If you want to be cited by AI engines, most of the work does not happen on your website. Third-party AI citations dominate every dataset published this month, and the practical question is no longer whether to invest off-domain — it is which off-domain surface, because the answer changes completely depending on what you sell.
Two pieces of research landed eight days apart and, read together, they settle an argument the industry has been having in generalities. Aleyda Solis published a vertical breakdown on August 2 covering 15 leading brands across three verticals, using Semrush Enterprise data to identify the top 10 cited source domains for each. Kevin Indig and Amanda Johnson published a community-signals analysis on August 10, drawing on roughly 35,000 citation URLs captured via Profound for a study conducted for G2 in February 2026, US and ChatGPT only, covering December 2025 prompts sorted into discovery, exploration, evaluation and focused-evaluation stages.
The two studies that frame the problem
Solis's framing is the sharper of the two: AI search is a third-party citation problem with an on-page corroboration base. Her numbers put owned domains at 17.7% of top cited sources in SaaS, 30.4% in ecommerce and 20.6% in finance. Whatever you publish on your own site is, at best, a fifth to a third of the surface an engine draws from when it answers a question about your category.
Indig and Johnson attack it from the platform side and find the same asymmetry expressed differently. User-generated content platforms account for 17.1% of cited domains — more than four times the 4.0% attributable to publishers. Wikipedia alone contributes between 10.1 and 14.0 points of that floor depending on journey stage. Their sample also produced two numbers worth pinning to a wall: 83% of AI citations for commercial queries come from pages updated in the past twelve months, and only 30% of brands stay visible from one AI answer to the next.
Third-party AI citations look different in every vertical
Here is where the generalities break. Solis's vertical split shows that the composition of the non-owned surface is not remotely consistent, and a digital PR programme optimized for one vertical will be aimed at the wrong platforms in another.
| SOURCE TYPE | SAAS | ECOMMERCE | FINANCE |
|---|---|---|---|
| Non-owned domains, share of top 10 cited | 82.3% | 69.6% | 79.4% |
| Social, community and creator platforms | 45.7% | 27.5% | 29.8% |
| News, review and comparison sites | 10.4% | 0.7% | 19.4% |
| Competitor domains | 8.1% | 36.6% | 19.8% |
| Owned domains | 17.7% | 30.4% | 20.6% |
Look at the news, review and comparison row. It ranges from 19.4% in finance to 0.7% in ecommerce — a difference of nearly thirty times. A retail brand running a press-release-led digital PR programme is buying placements on a surface that supplies well under one percent of its top citation sources. That is not a marginal inefficiency. That is the wrong channel.
SaaS: the community problem
For SaaS the dominant surface is social, community and creator platforms at 45.7% — nearly half of the top cited domains. Indig and Johnson's UGC finding points the same direction, and the mechanism is not mysterious. Software buyers ask other software buyers, in public, in threads that persist and get indexed and get retrieved.
The uncomfortable part is that this is the surface agencies are worst at serving, because it cannot be bought cleanly and it punishes anything that reads like marketing. What works is unglamorous: making sure your product is accurately described in the places practitioners already gather, correcting outdated claims, getting your team visibly useful under their real names, and earning review-platform presence with enough recent volume that an engine treats it as current. Indig and Johnson also found sequential heading structure associated with a 2.8x citation lift, which is a reminder that the on-page corroboration base still has to be there when a third-party mention sends an engine looking for confirmation.
One more number from their data that should reshape SaaS priorities: more than half of SaaS prompts use commercial language, but only 1.5% actually name a vendor brand. The category conversation is enormous and mostly unbranded. That is the opportunity and the reason comparison content earns citations at the rate it does — it answers the unbranded question directly.
Ecommerce: your competitors are your citation surface
The ecommerce column contains the single strangest number in the dataset: competitor domains account for 36.6% of top cited sources. More than a third of what the engine reads when it answers a question about your category is published by the people you are competing with.
That is a genuinely different strategic problem, and traditional link building barely touches it. You cannot outreach your way onto a competitor's domain. What you can do is contest the surfaces where comparison happens — marketplace listings, review platforms, category roundups, and the creator content that 27.5% of the surface represents. And you can make sure that when an engine lands on a competitor's comparison page and reads about you, the description it finds is accurate, because that page is now part of your brand's evidence base whether you like it or not.
“In ecommerce, a third of your citation surface is written by your competitors. You do not own the narrative, so the job is making the version they publish factually correct.”
Finance: the one vertical where classic digital PR still maps
Finance is the vertical where the traditional playbook survives contact with the data. News, review and comparison sites at 19.4% is the highest of the three, competitor domains at 19.8% are meaningful but not dominant, and community sits at 29.8%. A programme that mixes earned media placement with review-platform presence and community credibility is defensible here in a way it is not in retail.
This is also where the industry's competing headline numbers need care. Muck Rack, analyzing more than 25 million AI-cited links, reports that earned media accounts for about 84% of AI citations. Meltwater, via AMEC in July, put news media at 48% across all LLMs and 66% on ChatGPT. BuzzStream's March analysis of 4 million citations across 3,600 prompts put news at 14.09%. Those are not the same measurement: earned media is a far broader category than news media, and cited links, cited domains and citation instances are three different denominators. Before you act on any of them, ask which one is being counted.
Marina Grudeva's framing at Muck Rack is the right procurement question to carry into any vendor conversation: a mention and a citation tell you different things, and the test of a tracker is whether it names the source or just counts the mention. We made a related argument about why PR's AI citation KPI needs an agreed unit, and the vertical data here is the reason it matters — a tool that reports one blended number cannot tell a retail brand that its news placements are landing on 0.7% of the surface.
The bars nobody puts on a slide
Charted side by side, the vertical differences stop being a table of percentages and start looking like three separate jobs that happen to share a name. The community column is the one that should reorganize budgets fastest, because it is the largest single non-owned surface in two of the three verticals and the one almost no digital PR retainer is structured to serve.
Share of top 10 cited source domains that are social, community or creator platforms, by vertical (Solis, Aug 2, 2026, 15 brands)
The last bar is rounded from 0.7% and it is deliberately included to make the comparison uncomfortable. Any retail marketing team looking at that chart should ask what proportion of its earned media budget is currently aimed at the shortest bar on the page, and whether anyone has ever been asked to justify that split against citation data rather than against impressions or domain authority.
One caveat on all of these figures, stated plainly because the studies state it: Solis's breakdown covers 15 brands across three verticals, and the Indig and Johnson analysis is US-only, ChatGPT-only, and drawn from a December 2025 prompt set. Both are the best public evidence available on this question and neither is a census. Use them to form a hypothesis about your own citation surface, then test it on your own prompts before moving real money.
Building for third-party AI citations
Do this next
Pull your twenty highest-intent prompts this week and classify every cited domain into the four buckets. If your spend does not match the resulting distribution, you have found your reallocation, and you can defend it with a table rather than an argument.
Then pick the single largest non-owned surface in your vertical and put a named owner on it for the quarter. Our link building and digital PR work starts from this map rather than a placement target, our B2B SaaS practice runs the community-first version of it, and our earlier analysis of the six link types AI engines actually trust covers which of those surfaces carry weight once you get there. Solis published her full vertical breakdown with the per-brand tables if you want the underlying detail.
See where you are cited today
A free snapshot audit of your rankings and AI citations before we ever talk.
Tyler leads work at the intersection of SEO and generative engines at Something Inc., helping B2B brands get ranked and cited across every major AI engine.