rather than a tier someone believes in, is Machine Relations. When I say observation count I mean a counted citation, and the definition I would hand a vendor who says "visibility" instead is in the glossary.

The decision

I would rebuild the AI citation target list in two columns: observation count over the May 10 to September 23 window, and category span out of 20. I would keep a separate human-media list, tiered the way the guides above describe, for reporters. I would not let the human list's tiers set the order of the citation list. I would pursue the 32 deep category-local domains when they match our category, and the domains that travel when our buyers ask questions those domains already answer.

Scope: placement targeting where the success condition is an AI answer citation, on this release's six engines and 130 observed days. It does not govern a launch aimed at a human trade audience, and it does not say a single-category domain with 105 observations is a worse target than a generalist with a famous name and a thin count.

What would change my mind: a conditioned cut of a later release in which category span falls as observation count rises. That is the reversal. I would also reopen the "do not reconcile the third-party percentages" rule if two of those publishers disclosed matching denominators, one engine or five, answers or links, domains or URLs, and the figures then agreed. Until then they stay in the comparison table as incomparable, which is the reason the table exists.

Correction, November 11, 2026

An earlier version of this post credited the 68% top-15 figure to 5W's AI Platform Citation Source Index as if it had been measured, and linked the release. The Index's publisher, Everything-PR, calls the number its own calculation from a composite ranking of six studies, and says their raw citation counts were not added together. The table now says where the number comes from, and the link is removed. The post's conclusion, that these figures cannot share a slide, is unchanged.

The same table also credited "the top 1% of domains, roughly 12 sites, capture 47% of citations" to Everything-PR's AI Overviews Citation Source Index. It traces to a study of 1,000 AI Overviews that the index lists as Digital Applied's. Digital Applied's page for that study now redirects to its comparison of other studies, which leaves it out, and another firm, theStacc, publishes the same figures as its own study. I would not put a number on the slide when I cannot say whose count it is. The row now carries a concentration figure its publisher still stands behind: Pew Research Center found that Wikipedia, YouTube and Reddit together made up 15% of the sources listed in the Google AI summaries it examined. The FAQ answer names the same five figures. The conclusion is unchanged.

FAQ

Should the pitch list be our trade publications and nothing else?

For a human editor, vertical relevance still filters the list. For an AI citation target, I would sort by observation count first. At 50 or more observations only 14.0% of domains are single-category, the median spans 5 of 20, and 46.9% of all 23,280 cited domains were seen once, which is why the unconditioned 86.6% single-category figure overstates how local the real sources are.

Which outside percentages can share a slide with the Index?

None of the five in the comparison table, until their denominators match. The 68% top-15 figure, the 97.4% non-Tier-1 figure, the "no domain above 5%" figure, Wikipedia at 47.9% of ChatGPT's top ten cited sources, and Pew's 15% for Wikipedia, YouTube and Reddit in Google's AI summaries are computed on different populations. The same Webserv page that gives Wikipedia 47.9% of ChatGPT's top ten puts it at 7.8% of all ChatGPT citations. I would show them as a warning about denominators, not as a market size.

Are category-specific outlets ever the right target?

Yes. Thirty-two domains in this release cleared 50 observations inside one category, including omnimd.com at 105 in Healthcare Services, huntress.com at 94 in Cybersecurity, and allaboutcookies.org at 91 in Family Software. Forty-one cleared 50 across ten or more categories. I would pitch the 32 when they are our category. I would find them from the count, not from a directory of every title that covers the industry.

What do I ask an agency that hands me a tiered outlet universe?

For each domain, the observation count, the window, and the category span. A domain that cannot produce those is a publication someone thinks is important. The release I would check them against is methodology v2.0, May 10 to September 23, 2026, 16,475 answer runs, 23,280 cited domains, evidence floor of 10 observations on 7 dates. The 228 domains at 50 or more observations hold 25.0% of 115,066 citation observations. That is the head of the list I would rather be on.

Correction, November 12, 2026

An earlier version of this post presented Webserv's behavioral-health figures, DR 50–70 vertical publications carrying 64% of citations and DR 70+ generalist press under 3%, as a 2026 analysis that found them. Webserv credits them to Profound and SEMrush data on 2026 behavioral-health AI citations, which it does not link, and I found no Profound or Semrush publication of that split. The sentence now says whose figures they are, as Webserv states them. My judgment about vertical targets is unchanged.