Medium, Substack and DEV show up in AI answers mainly on two technical topics. In the Machine Relations Index answer ledger, twelve open publishing hosts supplied 1,692 of 201,266 cited-URL events (0.84 percent). On AI infrastructure questions they were 6.0 percent of the events over the whole window (8.4 percent in July, about 4 percent from September). Outside AI infrastructure and AI security they were 0.39 percent.
Medium alone was 863 of the 1,692. The citations are scattered: 419 different Medium author or publication names appear, 265 of them once, and no name is more than a sliver of the whole. A few individual URLs are cited again and again.
This page counts addresses an engine listed. It does not show that publishing on one of these hosts earns a citation, and it does not measure how many posts exist on the hosts or what their authors wanted. "Open publishing" describes the hosts, not every page on them: GitHub Pages, for example, carries organisation documentation (owasp.github.io, microsoft.github.io) as well as personal blogs. Without GitHub Pages the total is 1,558 events (0.77 percent).
What was counted #
- Population. Successful runs in the Index's buying-question panel from 2026-07-06 to 2026-11-09 on its 1,008 buying questions, 168 in each of six question shapes (best X, top lists, how to choose, is it worth it, problem first, X vs Y) across 24 verticals. Runs flagged as naming a brand in the prompt are excluded. The ledger export is dated 2026-11-10 08:10 UTC. It is the same population as the PDF, Markdown and llms.txt study.
- Engines. ChatGPT, Claude, Gemini, Google AI Mode and Perplexity. Google AI Overviews are left out because this analysis was not run on them.
- Unit. One cited URL in one answer run, counted once however many times the run repeated it. The same URL cited in two runs counts twice. The set holds 201,266 events from 23,729 runs that cited at least one address, 79,206 distinct URLs and 29,587 root domains. These are address strings as written. Under the Index's page-counting rule the same 201,266 events are 75,656 pages.
- An open publishing host. An address whose root domain is one of twelve hosts where anyone can publish under their own name, or a site or blog of their own, on the host's address: medium.com (including name.medium.com), substack.com (including name.substack.com), dev.to, github.io, hackernoon.com, vocal.media, wordpress.com, blogspot.com, hashnode.dev and hashnode.com, beehiiv.com, notion.site and ghost.io. GitHub.com is not counted (it is where code lives), nor are LinkedIn, Reddit, YouTube and Quora, which the social platform study covers. A publication that left a host for its own domain is not counted, and the ledger cannot tell one.
- The questions are the Index's own. They were written to cover buying decisions in 24 verticals. They are not a sample of what buyers type, and the panel changed over the window (see the last section).
Medium is half of it; the long tail is the rest #
Cited-URL events on each host (share of all 201,266; runs and questions are those with at least one such citation):
| Host | Events | Share of all events | Distinct URLs | Runs | Questions |
|---|---|---|---|---|---|
| Medium | 863 | 0.43% | 475 | 672 | 207 |
| DEV (dev.to) | 403 | 0.20% | 202 | 359 | 107 |
| Substack | 204 | 0.10% | 100 | 197 | 99 |
| GitHub Pages (github.io) | 134 | 0.07% | 74 | 122 | 47 |
| HackerNoon | 30 | 0.015% | 14 | 30 | 18 |
| Vocal | 21 | 0.01% | 12 | 21 | 13 |
| WordPress.com, Blogger, Hashnode, Beehiiv, Notion, Ghost | 37 | 0.02% | 26 | 37 | 28 |
| All twelve | 1,692 | 0.84% | 903 | 1,277 | 331 |
The 95 percent interval for the all-host share, resampling the 1,008 questions, is 0.70 to 0.99 percent. 331 of the 1,008 questions (33 percent) had at least one open publishing host cited on at least one run; 677 had none. Medium is 51 percent of the events and DEV 24 percent; the other ten hosts together are 25 percent. Distinct URLs are counted as written; stripping query strings such as ?utm_source=openai leaves 899.
For scale, in the same window LinkedIn was 1,406 events (0.70 percent), Reddit 3,341 and YouTube 2,996, and Wikipedia 265. The twelve hosts together are 20 percent more events than LinkedIn and about half of Reddit's.
The share is 6.0 percent in AI infrastructure and almost nothing in most verticals #
Share of cited-URL events that sat on an open publishing host, all 24 verticals (events on the twelve hosts, out of all events; the interval resamples the vertical's 42 questions):
| Vertical | Cited URLs | On an open publishing host | Share |
|---|---|---|---|
| AI infrastructure | 10,500 | 627 | 5.97% (4.59 to 7.45) |
| AI security and privacy | 11,728 | 365 | 3.11% (2.26 to 3.97) |
| Deep-tech hardware | 9,938 | 98 | 0.99% |
| AI visibility and GEO | 13,237 | 124 | 0.94% (0.61 to 1.32) |
| Cybersecurity | 11,110 | 101 | 0.91% |
| Fintech | 9,205 | 54 | 0.59% |
| Education and training | 9,405 | 50 | 0.53% |
| Emergent prosumer | 9,688 | 49 | 0.51% |
| Martech and advertising | 4,236 | 20 | 0.47% |
| Family software | 9,501 | 44 | 0.46% |
| Consumer finance | 11,850 | 48 | 0.41% |
| VC and private equity | 2,880 | 10 | 0.35% |
| Enterprise software | 9,252 | 26 | 0.28% |
| Professional services | 2,699 | 6 | 0.22% |
| HR and talent | 8,626 | 17 | 0.20% |
| Consumer products | 12,423 | 24 | 0.19% |
| iGaming and betting | 8,412 | 9 | 0.11% |
| Healthcare services | 8,379 | 7 | 0.08% |
| Local services | 6,438 | 3 | 0.05% |
| Physical consumer goods | 3,756 | 2 | 0.05% |
| Consumer health | 12,832 | 5 | 0.04% |
| Sales, GTM and CRM | 3,459 | 1 | 0.03% |
| Logistics and freight | 5,379 | 1 | 0.02% |
| Review-sparse industrial | 6,333 | 1 | 0.02% |
AI infrastructure and AI security together are 22,228 of the 201,266 events (11 percent) and 992 of the 1,692 open publishing events (59 percent). Their combined share is 4.46 percent. Across the other 22 verticals it is 700 of 179,038 events: 0.39 percent (0.33 to 0.45); 16 of those 22 are under 0.5 percent and six are between 0.5 and 1.0. In AI infrastructure, 372 of 1,211 runs that cited anything (30.7 percent) cited at least one such post; in AI security, 268 of 1,322 (20.3 percent); across all verticals, 1,277 of 23,729 runs (5.4 percent).
The pooled 0.84 percent is not a stable rate. By month, the share of all events was 2.22 percent in July, 0.70 in August, 0.41 in September, 0.58 in October and 0.56 in November. The panel's mix moved too: about 30 percent of July events were on AI infrastructure or security questions, none in August, and 7 to 9 percent from September. The share fell inside both groups of verticals:
| Month | AI infrastructure and security | Other 22 verticals |
|---|---|---|
| July (from the 6th) | 5.92% (606 of 10,229) | 0.63% (151 of 23,858) |
| August | no events | 0.70% (168 of 24,008) |
| September | 2.59% (63 of 2,436) | 0.26% (88 of 34,155) |
| October | 3.37% (268 of 7,946) | 0.31% (247 of 80,548) |
| November (1 to 9) | 3.40% (55 of 1,617) | 0.28% (46 of 16,469) |
AI infrastructure alone was 8.4 percent in July and 3.8 to 4.2 percent in September to November. The 6.0 percent headline is therefore a pooled figure that leans on July: 606 of the 992 technical-vertical events are from that month. Read the vertical and monthly rows, not the pooled figure. This page did not test why the share fell; the October change in Perplexity's collection (see what changed in Perplexity's collection) is one of several things that moved.
The engine matters, and the dev verticals do not explain it away #
Share of each engine's cited URLs on an open publishing host, split into the two technical verticals and the other 22 (events on the hosts, out of events):
| Engine | AI infrastructure and security | Other 22 verticals | All |
|---|---|---|---|
| Google AI Mode | 9.44% (140 of 1,483) | 0.53% (25 of 4,695) | 2.67% |
| Gemini | 8.07% (243 of 3,013) | 0.60% (87 of 14,567) | 1.88% |
| Claude | 5.71% (207 of 3,627) | 0.57% (173 of 30,443) | 1.12% |
| Perplexity | 3.02% (317 of 10,491) | 0.35% (334 of 96,661) | 0.61% |
| ChatGPT | 2.35% (85 of 3,614) | 0.25% (81 of 32,672) | 0.46% |
In the technical verticals Google AI Mode lists these hosts four times as often as ChatGPT and Gemini 3.4 times; outside them, the ratio is about two. ChatGPT and Perplexity are the lowest in both columns, so the gap is not only a matter of which engine saw more infrastructure questions. Engines also differ in how many sources they list (see how many sources each engine cites), so the per-run figures are less comparable than the per-event ones: a run on Gemini cited an open publishing host 211 times in 1,409 runs (15.0 percent), Perplexity 525 in 7,964 (6.6 percent), Google AI Mode 71 in 883 (8.0 percent), Claude 321 in 6,127 (5.2 percent) and ChatGPT 149 in 7,346 (2.0 percent).
By question shape: top lists cite these hosts least #
Share of cited URLs on an open publishing host by question shape (168 questions each; events on the hosts, out of events), pooled and split by the two technical verticals:
| Question shape | All verticals | AI infrastructure and security | Other 22 verticals |
|---|---|---|---|
| Problem first | 1.02% (273 of 26,691) | 4.96% (141 of 2,845) | 0.55% (132 of 23,846) |
| X vs Y | 1.01% (310 of 30,672) | 5.79% (181 of 3,125) | 0.47% (129 of 27,547) |
| Best X | 0.97% (383 of 39,663) | 6.21% (273 of 4,399) | 0.31% (110 of 35,264) |
| Is X worth it | 0.92% (288 of 31,296) | 3.81% (136 of 3,570) | 0.55% (152 of 27,726) |
| How to choose | 0.83% (266 of 32,031) | 4.43% (168 of 3,789) | 0.35% (98 of 28,242) |
| Top lists | 0.42% (172 of 40,913) | 2.07% (93 of 4,500) | 0.22% (79 of 36,413) |
Top-list answers have the lowest share in all three columns; pooled, it is about half the next lowest shape. The order of the other five changes between columns, so it is not a stable ranking. Engines are not evenly spread across shapes either, so this page does not attribute the gap to the shape alone. In AI visibility and GEO, 124 of the 13,237 events were on the hosts (Medium 72, DEV 27, Substack 15, HackerNoon 4, Vocal 3, GitHub Pages 3): 34 of 1,901 on head-to-head questions (1.8 percent) and 11 of 2,757 on best-X questions (0.4 percent).
Who is cited: many names, a few URLs cited again and again #
The author is the name in the address: medium.com/name, medium.com/@name or name.medium.com, name.substack.com and dev.to/name. On Medium the first path segment can be a publication rather than a person, so the count is of names, not people. Medium shows 419 distinct lowercased names in its 863 events, DEV 173 in 403 and Substack 79 in 204. 265 of the 419 Medium names appear once. The ten most-cited Medium names hold 14.6 percent of the Medium events, so there is no handful of authors who own the platform's citations.
The URLs are less spread out. Of the 903 distinct URLs, 590 (65 percent) were cited in a single run. At the other end, 65 URLs (7 percent) were cited five or more times and account for 482 of the 1,692 events (28 percent). The most-cited URL, a Medium explainer on differential privacy, appears 20 times on two questions and two engines. The top fifteen URLs were each cited on one to five questions, so none was a source across many questions.
Of the 863 Medium events, 698 were medium.com/name/post-title addresses and 165 were on a name.medium.com subdomain.
What this changes for how pages are published #
- In AI infrastructure and security, these hosts are a visible share of what engines list. Across the window 5.97 percent and 3.11 percent of the cited URLs sat on them (3.8 to 4.2 percent and 1.5 to 2.9 percent since September), and three in ten AI infrastructure runs cited at least one. If your buyers ask technical how-to and comparison questions, these hosts are among the sources listed on them. The counts do not say that a new post there would be listed.
- Elsewhere they are a small share. Across the 22 other verticals 0.39 percent of cited URLs were on these hosts; 16 of the 22 were under 0.5 percent, and in consumer health, sales and CRM, logistics and review-sparse industrial questions the count was five or fewer events in a vertical of thousands. The ledger cannot separate this from how few relevant posts exist.
- Most cited URLs were cited once. 65 percent appeared in a single run. This page did not read any post, so it says nothing about what made the repeated ones different.
- Check the engine first. Google AI Mode and Gemini list these hosts several times more often than ChatGPT does in the technical verticals and about twice as often elsewhere.
Machine Relations and AuthorityTech do not appear in this part of the ledger. No open publishing-host address in the window carried authoritytech, machinerelations, jaxon, parrott or lehman in the URL. The check is by address only, so an unlabelled post by us would not show, and the count is too small to say anything about what a post by us would earn.
What this page does not show #
- Cause. A post may be cited because of its content or because of a link from somewhere else, not because of where it sits. No page was read.
- Supply. The ledger cannot see how many posts exist on each host for each question. A low share can mean that few relevant posts exist as much as that engines avoid the host.
- Custom domains. A Medium or Substack publication on its own domain is counted as its own site, so these figures understate what is hosted on those platforms.
- Who wrote it. An address shows an author name but not the person's employer or whether the post is promotional. This page did not check either.
- A trend. The share fell across the window in both groups of verticals while the panel's mix and the Perplexity collection changed. This page did not separate those from a change in the engines.
- Buyer demand. The 1,008 questions are the Index's own panel, not a sample of what buyers type.
- Consumer apps and Google AI Overviews. Not included. The five engines are API routes.
The same ledger's other address studies: PDFs, Markdown and llms.txt, the slug study, site sections and subdomains and subfolders.
Method note #
Source: the Index answer ledger's citation-events table, daily export of 2026-11-10 08:10 UTC, read with DuckDB and joined to the run table for success and to the question table for the question shape. Filters: runs on the buying-question panel (mri_v2) that succeeded, run dates 2026-07-06 to 2026-11-09, events not flagged as brand-prompted, a non-empty cited URL, engines ChatGPT, Claude, Gemini, Google AI Mode and Perplexity. Unit: distinct (answer run, cited URL) pairs. A open publishing event is one whose root domain, as recorded in the ledger, is medium.com, substack.com, dev.to, github.io, hackernoon.com, vocal.media, wordpress.com, blogspot.com, hashnode.dev, hashnode.com, beehiiv.com, notion.site or ghost.io. The author name is read from the address with a pattern, lowercased: the first path segment on medium.com, dev.to and substack.com, or the subdomain on name.medium.com and name.substack.com, and the author count covers Medium, DEV and Substack only. Intervals resample the questions of the group (1,008 for all events, 42 for a single vertical, 924 for the 22 other verticals), 1,000 draws, fixed seed, with every cited URL of a question kept together; engine, shape and monthly rows report pooled counts and no interval. Run shares use the runs that cited at least one address (23,729) as the denominator. The address-only check for our names looked for the strings authoritytech, machine-relations, machinerelations, jaxon, parrott, christian-lehman and lehman in the lowercase cited URL.
Frequently asked questions #
Do AI engines cite Medium? #
Yes, and Medium is the largest open publishing host in the ledger. Medium addresses were 863 of 201,266 cited-URL events (0.43 percent), from 475 distinct URLs under 419 author or publication names. That is 51 percent of all cited-URL events on twelve open publishing hosts.
Do AI engines cite Substack? #
In a small share. Substack was 204 of 201,266 cited-URL events (0.10 percent) from 100 distinct URLs under 79 author or newsletter names, on 99 of the 1,008 questions. Medium was cited more than four times as often and DEV about twice as often.
Which topics do AI engines cite Medium, DEV and Substack for? #
Mostly AI infrastructure and AI security. Open publishing hosts were 5.97 percent of cited-URL events on AI infrastructure questions and 3.11 percent on AI security, against 0.39 percent across the 22 other verticals. Those two verticals hold 59 percent of the 1,692 events.
Which engine cites Medium and Substack most? #
Google AI Mode, then Gemini. On AI infrastructure and security questions, 9.4 percent of Google AI Mode's cited URLs and 8.1 percent of Gemini's sat on an open publishing host, against 2.4 percent of ChatGPT's. Outside those two verticals the shares were 0.25 to 0.60 percent for all five engines.
Should I publish on Medium or Substack to be cited by AI? #
The ledger cannot say that a post earns a citation. It shows where engines listed these hosts: a real share of cited URLs on technical infrastructure and security questions, and 0.39 percent in the other 22 verticals. 65 percent of the cited URLs appeared in a single run.